This Memo was sent to 26,313 subscribers. Welcome to +85 new readers! Subscribe to get the free memo weekly or upgrade to Premium for the full archive, research, frameworks, and templates.

Friendly shoutout from me this week to Aleyda Solis’ new venture Flinchling, which scans the news cycle and tells you exactly which stories you should react to, and why.

I normally write about strategy and search behavior, not labor markets. But the SEO job market is the clearest leading indicator I’ve seen of how companies are actually valuing AI skills, so I followed the data off the usual map.

946 SEO job postings show companies are willing to pay a premium for AI skills. But the signal is buried in descriptions, and the salary premium only truly activates at mid-level and above.

SEO jobs that mention AI in the title pay $113,625 at the median compared to $89,438 for jobs that don’t. That 27% gap is live in the market right now; it’s not a projection.

In this memo, I’m covering:

• Where the 25–27% AI pay premium actually shows up in SEO postings

• Why screening jobs by title filter misses 4 out of 5 of the roles paying more

• How to position your resume (or your job description if you’re a hiring manager) so the right opportunities land on your side of the table

This week, premium subscribers get a resume audit tool that translates existing SEO work into the AI-skill bullets these job descriptions are actually looking for.

Don’t miss this one… because every week you send a resume without AI evidence in the top third is a week you’re pricing yourself against an older market that doesn’t exist anymore. The AI SEO Resume Evaluator guides you to a fix.

About this data:

• 946 full-time SEO roles from SalaryGuide.com were included in this analysis, posted December 2025 through March 2026, deduped at company + job title.

• Salary midpoints from the 41.8% of roles that disclosed pay.

• “AI mention” means the title or description contains “AI,” “LLM,” “AEO,” “GEO,” “Answer Engine Optimization,” or “Generative Engine Optimization.”

Only 22.5% of teams have fully integrated SEO and AI search workflows. Is yours one of them?

Most brands are already losing in AI search without knowing it. And most marketing leaders can’t pinpoint where.

Semrush published two new playbooks to solve this, giving marketers the foundation to understand the landscape, secure internal buy-in, and start building toward execution.

Brands that treat SEO and AI visibility as separate workstreams are building on a fault line. The ones pulling ahead are making visibility an enterprise operating model.

Get the playbooks

Companies pay 27% more salary for AI skills

AI in the job title commands the bigger salary premium, but the description signal covers far more ground. Only 146 jobs carry AI in the title. 563 include it in the description. The description bucket captures 4x more roles and still delivers a 25% median salary lift over non-AI descriptions ($100,000 vs. $80,000).

The dollar deltas are $24,187 for the title bucket and $20,000 for the description bucket. Compounded across salary negotiations over a career, neither is marginal.

The AI requirement is hidden in the job description

Only 15.5% of SEO postings include AI in the title. 59.5% require it somewhere in the description. Employers are building AI into the role without putting it in the headline.

At senior levels, the pattern becomes near-universal:

• 78.3% of director/executive descriptions mention AI.

• 67.4% of manager descriptions do.

Even at mid level, 1 in 2 job postings include it.
A hangup here? Filtering job searches by AI in the title misses 80% of AI-required roles. The requirement sits in the body text, not the headline.

The AI skill premium grows with seniority

At entry level positions, AI skills in the description carry a slight negative premium (-2.3%). Employers don’t pay new grads more for knowing AI.

The signal flips at mid level (+14.3%), then compounds sharply at the management layer.

A director with AI in the description earns $35,250 more at the median than one without. Senior roles may earn more, but the premium is due AI judgment (instead of tool skills). The market pricing is applied accordingly. Junior candidates may need AI on their resume to get the interview, but getting paid more for AI skills happens at mid level and above.

If you’re a Manager or Director, the AI SEO Resume Evaluator tool (included below) lets you put your current work into the verbs and scope hiring managers at that level are writing. Upgrade to premium for access.

9+ years in, AI skills are assumed

Experience requirements tell the same story with a steeper slope: For junior 0-1 year roles, 40.9% mention AI in the description. For roles requiring 9+ years of experience, that number is 92%.

At 9+ years, AI isn’t listed as a differentiator. Instead, it’s embedded in the role definition.

The 8% of senior postings that don’t mention it are the outliers.

The market has decided, but the titles haven’t caught up

Even if the salary premium compresses later, pricing your skills against job description-level signals is still the right move today.

1/ If you’re a job candidate: screen descriptions, not titles. The title filter misses 80% of the AI-required roles and the 25–27% premium that rides with them. Put AI evidence in the top 1/3 of your resume, or it won’t register for the postings that pay more.

2/ If you’re a hiring manager: Your pay bands are already 2-tier, whether you’ve formalized it or not. Roles requiring AI pay more at the median, and most of yours don’t say so upfront. Close that gap now.

3/ Mid-career and up: This is where the premium actually compounds. If you’re 4+ years in and AI doesn’t appear in the first 1/3 of your resume, you’re pricing yourself against an outdated market.

Quote from Josh Peacock, founder of Search for Hire:

Having been on hundreds of discovery calls with companies hiring SEOs and having built out hundreds of search teams at Search for Hire, the pattern is undeniable: SEO talent is being priced on two axes now - fundamentals and AI capability. The candidates commanding a premium aren’t the ones who can use ChatGPT, they’re the ones who can build scalable systems with it. But AI without precision judgement can take you a long way in the wrong direction, fast. The real unicorns combine that build capability with deep technical skill, strategic thinking and the ability to sit in front of a client. That combination barely exists and when it does, it doesn’t stay on the market long.

Premium: Translate your SEO work into the bullets the market pays a premium for

Put this data analysis to work on your resume.

Get an evaluation of your current tasks and work, with recommendations on how to shift your resume copy to showcase your AI SEO skills.

Read more

This Memo was sent to 26,187 subscribers. Welcome to +95 new readers! Subscribe to get the free memo weekly or upgrade to Premium for the full archive, research, frameworks, and templates.

When an AI answers a question using your content, it usually cites you with a source link. What it doesn’t do, 62% of the time, is say your name. The link is there. The brand mention is not. This is what I like to call a ghost citation: the AI using your content doesn’t mention you in the answer.

This week, I’m sharing:

• Why being cited and being mentioned are 2 different outcomes that require different strategies

• Which LLMs name brands vs. which treat them as anonymous source material

• The query format and content type that produce 30x more brand mentions

A note from Kevin: I’m a big fan of HubSpot’s Marketing Against the Grain. I had Kieran, one of the co-hosts, on my Tech Bound podcast back in 2023. Now, they launched a newsletter with smart experiments, fresh perspectives, and practical lessons on what’s working right now. So, I thought I would give a friendly shoutout: Check it out.

Hertz drove +50% more bookings from AI search. Here’s the blueprint.

Leading brands are using Semrush for Enterprise to control their brand visibility and convert it into customers.

In a few months, Hertz Iceland secured executive buy-in, delivered a complete seasonal strategy, and grew bookings from AI search significantly.

Their team has a brand visibility blueprint backed by the market’s most complete digital visibility data. And now they’re scaling it year-round.

See how they did it

This analysis draws on 3,981 domains across 115 prompts, 14 countries, and 4 AI search engines (ChatGPT, Google AI Overviews, Gemini, AI Mode), using data from the Semrush AI Visibility Toolkit. Every appearance is tagged as ‘cited’ (source link present) and/or ‘mentioned’ (brand name appears in the answer text). The gap between those 2 states is the ghost citation problem.

1. 62% of your brand’s LLM citations are functionally invisible.

Most brands assume being cited means being seen. The data says otherwise.

74.9% of domains were cited, and 38.3% mentioned. 61.7% of citations are ghost citations: the domain gets a source link but zero name recognition in the answer text.

Only 13.2% of appearances convert into both a citation and a mention. Not a single domain was cited, but not mentioned at all or vice versa.

2. Every LLM shows a different behavior

The 4 AI engines treat citations and mentions in fundamentally different ways:

• Gemini names brands in 83.7% of appearances, but only generates a citation link 21.4% of the time. It operates more like a conversationalist drawing on brand knowledge.

• ChatGPT is the opposite: It cites 87.0% of the time but mentions brands in only 20.7% of answers, functioning more like an academic paper with footnotes.

• Google AI Overviews (AIOs) sit in the middle but lean toward citation.

• Google’s AI Mode offers about 17% more brand mentions than ChatGPT in its outputs, but also functions closer to an academic paper than its Gemini sibling.

For brands, this means Gemini visibility and ChatGPT visibility are not the same thing. (This data set showed clear evidence that there wasn’t much overlap with ChatGPT citations/mentions and Gemini citation/mentions for the same prompts.) Optimizing for one does not help with the other. There is no single “AI visibility metric.” There are at least 4 different behavioral systems running in parallel.

3. Strong brands get named in the text

A clear pattern emerges among domains appearing 3 or more times: Content aggregators and academic sources are cited repeatedly but almost never mentioned.

• Medium.com was cited 16 times for the same prompts across 3 different engines and named zero times.

• Wikipedia.org was cited 27 times and mentioned in only 2 answers, both times for the same conversational query (“what is the most dangerous creature in the world?”).

• Wired.com, sciencedirect.com, harvard.edu: same pattern.

Consumer brands with strong public identity get mentioned in the output at near 100%. The AI doesn’t feel the need to cite. Instead, it mentions consumer brands outright. It knows the data about the brands came from somewhere but doesn’t feel the need to explicitly say so to users. For publishers whose value proposition is information authority, this is a structural problem.

* Mention rate above 100% means the brand is named in the answer text even when not cited as a source link - the engine references the brand by name without linking to it. For values in this data set over 100%, think about being cited 10x and mentioned 10x as = 100%. If a brand is mentioned 12x and cited 10x, that’s 120%.

4. LLMs disagree on the same brand 22% of the time

454 prompt+domain combinations were tested across multiple engines. In 22% of those outputs (100 total), LLMs disagreed on whether to mention the brand:

• Instagram.com was mentioned by ChatGPT and Gemini but only cited (not named) by Google.

• Facebook.com was mentioned by Gemini in 3 out of 3 appearances.

• Google AI cited Facebook 9 out of 9 times, but named it in only 1.

The same brand, the same query, but different engines and different outcomes. This matters for measurement: A brand can appear “visible” in one engine’s data while being completely anonymous in another. Aggregate AI visibility metrics mask this divergence.

5. In-text brand mention rates vary by geography

Controlling for the LLM, country-level differences in mention rates are meaningful:

• India and Sweden show the highest mention rates (50%), suggesting more conversational or brand-forward query patterns in those markets.

• Italy, Brazil, and the Netherlands show the lowest mention rates (18–22%), with very high citation rates (82–94%).

• The UK and Canada are mid-range but above the global average.

* Note: the dataset uses localized prompts confirmed by Semrush, so language is not a confound.

Being cited and being named are not the same, and require a different approach

From this analysis, 4 takeaways stood out to me the most for brands and their content strategies:

1. Being cited means an AI is drawing on your content. Being mentioned means it is naming you. We don’t yet know enough about the implications of mentions and citations, but we can say for sure that there’s a system that decides when you’re cited vs. mentioned.

2. Your strategy must be LLM-specific. A Gemini-first strategy is different from a ChatGPT-first strategy. Any AI visibility report that aggregates across LLMs is misleading.

3. Comparative content gets brands named. Informational content feeds the machine anonymously. If the goal is brand mentions, not just citations, focus your content strategy toward evaluation, comparison, and recommendation.

4. Prompt format matters. Brands should map not just which topics they want to appear in, but specifically which phrasing patterns produce mentions vs. ghost citations. Short conversational queries and long structured queries behave like different products.

Methodology

Data source: Semrush AI Visibility Toolkit: 3,981 domain appearances across 115 prompts, 14 countries, and 4 AI search engines (ChatGPT, Google AI Overviews, Gemini, Google).

Every row in the dataset represents a domain that appeared in an AI answer. Each appearance is tagged as “cited” (the domain appears as a source link) and/or “mentioned” (the brand name appears in the answer text). The gap between those 2 states is what this analysis calls a ghost citation: the AI used your content but did not say your name.

For premium subscribers: What creates a 30x difference in brand mentions

The ghost citation problem has a solution. In fact, 2 findings in this analysis point directly to it: one about how you phrase queries, one about what type of content you produce. Together, they explain a 30x difference in mention rates on the same topic.

Premium subscribers get both findings, plus the content strategy implications.

The first 5 findings describe the problem. These 2 explain what drives it, and what to do differently.

Read more

Welcome to another Growth Intelligence Brief, where organic growth leaders discover what matters - getting insights into the bigger picture and guidance on how to stay ahead of the competition.

As a free subscriber, you’re getting the first big story. Premium subscribers get the whole brief.

Subscribe now

Today’s Growth Intelligence Brief went out to 639 (+9) marketing leaders.

This week, we’re looking at:

• New research shows that 65-85% of what people ask ChatGPT is invisible to traditional keyword tools

• Why AI Overviews are wrong far more often than Google’s scale can afford

• Shopify’s bet that the next product discovery layer will be built for AI agents, not humans

I’ll also connect the dots on what this all means for you.

Most of what people ask ChatGPT can’t be tracked in traditional keyword tools

Here’s what happened:

Semrush published a clickstream analysis of 17 months of ChatGPT usage data. The headline finding: between 65% and 85% of ChatGPT prompts have no matching keyword in Semrush’s database. They’re queries that simply don’t exist in traditional search.

Meanwhile, ChatGPT’s outbound referral traffic grew 206% year over year (Jan ‘25 vs Jan ‘26), and the share of that referral traffic going back to Google climbed from 14% to over 21%.

Why this news matters:

This data puts hard numbers to a problem: The majority of AI-driven discovery is invisible to existing measurement tools. If 65-85% of prompts don’t match any keyword, then optimizing for keywords only captures a fraction of the surface where your brand could show up.

The referral traffic growth also flips the narrative that ChatGPT is purely cannibalizing Google. It’s creating a new discovery loop where users bounce between AI and search.

My take on this:

This is the strongest evidence yet that AI search and traditional search are becoming complementary surfaces, not substitutes. The keyword-invisible prompts are where the opportunity lives: long, conversational, context-rich queries that reward brands with deep, authoritative content. The fact that ChatGPT sends a growing share of its referral traffic back to Google suggests users treat ChatGPT as a starting point for research, not a final destination. Sites need to show up in both places.

Here’s what to do:

Stop treating your keyword list as the complete map of how people find you.

1. Start monitoring your AI citations (tools like Ahrefs and Semrush now offer this).

2. Look at your analytics for referral traffic from chat.openai.com and chatgpt.com.

3. Find real user questions in sales conversations or customer research.

If these are both growing, that’s signal that your content is being cited in prompts you’ll never see in a keyword report. Build content for the questions your customers actually ask, not just the queries they type into Google.

AI Overviews are wrong a lot more than Google wants you to think

Read more

This Memo was sent to 26,074 subscribers. Welcome to +105 new readers! Subscribe to get the free memo weekly or upgrade to Premium for the full archive, research, frameworks, and templates.

Subscribe now

For years, SEOs have operated on a simple assumption: The more ground your content covers, the more likely it is to surface in AI-generated answers. In fact, every “best practice” in classic SEO content pushes you toward more: more subtopics, more sections, more words. Build the “ultimate guide.”

An analysis of 815,000 query-page pairs across 16,851 queries and 353,799 pages says otherwise:

• Fanout coverage is nearly irrelevant to citation rates

• 2 signals actually predict whether ChatGPT cites your page

• 6 concrete changes to your existing content library help

Premium subscribers get 3 additional findings: whether the density-vs-query match tradeoff shifts by vertical, heading patterns from 6.8 million H2-H4s, and what the data says about “best X” searches.

ChatGPT judges a page by its cover.

Pages with headlines that directly answer the question get cited 41% of the time. Pages with loosely related headlines drop to 29%.

I partnered with AirOps on a study of 16,851 ChatGPT queries and 353,799 pages across 10 industries. Several more findings should change how you approach AI visibility:

1. Retrieval rank is the #1 signal: A page at position 1 has a 58% chance of being cited. By position 10, that drops to 14%.

2. Do comprehensive guides still win? Not exactly. Pages covering 26-50% of ChatGPT’s fanout sub-queries get cited more than pages covering 100%.

3. Domain authority predicts nothing: Always-cited pages have lower DA than never-cited pages. Content quality is what counts.

The full report covers 20+ signals, with controlled comparisons across each. Get a head start on how you can win visibility.

Read the full report

1/ The study

AirOps ran 16,851 queries through ChatGPT 3 times each through the UI, capturing every fanout sub-query, every URL searched, every citation made, and every page scraped. Oshen Davidson built the pipeline. I analyzed the data.

Each query generates an average of 2 fanout queries. ChatGPT retrieves roughly 10 URLs per sub-search, reads through them, then selects which ones to cite. We scored how well each page’s H2-H4 subheadings matched those fanout queries using cosine similarity on bge-base-en-v1.5 embeddings. That score is what we call fanout coverage: the share of subtopics a page addresses at a 0.80 similarity threshold. (The 0.80 similarity threshold cutoff was used to decide whether a subheading counts as a match to a fanout query. Think of it as a relevance bar.)

The question: Do pages with higher fanout coverage get cited more?

You’ll find even more information in the co-written AirOps report.

2/ Density barely moves the needle

Across 815,484 rows, the relationship between fanout coverage and citation is weak.

Covering 100% of subtopics adds 4.6 percentage points over covering none. That gap shrinks further when you control for query match (how well the page’s best heading matches the original query). Among pages with strong query match (>= 0.80 cosine similarity):

Moderate coverage (26-50%) outperforms exhaustive coverage. Pages that cover everything score lower than pages that cover a quarter of the subtopics. The “ultimate guide” strategy produces worse results than a focused article that covers 2-3 related angles well.

3/ What actually predicts citation

These 2 signals dominate: retrieval rank and query match.

1/ Retrieval rank is the strongest predictor by a wide margin. A page at position 0 in ChatGPT’s web search results (the first URL returned by its search tool) has a 58% citation rate. By position 10, that drops to 14%. We ran each prompt 3 times consecutively for this analysis, and pages cited in all 3 runs have a median retrieval rank of 2.5. Pages never cited: median rank 13.

2/ Query match (cosine similarity between the query and the page’s best heading) is the strongest content signal. Pages with a 0.90+ heading match have a 41% citation rate compared to the 30% rate for pages below 0.50. Even among top-ranked pages (position 0-2), higher query match adds 19 percentage points.

Fanout coverage, word count, heading count, domain authority: all secondary. Some are flat. Some are inversely correlated.

4/ The Wikipedia exception

One site type breaks the pattern. Wikipedia has the worst retrieval rank in the dataset (median 24) and the lowest query match score (0.576). It still achieves the highest citation rate: 59%.

Wikipedia pages average 4,383 words, 31 lists, and 6.6 tables. They are encyclopedic in the literal sense. ChatGPT cites Wikipedia from deep in the search results where every other site type gets ignored.

This is density working as a signal, but at a scale no publisher can replicate. Wikipedia’s content is exhaustive, richly structured, and cross-linked across millions of topics. A 3,000-word corporate blog post with 15 subheadings is not the same thing.

5/ The bimodal reality

58% of pages retrieved by ChatGPT in this dataset are never cited. 25% are always cited when they appear. Only 17% fall in between.

The always-cited and never-cited groups look nearly identical on most content metrics: similar word counts (~2,200), similar heading counts (~20), similar readability scores (~12 FK grade), similar domain authority (~54). The on-page signals we can measure do not separate winners from losers.

What separates them is retrieval rank. Always-cited pages rank near the top when they surface. Never-cited pages rank in the bottom half. The retrieval system, whatever signals it uses internally, is the gatekeeper. Everything else is a tiebreaker.

6/ What this means for your content

Conventional SEO content writing wisdom says cover more subtopics, add more sections, build density. The data says the conventional approach produces “mixed” pages, the 17% in the middle that get cited sometimes and ignored other times.

Mixed pages have the highest word counts, the most headings, and the highest domain authority in the dataset. They are the “ultimate guides.” They are also the least reliable performers in ChatGPT.

The pages that win consistently are focused. They:

• Match the query directly in their headings,

• Tend to be shorter (the citation sweet spot is 500-2,000 words), and

• have enough structure (7-20 subheadings) to organize the content without diluting it.

Build the page that is the best answer to one question. Not the page that adequately answers twenty.

Premium subscribers receive 3 more findings:

1. If this density-vs-query match depends on the vertical

2. Common successful patterns for writing headings for AI citation (with 6.8 million H2-H4 headings analyzed), and

3. What you need to know about “Best x” type searches.

Subscribe now

Read more

AI Mode is compressing the stage where buyers compare, reject, and discover brands on their own. Our new usability study of 185 documented purchase tasks shows that 74% of AI Mode final shortlists came directly from the AI’s output - no external check, no triangulation, no second opinion.

This analysis will cover:

• How the comparison search phase has collapsed

• What this means for brands competing in categories with high competitor AI Mode saturation

• The 3 levers that determine whether your brand shows up

Premium subscribers receive exclusive screen recordings of real user purchase tasks in AI Mode, providing valuable, detailed insight into user behavior and illustrating the three core findings. These clips can be used to illustrate to stakeholders the importance of securing brand visibility in the AI Mode outputs.

The market’s most complete brand visibility data

Semrush for Enterprise is built for brands who refuse to be second.

It unifies the industry’s largest keyword, backlink, and clickstream datasets into a single connected system. Think 27B keywords, 43T backlinks, 213M prompts, and 500TB of web activity data at your fingertips.

Act on shifts first and prove impact with insights from intent to purchase. This is what visibility leadership looks like.

Explore Semrush for Enterprise

Why we conducted the study

AI transforms Search from a list of results to a list of recommendations (shortlist). Until now, we have no idea how users treat AI shortlists. Do they take it at face value or thoroughly validate it?

That’s why I partnered with Citation Labs and Clickstream Solutions to record real users and their interactions when facing high-stakes purchases. This usability study of 48 participants completing 185 major-purchase tasks reveals that AI Mode operates as a recommendation environment, not a comparison one.

In traditional search, people click through results, comparing across sources to assemble a candidate set. In AI Mode, they accept the AI’s candidates and move on. 74% of AI Mode shortlists came directly from the AI’s output with no external check. In traditional search, more than half of users built their own shortlist from scratch.

The study covers four categories (televisions, laptops, washer/dryer sets, and car insurance). Participants completed tasks using both AI Mode and traditional search in a within-subjects A/B design, producing 149 AI Mode task observations and 36 search observations. The behavioral patterns are consistent enough across categories and participants to carry weight. (Full study design is at the end.)

From Garret French, founder of Citation Labs:

“In AI Mode, buyers often use a shortlist synthesis to shortcut the cognitive effort of Standard Searching and comparing. This raises the value of onsite decision assets and third-party sources that provide AI with clear trade-offs, specific evidence, and sufficient contextual structure to describe a brand’s offering with confidence.”

From Eric Van Buskirk

The absence of narrowness frustration is the most intellectually significant finding. 15% in AI Mode vs 11% in Search, with no meaningful statistical difference. That’s the finding that rules out the obvious alternative explanation: that users accepted the AI’s shortlist because they felt trapped. They didn’t push back. They weren’t frustrated. They were satisfied. That makes the acceptance harder to dismiss.

Here’s what happened.

1. 88% of users took the AI’s shortlist outright

Across the laptop and insurance tasks, where participants used both search surfaces (classic search and AI Mode), the gap in constructing a product shortlist was stark.

Definitions:

• AI Adopted: The participant took the AI’s recommended candidates as their shortlist with no changes or external verification.

• User Built: The participant ignored the AI’s (or Search’s) suggestions and assembled their own candidate list from independent sources.

• AI Verified: The participant started with the AI’s candidates but checked them against an outside source (a retailer site, a review, a manufacturer page) before finalizing.

• Hybrid: The participant combined AI-suggested candidates with at least one candidate they found independently.

In classic search, 56% of participants built their own shortlist from multiple sources. In AI Mode, only 8 out of 147 codeable tasks produced a genuinely self-built shortlist. The user’s comparison process didn’t just shrink when using AI Mode. For most participants, it didn’t happen at all.

64% of AI Mode participants clicked nothing at all during their task. They read the AI’s text, sometimes scrolled through inline product snippets, and declared their finalists. The no-click rate varied by category:

Insurance participants delegated most heavily. Washer/dryer participants clicked the most, likely because appliance decisions involve specific physical constraints (capacity, stacking compatibility, dimensions) that the AI summary didn’t always resolve.

The 36% who did interact with individual results within AI Mode broke into 2 groups:

• About 15% of the AI Adopted group (17 of 117 participants) verified inside AI Mode: They opened inline product cards or merchant pop-ups to check a price or spec, then returned to the AI’s list.

• Others used follow-up prompts as verification tools, asking the AI for prices or narrowing by constraints.

A separate 23% of all AI Mode tasks involved at least one visit to an external website, mostly retailers (Best Buy appeared in 10 of 34 tasks with external visits) and manufacturer sites. The destination pattern matters: Users left AI Mode to confirm a candidate they’d already accepted from the AI’s list, not to find new ones.

Of the 117 participants who adopted the AI’s shortlist directly, roughly 85% showed no internal verification behavior at all. Participants who built their own lists took an average of 89 seconds longer and consulted more than twice as many sources.

• “Given that the first paragraph says Lenovo or Apple... going with that,” said one user about laptops when searching via AI Mode. Position one in the AI response was the entire decision.

• Another AI Mode user remarked: “I liked it more than anything else I’ve ever used for product searching. It made it a lot quicker to find the options.” They experienced speed as a valuable feature, not a shortcut.

In classic search, the pattern reversed. Nearly 89% of participants clicked something.

• One insurance participant clicked out to Progressive and GEICO independently, read both landing pages, consulted an Experian article, and then arrived at a shortlist.

• A laptop participant applied hardware filters and flagged a review score discrepancy: “It shows 4.6 out of 5 stars for the reviews, but when you actually click the link: not reviewed yet.” Active skepticism of aggregated data was a behavior absent from AI Mode transcripts.

2. The AI’s top pick becomes the user’s top pick 74% of the time

Just like in classic search, the top answer carries outsized weight. 74% of participants chose the item ranked first in the AI’s response as their top pick. The mean rank of the final choice was 1.35. Only 10% chose something ranked third or lower.

Position one in the AI’s output carries an outsized advantage because of where it sits: inside a curated section that typically contains 2 to 5 items, after the AI has already done the filtering. The first item is the AI’s top pick. When people engage with AI mode, we know they read almost all of the output: The first AI Mode study found users spend 50 to 80 seconds reading AI Mode output, more than double the dwell time on AI Overviews. Users are reading carefully. They just read within a set the AI already narrowed.

However, 26% of participants in this study overrode rank order. The driver: brand recognition. They spotted a brand lower on the list and preferred it regardless of where the AI placed it. TV and laptop categories saw this most, where participants arrived with existing preferences for Samsung, LG, Apple, or Lenovo. But overriding rank did not mean rejecting the AI’s output: 81% of rank-override participants still chose from the AI’s candidate set.

3. The AI’s words become the trust signal

“Travelers and USAA actually tell me how much, whereas State Farm and GEICO give percentages. Just knowing the exact amount makes me want to pick Travelers or USAA right off the bat.”

That quote captures a core pattern in AI Mode trust. The AI’s formatting shaped the decision: Dollar amounts versus percentage discounts determined which brands made the shortlist.

AI framing (37%), meaning how AI talks about the product, and brand recognition (34%) were the top 2 trust drivers in AI Mode. They run nearly even:

• Brand recognition led when participants arrived with brand preferences

• AI’s wording filled the gaps where participants didn’t already have preferences

In classic search, the dominant trust mechanism was multi-source convergence: Participants built confidence by checking whether multiple independent sources agreed about a product.

Essentially, users triangulated. One checked Progressive, then GEICO, then an Experian article. Another compared aggregated star ratings against reviews on the actual site. They were building a case from separate inputs.

That behavior was almost absent in AI Mode (5%). Instead, AI framing (how the AI worded its description of a product) and brand recognition were the top 2 trust drivers.

The split between these 2 signals tracked closely with product category:

For televisions and laptops, where most participants arrived with existing brand preferences, brand recognition dominated. For insurance and washer/dryer, where participants had less prior knowledge, AI framing dominated.

When you lack a prior view, the AI’s description becomes the trust signal. In AI Mode, the synthesis is the corroboration. Participants treated the AI’s summary as if the cross-checking had already been done for them.

The first study showed a related pattern from the supply side: AI Mode matches site type to intent, surfacing brands for transactional queries and review sites for comparisons. This study shows the demand side of the same behavior: When the AI surfaces a brand the user already knows, brand recognition drives the decision; when it doesn’t, the AI’s own framing fills that role. The site-type matching and the trust mechanism reinforce each other.

4. If you’re not in the list, you don’t exist

Purchase outcomesBrand outcomes in AI Mode concentrated heavily. For laptops, three brands captured 93% of all AI Mode final choices. In classic search, the distribution was broader: HP EliteBook variants appeared three times, ASUS once, and other brands got consideration they never received in AI Mode.

Two distinct problems emerged:

1. Brands that never appeared in the AI’s output were never considered. Participants didn’t see them, so they couldn’t evaluate them. The AI decided who made the list, not the buyer.

2. Brands that did appear but lacked recognition faced a different problem: They weren’t seriously considered. Erie Insurance showed up in AI Mode results, but multiple participants eliminated it on name recognition alone. The brand was present but hadn’t built enough awareness to survive the moment of selection. One participant dropped a brand because it lacked a hyperlink in the AI output, reading that formatting gap as a credibility signal: “There’s not even a link there.”

Another participant said when using AI Mode: “I’m already eager to believe these are good recommendations because it mentions LG and Samsung, two brands I consider very reliable.” The AI didn’t say those brands were better. The participant inferred it from familiarity.

Participants didn’t feel constrained by the narrower set. Narrowness frustration appeared in 15% of AI Mode tasks and 11% of classic search tasks, statistically indistinguishable. The option set shrank, but the feeling of having enough options didn’t change. The most skeptical AI Mode participant in the comparison set, who complained the AI kept pointing to “teen drivers, teen drivers, teen drivers,” still chose GEICO and Travelers: the consensus AI result.

5. Users leave to buy, not to research

23% of AI Mode tasks involved an external site visit, but keep in mind these prompts reflect high-stakes situations. In standard search, that figure was 67%.

The volume difference matters less than the intent difference:

• AI Mode participants who left went to retailer sites and manufacturer pages to verify a price or spec for a candidate they’d already selected.

• Standard Search participants left to discover candidates: Reddit for peer opinions, editorial review sites for expert takes, insurance aggregators for comparison.

In the first AI Overviews study, we found that high risk leads users to verify AI claims more and reference against answers from other users on UGC platforms (like Reddit).

In this study, Reddit appeared in 19% of standard search tasks and only twice across all 149 AI Mode sessions. The peer-opinion layer that shapes a large share of traditional Search barely exists in AI Mode behavior.

There’s irony in that pattern. Google leans heavily on Reddit content to train its models. However, the source that users rely on most in standard search is the one they almost never visit when the AI synthesizes those same sources for them.

The first study found the same pattern at a different scale. Across 250 sessions, clicks were “reserved for transactions:” Shopping prompts drove the highest exit share, while comparison prompts drove the lowest. The exit destinations were retailers and brand sites, not editorial or peer-opinion sources. Six months and a different task set later, the pattern holds: When users leave AI Mode, they leave to buy.

6. 3 levers: Visibility, framing, and pricing data

3 things that excite me most about the study:

First, we can apply the mental model of rankings (higher = better) to AI Mode as well. Most users choose the first product. Now, we can apply this to prompt tracking by focusing more on prompts that lead to shortlists and use our position as goalpost.

Second, trust trumps rank. We know this since the first user behavior studies I published, but this study reinforces the importance of building trust with users before they search. It’s the ultimate cheat code.

Third, we now know buyers trust AI’s recommendations. Obviously, there’s a high risk here if the AI is wrong, but seeing how quickly buyers take the AI’s recommendation also shows us how fast consumers adopt AI. It truly is the future of Search.

Keep in mind:

1/ Visibility at the model layer is the new threshold. If AI Mode doesn’t surface your brand, you have a visibility problem at the model layer. Query your own category the way a buyer would (i.e., “best car insurance for a family with a teen driver,” “best washer dryer set under $2,000”) and document which brands appear, in what order, and with what framing. Do this across multiple prompt variations. Do it regularly, because AI responses shift over time.

2/ How the AI describes you matters as much as whether it appears. Brands cited with concrete attributes (specific model, specific price, named use case) held stronger positions than brands described generically. The content on your site that the AI draws from not only affects whether you show up, but also how confidently and specifically you show up. A brand with structured pricing data, clear product specs, and explicit use cases gives the AI better material to work with.

3/ For categories with context-dependent pricing, AI Mode creates a false-confidence problem. 63% of insurance participants were rated overconfident about pricing. They accepted AI-quoted rate estimates without checking whether the figures applied to their actual state, driving record, or current insurer. They made elimination decisions based on numbers that may not have applied to them. Where shopping panels showed explicit retailer-confirmed prices (washer/dryer), 85% of participants understood pricing clearly. Where they didn’t (insurance, laptops), confusion and overconfidence filled the gap. Structured pricing data through Merchant Center feeds and schema markup is the most direct lever for brands selling physical products. For services, the lever is editorial: Make sure your landing pages and FAQ content frame pricing as conditional (”your rate depends on X, Y, Z”) so the AI has that framing to draw from.

Study design

Citation Labs and Clickstream Solutions ran this as a remote, unmoderated usability study with 48 U.S.-based participants recruited through Prolific. Each participant completed up to 4 major-purchase shortlisting tasks across televisions, laptops, washer/dryer sets, and car insurance.

The comparison between AI Mode and traditional standard search used a within-subjects A/B design: participants used both surfaces, not one or the other. Significance calculations were normalized for the exact number of participants in each group (149 AI Mode task observations, 36 standard search task observations). This matters because the groups are unequal in size, and raw percentage comparisons between them would overstate confidence without that correction.

Sessions were screen-recorded with think-aloud audio. Trained analysts annotated each recording for behavioral markers (click-through, shortlist origin, trust signals, external site visits) and qualitative markers (stated reasoning, brand mentions, frustration signals). The 185 task-level observations provide a larger analytical base than the 48-participant headcount suggests, but confidence intervals remain wider than a large-scale survey. Findings are directional, not population-level estimates.

Notes on terminology used throughout this report:

• Shortlist: The final set of brands a user would consider buying from

• AI Adopted: The participant took the AI’s recommended candidates as their shortlist with no changes or external verification.

• User Built: The participant ignored the AI’s (or Search’s) suggestions and assembled their own candidate list from independent sources. In Search, when there was no AIO present, they had no option for relying on AI suggestions.

• AI Verified: The participant started with the AI’s candidates but checked them against an outside source (a retailer site, a review, a manufacturer page, further prompting, or interaction with a panel outside the main AI text block ) before finalizing.

• Hybrid: The participant combined AI-suggested candidates with at least one candidate they found independently.

• AI framing: The specific words and structure the AI used to describe a product, such as labels like “best for affordability” or explicit price comparisons.

• Brand recognition: The user chose or eliminated a brand based on prior familiarity, not the AI’s description or any external research.

• AI trust (general): The user accepted the AI’s output as credible without citing a specific reason, such as a particular label or description.

• Source trust: The user trusted a recommendation because of where it came from, such as a retailer, manufacturer, or named publication surfaced in results.

• Multi-source convergence: The user built confidence by checking whether multiple independent sources agreed on the same recommendation.

• Rank override rate: The share of users who chose a brand other than the AI’s top-ranked option, regardless of whether they stayed within the AI’s candidate list.

Premium: See real user behavior from the study—and take action

The challenge? Getting out of our own headspace and into how real people (not SEOs and growth marketers) use these new search surfaces. The data above tells you what happened across 185 tasks, but premium subscribers get access to screen recordings below.

Each clip is a real participant thinking aloud as they build a shortlist in AI Mode. We grouped them by the finding they illustrate:

1/ “The AI said it, so I’ll go with it”
2/ “I know that brand.”
3/ “I didn’t question the numbers.”

Each of these recordings gives valuable, detailed insight into how users are experiencing brands in the AI Mode outputs, but you can also use them to illustrate to stakeholders the importance of securing brand visibility in the outputs themselves, not just LLM citations.

Read more

As a free subscriber, you’re getting the first big story. Premium subscribers get the whole brief.

Subscribe now

Today’s Growth Intelligence Brief went out to 637 (+18) marketing leaders.

This week, we’re looking at:

• New research reveals how much of ChatGPT’s citation surface sits outside standard keyword tracking

• What Bing’s latest AI Performance update means for the teams trying to close that gap

• Why Google Personal Intelligence is creating a measurement layer that third parties may never reach

I’ll also connect the dots on what this all means for you.

95% of the queries driving AI citations don’t show up in your keyword tools

Here’s what happened:

AirOps analyzed 548,534 retrieved pages across 15,000 original prompts: The headline finding is that 85% of pages ChatGPT retrieves are never cited in the final answer.

The more operationally significant finding is about fan-out: ChatGPT generates 2 or more follow-up searches on 89.6% of queries... and 95% of those fan-out queries had zero monthly search volume by traditional metrics. Zero volume search queries matter.

That gap has direct consequences, though. 32.9% of cited pages appeared only in SERPs for a fan-out query, not the original prompt. Brands tracking primary keywords only are missing nearly a third of the citation surface entirely.

AirOps found that Google rankings still carry over. Pages ranking #1 in Google were cited 3.5x more often than pages outside the top 20, and 55.8% of all cited pages ranked in the top 20 for at least one query.

Pages with 50% or greater title-to-query overlap had a 20.1% citation rate versus 9.3% for pages with less than 10% overlap - a 2.2x lift from alignment alone.

Much of their findings track with the analysis I did with Gauge on The science of how AI pays attention.

Why this news matters:

Most AEO / AI SEO strategies are built around the queries teams already track or have tracked for years. And this data says that approach accounts for roughly 2/3 of citation opportunities at best, and misses the follow-up searches that ChatGPT generates while building an answer.

Fan-out queries behave differently by intent type. Commercial queries decompose into sub-queries covering pricing, comparisons, alternatives, and features - meaning a brand optimizing only for the head term is absent from the supporting research ChatGPT does before it writes its answer.

My take on this:

The data shows how important fan-out queries are for visibility in ChatGPT. Where it gets interesting is that about ⅓ of queries would not get caught by classic keyword research because they have zero search volume. For us, this means we need to factor fan-out queries into our research process and then systematically target those keywords. Thinking one step further: how many of those queries really don’t have any searches (always hard to tell without access to GSC, and even then it can be sampled). And how would Google treat domains that have content that has no search demand from humans but from bots?

Here’s what to do:

• Build content for zero volume searches that are crucial to your brand authority

• Audit your content against the fan-out pattern, not just head terms. For your top 10-20 commercial keywords, manually prompt ChatGPT and document the follow-up searches it generates. Those sub-queries (pricing, comparisons, alternatives, “[brand] vs [competitor]”) are your actual citation surface. If you don’t have content for them, you’re invisible for ~1/3 of citations.

• Prioritize title-to-query alignment as a discrete optimization lever. AirOps found a 2.2x citation lift from title overlap alone. Review your top pages and tighten title tags to match the phrasing AI models use in their sub-queries, not just what humans type into Google.

• Don’t abandon traditional SEO to chase AI citations. Pages ranking #1 in Google still get cited 3.5x more than pages outside the top 20. The playbook isn’t “pivot to AI optimization”; it’s “keep winning in Google AND cover the fan-out surface you’ve been ignoring.”

Bing just gave brands a map of which queries drive citations to which pages. Google hasn’t.

Read more

This Memo was sent to 25,823 subscribers. Welcome to +224 new readers! Subscribe to get the free memo weekly or upgrade to Premium for the full archive, research, frameworks, and templates.

Subscribe now

In The science of how AI pays attention, I analyzed 1.2 million ChatGPT responses to understand exactly how AI reads a page. In The science of how AI picks its sources, I analyzed 98,000 citation rows to understand which pages make it into the reading pool at all.

This is Part 3.

Where Part 1 told you where on a page AI looks, and Part 2 told you which pages AI routinely considers, this one tells you what AI actually rewards inside the content it reads.

The data clarifies:

• Most AI SEO writing advice doesn’t hold at scale. There is no universal “write like this to get cited” formula - the signals that lift one industry’s citation rates can actively hurt another..

• The entity types that predict citation are not the ones being targeted. DATE and NUMBER are universal positives. PRICE suppresses citation in 5 of 6 verticals and KG-verified entities are a negative signal.

• The one writing signal that holds across all 7 verticals: Declarative language in your intro, +14% aggregate lift.

• Heading structure is binary. Commit to the right number for your vertical or use none. 3-4 headings is worse than zero in every vertical.

• Corporate content dominates. Reddit doesn’t. AI citation behavior does not mirror what happened to organic search in 2023-2024.

Premium subscribers get a SEO content audit tool that scores any page against all 3 parts of this series and generates a custom rewrite prompt.

LinkedIn is an AI citation machine. Take advantage of it.

LinkedIn ranks #2 for AI citations across ChatGPT Search, Perplexity, and Google AI Mode, appearing in 11% of AI responses. With semantic similarity scores of 0.57–0.60, AI models are closely mirroring your LinkedIn content in their answers. Are you taking advantage?

Enterprise AI Optimization lets you track and steer how your brand shows up in AI search, across your markets and domains. See what Semrush sees and stay ahead.

Read the full research

1/ Specific writing signals influence citation, while others harm it.

While The science of how AI pays attention covers parts of the page and types of writing that influences ChatGPT visibility, I wanted to understand which writing-level signals - word count, structure, language style - predict higher AI citation rates across verticals.

Approach

1. I compared high-cited pages (3+ unique prompt citations) vs low-cited across 7 writing metrics: word count, definitive language, hedging, list items, named entity density, and intro-specific signals.

2. I analyzed the first 1000 words for list item count, named entity density, intro definitive language token density, and intro number count.

Results: Across all verticals, definitive phrasing and including relevant entities matter. But most signals are flat.

What the industry patterns showed:

When splitting the data up by vertical, we suddenly see preferences:

• Total word count was strongest in CRM/SaaS (1.59x).

• Finance was an anomaly with word count: Shorter pages win (0.86x word count).

• Definitive phrases in the first 1K characters was positive for most verticals.

• Education is a signal void. Writing style explains almost nothing about citation likelihood there.

Top takeaways:

1/ There is no universal ‘write like this to get cited’ formula. For example, the signals that lift CRM/SaaS citation rates actively hurt Finance. Instead, match content format to vertical norms.

2/ The one universal rule: open with a direct declarative statement. Not a question, not context-setting, not preamble. The form is “[X] is [Y]” or “[X] does [Z].” This is the only writing instruction that holds regardless of vertical, content type, or length.

3/ LLMs “penalize” hedging in your intro. “This may help teams understand” performs worse than “Teams that do X see Y.” Remove qualifiers from your opening paragraph before any other optimization.

2/ The entity types that predict citation are not the ones being targeted.

Most AEO advice focuses on named entities as a category: pack in more known brand names, tool names, numbers. The cross-vertical entity type analysis below tells a more specific (and more useful) story.

Approach

1. Ran Google’s Natural Language API on the first 1,000 characters (about 200-250 words) of each unique URL.

2. Computed lift per entity type: % of high-cited pages with that type / % of low-cited pages.

3. Analyzed 5,000 pages across 7 verticals.

* A quick note on terminology: Google NLP classifies software products, apps, and SaaS tools as CONSUMER_GOOD, a legacy label from when the API was built for physical retail. Throughout this analysis, CONSUMER_GOOD means software/product entities.

Results: DATE and NUMBER are the most universal positive signals. Interestingly, PRICE is the strongest universal negative.

What the industry patterns showed:

• DATE is the most universal positive signal, with the exception of Finance (0.65x).

• NUMBER is the second most universal. Specific counts, metrics, and statistics in the intro consistently predict higher citation rates. Finance (0.98x) and Product Analytics (1.10x) mark the floor and ceiling of that range.

• PRICE is the strongest universal negative. Pages that open with pricing signal commercial intent. Finance is the sole exception at 1.16x, likely because price here means fee percentages and rate comparisons, which are the actual reference data financial queries are looking for.

• CONSUMER_GOOD (software/product entities) is mixed. In Healthcare, product entities signal established brands and tools. In Crypto, naming specific protocols and products is core to answering technical queries.

• PHONE_NUMBER is a positive signal in Healthcare (1.41x) and Education (1.40x). In both cases, it is almost certainly a proxy for established brands/institutions/providers with real physical presence, not a literal signal to add phone numbers to your pages.

The Knowledge Graph inversion deserves its own note here:

• The data showed that high-cited pages average 1.42 KG-verified entities vs. 1.75 for low-cited pages (lift: 0.81x).

• Pages built around well-known, KG-verified entities (major brands, institutions, famous people) tend toward generic coverage, which isn’t preferred by ChatGPT.

• High-cited pages are dense with specific, niche entities: a particular methodology, a precise statistic, a named comparison. Many of those niche entities have no KG entries at all. That specificity is what AI reaches for.

Top takeaways

1/ Add the publish date to your pages and aim to use at least one specific number in your content. That combination is the closest thing to a universal AI citation signal this dataset produced. But Finance gets there through price data and location specificity instead.

2/ Avoid opening with pricing in non-Finance verticals. Price-dominant intros correlate with lower citation rates.

3/ KG presence and brand authority do not translate to AI citation advantage. Chasing Wikipedia entries, brand panels, or KG verification is the wrong lever. Specific, niche entities (even ones without KG entries) outperform famous ones.

3/ Heading structure: Commit to one or don’t bother.

We know headings matter for citations from the previous 2 analyses. Next, I wanted to understand whether heading count predicts citation rates and whether the optimal structure varies by vertical.

Approach

1. Counted total headings per page (H1+H2+H3) across all cited URLs.

2. Grouped pages into 7 heading-count buckets: 0, 1-2, 3-4, 5-9, 10-19, 20-49, 50+.

3. Computed high-cited rate (% of URLs that are high-cited) per bucket per vertical.

Results: Including more headings in your content is not universally better. The sweet spot depends on vertical and content type. One finding holds everywhere: Strangely, 3-4 headings are worse than zero.

What the industry patterns showed:

• CRM/SaaS is the only vertical where the 20+ heading lift is confirmed: 12.7% high-cited rate at 20-49 headings vs. a 5.9% baseline. The 50+ bucket reaches 18.2%. Long structured reference pages and comparison guides with one section per tool outperform everything else here.

• Healthcare inverts most sharply. The high-cited rate drops from 15.1% at zero headings to 2.5% at 20-49 headings. A page with 30 H2s on telehealth topics signals optimization intent, not clinical authority.

• Finance peaks at 10-19 headings (29.4% high-cited rate). Structured but not exhaustive: think rate tables, regulatory breakdowns, and advisor comparison pages with moderate heading depth.

• Crypto peaks at 5-9 headings (34.7% high-cited rate). Technical documentation in this vertical tends toward dense prose with moderate navigation structure. Over-structuring breaks up the technical depth.

• Education is flat across all heading counts, which is consistent with the writing signals finding. Heading structure explains almost nothing about citation likelihood in education content.

• The 3-4 heading dead zone holds across every vertical without exception. Partial structure confuses AI navigation without providing the full benefit of a committed hierarchy.

Top takeaways:

1/ The 20+ heading finding from Part 1 is a CRM/SaaS finding, not a universal one. Applying it to Healthcare, Education, or Finance could actively suppress citation rates in those verticals.

2/ The principle that holds everywhere: Commit to structure or don’t use it. The middle ground costs you in every vertical. A fully-structured page with the right heading depth outperforms a half-structured page in every vertical.

3/ Use the optimal heading range for your vertical. Crypto: 5-9. Finance and Education: 10-19. CRM/SaaS: 20+ (with H3s). Healthcare: 0 or 5-9 at most. Long CRM reference pages with 50+ sections are the one case where maximum heading depth pays off.

4/ UGC doesn’t dominate

The “Reddit effect” reshaped organic search between 2024 and 2025. I wanted to understand whether ChatGPT cites user-generated content (Reddit, forums, reviews) at meaningful rates or whether corporate/editorial content dominates.

The common industry assumption - that AI also preferentially cites community voices - is not what we found in the data.

Approach

1. Classified these cited URLs as (1) UGC: Reddit, Quora, Stack Overflow, forum subdomains, Medium, Substack, Product Hunt, Tumblr, or (2) community/forum prefixes or corporate/editorial by domain.

2. Computed citation share per category per vertical.

3. Dataset: 98,217 citations across 7 verticals.

Results: Corporate content accounts for 94.7% of all citations. UGC is nearly invisible.

What the industry patterns showed:

• Finance is the most corporate-locked vertical at 0.5% UGC. YMYL (Your Money, Your Life) content appears to systematically suppress citations to community opinion.

• Healthcare sits at 1.8% UGC for the same structural reason. Clinical, telehealth, and HIPAA content draws almost exclusively from institutional sources.

• Crypto has the highest UGC penetration in the dataset at 9.2%. Community-generated content (Reddit technical threads, Medium tutorials, developer forum posts) answers a meaningful proportion of analyzed queries. In a fast-moving technical niche where official documentation consistently lags, community posts fill the gap.

• Product Analytics and HR Tech sit at 6.9% and 5.8% UGC. Both are verticals where Reddit comparison threads and product review communities provide genuine signal alongside corporate content.

Top takeaways:

1/ The “Reddit effect” in SEO has not translated proportionally to AI citations. In most verticals, reddit.com captures 2-5% of total citations. This finding is in-line with other industry research, including this report from Profound.

2/ For Finance and Healthcare: UGC has near-zero AI citation value. Invest in structured, authoritative corporate content with clear sourcing. Community engagement may matter for other reasons, but it does not contribute meaningfully to AI citation share in these verticals.

3/ For Crypto, Product Analytics, and HR Tech: Community presence has measurable citation value. Detailed Reddit comparison threads, technical Medium posts, and structured developer forum answers can supplement corporate content reach.

What this means for how you strategize for LLM visibility

Across all 3 parts of this study, the consistent finding is that AI citation is not primarily a writing quality problem.

Part 2 showed it is a content architecture problem: Thin single-intent pages are structurally locked out regardless of how well they’re written. This piece shows the same logic applies inside the content itself.

The aggregate writing signals table is the most important chart in this analysis. Not because it shows you what to do, but because it shows how much of what the AI SEO/GEO/AEO industry is telling you doesn’t survive cross-vertical scrutiny. Word count, list density, named entity counts… all flat or negative at the aggregate. The signals that work are vertical-specific and smaller than our industry’s consensus implies.

The meta-lesson from this analysis is that findings are vertical (and probably topic) specific, which is no different in SEO.

This part concludes the Science of AI - for now. Because the AI ecosystem is constantly changing.

Methodology

We analyzed ~98,000 ChatGPT citation rows pulled from approximately 1.2 million ChatGPT responses from Gauge.

• Gauge is extending a one-time 75% discount for Growth Memo subscribers to help them grow their AI presence. Mention Growth Memo during a live demo or use GROWTHMEMO at checkout to redeem.

Because AI behaves differently depending on the topic, we isolated the data across 7 distinct, verified verticals to ensure the findings weren’t skewed by one specific industry.

Analyzed verticals:

• B2B SaaS

• Finance

• Healthcare

• Education

• Crypto

• HR Tech

• Product Analytics

For premium subscribers: Your content audit tool is below

Premium subscribers get an easy-to-use tool to check their actual pages against all 3 parts of this series: paste a URL and get a scored audit against the:

• Ski ramp position data

• Writing signal benchmarks

• Entity type signals, and

• Heading structure norms for your vertical

Plus, once the URL is audited, you’ll get a ready-to-use prompt to take into Claude to rewrite the page based on those findings.

Read more

This Memo was sent to 25,523 subscribers. Welcome to +148 new readers! Subscribe to get the free memo weekly or upgrade to Premium for the full archive, research, frameworks, and templates.

Subscribe now

In The science of how AI pays attention, I analyzed 1.2 million ChatGPT responses to understand exactly how AI reads a page. This is Part 2.

Where Part 1 told you where on a page AI looks, this one tells you which pages AI routinely considers.

The data clarifies:

• Why ~30 domains own 67% of citations in any topic

• The page structure that earns citations across 50+ distinct queries vs. the one that gets cited once

• Whether the ski ramp from Part 1 is actually steeper or flatter in your vertical

Premium subscribers get a checklist to integrate the study results into their workflows.

Among pages ranking #1 in Google, 43.2% were cited by ChatGPT. That was 3.5 times higher than the citation rate for pages ranking beyond Google’s top 20 results.

ChatGPT retrieves about 6x more pages than it cites.

In research across 548,534 retrieved pages and 15,000 prompts, AirOps found:

• 85% of pages ChatGPT retrieved were never cited.

• ⅓ of cited pages came from fan-out queries, and 95% of those had zero search volume.

• Among pages ranking #1 in Google, 43.2% were cited by ChatGPT. That was 3.5x higher than the citation rate for pages ranking beyond Google’s top 20 results.

Ranking well helps, but it doesn’t guarantee citations.

AirOps surfaces these fan-out queries so teams can see the full search path ChatGPT uses to build an answer and act on it.

Read the full report

1/ ~30 domains own 67% of AI citations per topic.

Classic search is a winner-takes-all game. The top result gets disproportionately more clicks than the second. Is that also true for ChatGPT answers? Is the distribution of cited domains democratic or totalitarian?

Approach:

1. Compute the citation share per domain per vertical

2. Calculate the cumulative share captured by the top 10% of domains

3. Dataset: 21,482 ChatGPT citation rows, 670 unique domains, 2,344 unique URLs, 127 unique prompts

Results: The top 10 domains take 46% of all citations in a topic. The top 30 take 67%.

AI citation is slightly less concentrated than traditional organic search, but still extreme:

• Effectively, there are ~30 seats (domains) at the citation table for any given topic. Everything else is nearly invisible.

• Example: storylane.io appears as a cited source across 102 distinct prompts (unique questions asked of ChatGPT), reprise.com across 98. Even though reprise.com has more total citations (1,369 vs. storylane.io’s 968), storylane.io shows up in answers to a broader range of different questions.

We confirmed these findings in product-comparison verticals (SaaS tools, financial advisors). However, you’ll see below that the pattern is weaker in healthcare and open web topics, where no single domain dominates. Notably, the education sector receives the most AI citations of any vertical we studied.

What the industry patterns showed:

The findings above are from product comparison verticals (SaaS, financial advisors), but the pattern is weaker in healthcare and open web topics, where no single domain dominates, and stronger in the education sector.

Education is winner-take-most: the top 10% of domains capture 59.5% of all citations.

• If you are not already in the top 5-10 domains in education, achieving citation breadth is exceptionally hard.

• tefl.org alone answers 102 unique prompts and holds 18.75% of all Education citations. The next three domains (internationalteflacademy.com 7.83%, gooverseas.com 5.87%, reddit.com 5.22%) leave the top 3 controlling about 32% of all citations.

Crypto is the second most concentrated at 43.0% for the top 10%.

• A small set of technical documentation and comparison sites (alchemy.com, quicknode.com, chainstack.com) dominate Solana RPC and infrastructure queries.

• The technical nature of Solana queries means few credible sources exist; once a domain earns trust in this niche, it captures a large share.

Finance sits at 29.4% for top-10%.

• Concentration is query-type specific: Financial advisor locator pages (forfiduciary.com at 139 unique prompts, smartasset.com at 168 unique prompts) dominate city-level advisor queries.

• But the long tail of financial product queries keeps total concentration moderate.

Healthcare is the least concentrated at 13.0% for the top 10%.

• No single domain dominates. New entrants have a realistic path to citation reach.

• The citation surface is spread across hundreds of domains, each covering a small slice of telehealth, HIPAA compliance, and healthcare app queries.

CRM/SaaS and HR Tech are similarly diffuse (16.1% and 14.4% top-10%).

• These are multi-product software categories where dozens of comparison sites, review platforms, and vendor pages split citations.

• monday.com leads CRM with only 2.88% of all citations (37 unique prompts). A genuinely open competitive field

Top takeaways:

1/ Breadth of topic coverage matters more than domain authority. A single well-structured comparison page (learn.g2.com: 65 unique prompts, 495 citations) can still outperform the entire domain portfolio of a well-known brand. The goal is not to rank for one query, but to answer a cluster.

2/ Concentration reflects category maturity. Fragmentation is an opportunity. Education and Crypto have narrow, well-defined query spaces where a few authoritative sources have locked in trust. Healthcare and CRM are broad, fragmented categories where no single domain dominates. That fragmentation is your opening.

3/ Citation reach (the number of distinct prompts a domain answers) is a more useful strategic metric than raw citation count. In low-concentration verticals like Healthcare and CRM, a focused 30-50 page strategy can realistically compete for a seat at the table. In high-concentration verticals like Education and Crypto, the path is narrower: become the definitive resource on a specific sub-topic or accept that you’re fighting for scraps.

2/ The citation advantage starts at 10,000 words.

In classic Search, word count and page length are somewhat indicative of ranks, as long as the quality is high. I wondered, again, if that is also true for showing up in ChatGPT answers?

Approach:

1. Measure raw text length of every cited page

2. Group length into 7 buckets

3. For each bucket, calculate average citations per page

Results: More words do indeed correlate with more citations, but there’s a ceiling.

The 5K-to-10K jump is the largest single step - nearly 2x. Pages above 20,000 characters average 10.18 citations each vs. 2.39 for pages under 500 characters.

What the industry patterns showed:

The length effect is vertical-specific: Finance inverts it entirely. High-cited Finance pages average 1,783 words vs. 2,084 for low-cited pages - a 0.86x lift. Authoritative compact sources, rate tables, and regulatory summaries outperform comprehensive guides there. The 10,000-character rule holds for SaaS and editorial content.

Finance peaks at 5K-10K words (10.9 citations/page), then drops sharply at 10K-20K (4.92 citations/page).

• Finance also shows the steepest absolute gain: Pages under 500 words earn only 3.84 citations/page while 5K-10K pages earn 10.9, which is a 2.8x multiplier from length optimization alone.

• Very long Finance pages may dilute the citation-triggering content with redundant detail.

Education shows the clearest length-wins-everything pattern.

• Citations per page climb steadily from 1.85 (under 500 words) to 6.05 (20K+ words) with no drop-off.

Crypto and Product Analytics behave similarly to Education.

• Length consistently pays off, plateauing around the 10K-20K tier (5.34 and 4.01, respectively). Both are technical verticals where comprehensiveness signals authority.

SaaS shows the weakest length effect: Citations per page ranges from 1.06 (1K-2K words) to 2.77 (20K+ words).

• Even the longest CRM pages only get 2.77 citations per page on average.

• In this vertical, length alone does not determine citations. Format, structure, and domain authority appear more important.

Healthcare shows a moderate length effect (1.74 to 3.92 citations/page).

• But with one anomaly: 5K-10K words (2.80) underperforms vs. 2K-5K words (3.36).

• Very long Healthcare pages may include too much clinical detail that dilutes citation-triggering content.

Top takeaways:

1/ Universal finding: Very short pages (under 1K words) underperform in every vertical. The underperformance of thin content is consistent, but the reward for long content is vertical-specific.

2/ Target your length based on industry, content type, and query intent, not a universal word count. For Finance verticals: Aim for 5K-10K words. Education, Crypto, and Product Analytics: Go as long as possible. CRM/SaaS: Prioritize structure over word count.

58% of cited URLs are cited once.

When we look at the citations within a topic, we often see many pages on a domain getting cited. So, how many citations can a single page get?

Approach:

1. Count the number of unique prompts for each page

• Classify number of citations into: 1, 2-5, 6-10, 11+

• Inspect the top URLs per vertical for structural patterns

Results: On average, 67% of cited URLs appear in only one prompt.

Think of it like a footprint game. Raw citation count tells you how popular a page is. Citation breadth tells you how strategically valuable it is. An evergreen page in AI citation is not one that gets cited a lot; it is one that keeps appearing across diverse queries.

The top 4.8% of URLs (cited 10+) are all category-level comparisons or guides answering “what is it,” “who uses it,” “how to choose,” and “pricing” in a single URL.

What the industry patterns showed:

The citation pool isn’t a meritocracy of the best answer, but the degree varies sharply.

• CRM/SaaS has the highest one-hit rate at 84.7%.

• Finance produces the highest-reach evergreen pages: forfiduciary.com covers 119 unique prompts.

• Crypto generates the most concentrated evergreen pages at 55.4% in the technical tier: chainstack.com/best-solana-rpc-providers-in-2026 (63 prompts), alchemy.com/overviews/solana-rpc (62 prompts), and rpcfast.com/blog/rpc-node-providers (61 prompts). All three are comparison pages covering the Solana RPC provider landscape from slightly different angles.

• Education evergreen pages follow a different logic: tefl.org, internationalteflacademy.com, and gooverseas.com get cited broadly because they answer TEFL-adjacent queries (cost, location, certification type) from a single resource. One URL serves many query angles.

Top takeaways:

1/ Evergreen pages share consistent structural patterns: Category-level guide format (best X for 2025/2026), broad topic coverage within a single page (what is X, how to choose X, top X vendors, pricing), and explicit year anchoring in URL or title. Pages that answer a class of questions earn citation breadth.

2/ The top 5 evergreen pages in every vertical are either comparison roundups, authoritative guides, or directory/listing pages. No thin single-topic page reaches the 11+ prompt tier in any vertical.

3/ A single evergreen page covering 10+ query intents is worth more in AI citation reach than 10 single-intent pages. The ROI of comprehensive content is front-loaded: one well-built page compounds citation reach over time. The long tail exists, but the top 5% of pages capture a disproportionate share of ongoing citation activity.

4/ The ski ramp is steeper in some verticals.

The science of how AI pays attention showed that ChatGPT cites 44.2% from the top 30% of any page. Does that trend hold across different verticals?

Approach: Re-run the same positional analysis across 7 verticals with 42,460 matched citations.

Results: The trend is real but varies by topic. One number holds everywhere: The bottom 10% of any page earns 2.4-4.4% of citations, roughly a quarter of what the peak band earns. The conclusion section is nearly invisible to AI, regardless of vertical.

What the industry patterns showed:

The true peak decile across all verticals is not the very opening. The 10-20% band is where AI reads hardest in every vertical. The first 10% is typically navigation, headlines, and intro fluff that AI skips.

• Finance is the extreme case. 43.7% of citations land in the first 30% of the page. Finance pages front-load rate data, percentages, and key figures. AI grabs them and rarely reads past the halfway point.

• Healthcare and HR Tech have the flattest ramps. Useful content is distributed more evenly across those pages.

• Education peaks at the 30-40% decile rather than 10-20%, because educational content tends to bury the key answer slightly deeper after the intro.

Top takeaways:

1/ Put your most citable claims and data in the first 30% of the page - no matter what industry you’re in. Summaries and conclusions rarely get cited.

2/ For Finance brands: Front-load your thesis and statistics as much as possible.

What this means for how you build LLM visibility

The domains that own citation share didn’t get there by writing better sentences. They built pages that hold true topical authority, addressing multiple queries in one place, and then repeated that authority across enough sub-topics to hold multiple seats at the table.

Getting cited across 30, 60, or 100 distinct prompts requires a targeted content architecture: pages built around query clusters and owning entire topics rather than individual keywords. Teams that keep the traditional “one keyword, one page” model will be structurally locked out of AI citation, even if their individual pages are beautifully written.

But as the data shows, there is no universal playbook. The tactics that work for a broad CRM platform could actively harm a Finance brand.

Methodology:

We analyzed ~98,000 ChatGPT citation rows pulled from approximately 1.2 million ChatGPT responses from Gauge.

• Gauge is extending a one-time 75% discount for Growth Memo subscribers to help them grow their AI presence. Mention Growth Memo during a live demo or use GROWTHMEMO at checkout to redeem.

Because AI behaves differently depending on the topic, we isolated the data across 7 distinct, verified verticals to ensure the findings weren’t skewed by one specific industry.

Analyzed verticals:

• B2B SaaS

• Finance

• Healthcare

• Education

• Crypto

• HR Tech

• Product Analytics

To reverse-engineer the citation selection, I ran the data through several layers of analysis:

• Structural parsing: I measured the raw character length of every cited page and mapped heading hierarchies (H1s, H2s, H3s) to see how information architecture impacts visibility.

• Positional mapping: I used Jaccard sliding-window similarity to pinpoint exactly where on the page the AI extracted its answers from, down to the specific decile.

• Entity & Sentiment extraction: I ran the opening text of unique cited URLs through the Google Natural Language API to classify named entities (dates, prices, products) and used TextBlob to score sentiment, comparing the performance of corporate content against user-generated content (UGC).

For premium subscribers: Your citation audit is below.

Most teams will see the 30-domain concentration figure, recognize they’re not in it, and move on without changing how they build content. Single-intent pages will keep shipping. Topic coverage gaps will stay unaddressed.

Premium subscribers get the tool to help change that: A citation-readiness checklist that scores your current content setup against the 4 signals from this analysis - domain footprint, page length by vertical, evergreen URL structure, and positional front-loading.

Read more

This Memo was sent to 25,371 subscribers. Welcome to +199 new readers! Get the free memo weekly. Upgrade to Premium for the full archive, research, frameworks, and templates.

Subscribe now

Branded search inflates your return on ad spend (ROAS) by taking credit for demand you already own, and every input in the paid acquisition model is getting worse simultaneously.

This week, you’ll cover:

1/ The math on how branded search distorts performance reporting across the industry

2/ Why AI-driven discovery will expose this distortion faster than any audit

3/ A concrete framework for separating real acquisition from expensive demand capture.

Premium subscribers also get the Brand Tax Calculator, an interactive tool that separates your real acquisition return from branded search inflation using your own spend data, along with 3 simple tests to run before your next budget review.

Be first everywhere customers search with Semrush for Enterprise.

Billions of data points across demand, authority, search, and customer journeys. Content optimization built for maxing out your SEO and AI search performance. And the infrastructure to scale it all.

The most comprehensive brand visibility platform gets more powerful.

Explore Semrush for Enterprise

The economics of performance marketing are deteriorating, but the metric most teams use to justify their budget is hiding the problem.

Contentsquare’s 2026 analysis of 99 billion sessions shows every paid acquisition input degrading simultaneously. Yet, while ad costs rose 30% and conversion rates fell, Google’s Q4 search revenue still grew 17%.

The data points to three hidden traps in how we measure performance. More importantly, it highlights why the financial case for AI SEO gets stronger with every dollar wasted on paid clicks that bounce

1/ Last year, ad costs rose 30%. Conversion rates fell 5%.

The visitors who convert best are the ones who already know you... and the visitors you pay the most to acquire are the ones most likely to leave.

Contentsquare measured the full acquisition funnel across 9 industries, and the picture is consistent: More money in, less value out. (I touched on this study briefly for premium subscribers in this Growth Intelligence Brief).

Cost per visit climbed 9.4% in 2025 alone, adding to a 30% cumulative increase over 3 years. Conversion rates fell 5.1%.

But their analysis showed paid search bounces at 59% and paid social bounces at 65%, while organic visits have a bounce rate of about 42%. Channel-level conversion rates are brutal: 2% for paid search, 1.6% for display, 0.4% for paid social, and 1.8% for organic search.

Those bounce rates mean more than half of every paid search dollar produces a visitor who leaves without seeing a second page. Paid social is worse. Every input in the acquisition model is degrading… all at the same time.

Gallant Chen, growth advisor to companies like Shopify, DocuSign, New Relic and others:

My client’s results were similar. Typically, sometime in 1st half of 2025, most of my clients saw a decrease in overall paid search traffic (brand and non-brand) combined with corresponding increases in CPCs (e.g., 20% drop in Paid Search clicks, but 20% increase in CPCs). Basically, this was Google rolling out AI Overviews and, in doing so, ensuring they retained steady state revenue. AI Overviews decreased clicks. But the advertisers that still got clicks ended up paying more per click. So net, net, Google did not have to sacrifice revenue to go all in on AI Overviews.

I predict Google’s AI Overviews and AI Mode will continue to accelerate this. Google shows AI-generated answers on roughly 16% of search results in Q4 of 2025, according to Semrush data, and that number is climbing.

Shrinking click inventory does not necessarily shrink demand for ads, of course - but it does concentrate bidding onto fewer clicks, which drives cost per click higher.

One Contentsquare finding sharpens the problem: Repeat visitors - the 13% who return within 30 days - account for the majority of conversions on many sites. AI-referred traffic, still just 0.2% of total visits when you look at the whole picture, bounces less and converts closer to organic rates.

2/ That means you’re likely taxing your own demand.

If every acquisition input is getting worse, why do most dashboards still show paid search as the top-performing channel? Because branded search is doing the heavy lifting, and branded search is not acquisition… it’s demand capture.

Dreamdata’s analysis of B2B Google Ads accounts found that 18% of search ad budget - an estimated $47 billion - goes to branded keywords. Branded campaigns returned 1,299% ROAS versus 68% for non-branded. That gap looks like a success story until you test whether the ad caused the sale.

In 2024, Rand Fishkin explained the attribution mechanism that makes this invisible: When people hear about a brand through social, podcasts, or word of mouth, they go to Google and search the brand name. Google gets attribution credit for the conversion. CFOs look at analytics and see that the best traffic comes from Google, which reinforces the investment in Google Ads.

The more a company invests in brand-building elsewhere, the better branded search numbers look, which makes Google look like the best channel… which leads to more Google spend.

Google’s collecting the toll on conversions it had nothing to do with, and if you’re not careful about measurement, this can distort what’s actually going on. In catching up with Rex Gelb, Founder & CEO at Summit Chase and Head of Paid Media at Cursor, he mentioned:

Branded search is one of the most misunderstood metrics in performance marketing. High ROAS on brand campaigns usually reflects demand that your marketing efforts already created elsewhere. That doesn’t mean branded search is useless - it often protects conversion paths and captures high-intent traffic. The real mistake is reporting blended ROAS without separating brand and non-brand. Once you split them, the economics of acquisition become much clearer.

Gallant Chen seconds that notion:

My preferred approach is for teams to think about Brand Paid Search as an “opex” item akin to other G&A elements that, unfortunately, you must invest in to run your business. Brand Paid Search does not drive incremental revenue. Focus on NonBrand, which does drive incremental revenue.

3/ Branded spend defends 70% of search - and ignores the rest.

The brand tax would be easier to justify if Google were the only place people search… but we know it’s not. Branded keyword defense does nothing on Amazon, YouTube, Reddit, or any AI surface.

SparkToro and Datos published new research this month analyzing desktop search behavior across 41 domains:

• Roughly 80% of searches happen on traditional search engines (Google was found responsible for 73.7% of all desktop searches).

• Commerce sites account for 10% (like Amazon and eBay), social 5.5% (TikTok, Youtube), AI tools 3% (ChatGPT, Claude).

Brands are paying to defend their name on a platform that represents 70% of search, all while it’s actively shrinking (albeit slowly)… and user discovery is shifting to surfaces where the brand tax does not apply:

The one I’m most excited about is invisible — it’s the 34 sites outside the top 7 growing their share of search — one of the only areas of web behavior we’ve investigated in the last decade(?!) where the biggest sites aren’t getting more dominant with time. Fingers crossed this trend continues.

A brand that spends 90% of its paid budget on Google is optimizing for one platform in a search economy that now spans 41 and counting - 34 smaller sites outside the top 7 are the fastest-growing segment of search. That’s risky.

The math does not hold when you account for where people actually look for products, answers, and recommendations.

4/ Increased ad costs and high bounce rates make the case for AI SEO.

If influence is more valuable than traffic - and it is, although harder to measure - brands should build presence on the platforms where their audience already spends time rather than (over)paying to pull them through a branded click.

Contentsquare’s 2026 retention data supports this: Repeat visitors who return within 30 days convert at multiples higher than first-touch paid visitors. AI-referred visitors, arriving with clearer intent from upstream AI conversations, bounce less and convert closer to organic rates.

The pattern is consistent: Brand familiarity built before the click can produce better economics than paid acquisition at the click.

And this is one of the biggest financial cases for AI SEO, even when the ROI of LLM visibility is hard to quantify today.

If more than half of every paid search dollar produces a bounce - and it’s likely AI Overviews will push that number higher - then investing in brand visibility and trust inside AI answers makes financial sense for many brands.

The comparison is not “AI SEO versus proven ROI.” The comparison is “AI SEO versus a high bounce rate that is getting worse.”

A channel that builds brand recognition upstream and balances your dependency on paid demand capture does not need to prove attribution the same way a direct-response campaign does.

It needs to prove that branded search spend went down while total revenue held. And that’s a test you can run.

5/ Premium: Brand Tax Calculator and 3 tests to run before your next budget review

Your dashboard reports a 5.9x return. Your true acquisition ROAS might be 7x, or it might be 1.2x. The difference is the brand tax: the capital you spend capturing demand you already own.

The Brand Tax Calculator lets you plug in your paid search spend, branded keyword share, and reported ROAS to see the gap in dollar terms, including how much revenue your paid campaigns are cannibalizing from organic.

Subscribe now

Read more

Today’s Growth Intelligence Brief went out to 620 (+38) marketing leaders. As a free subscriber, you’re getting the first big story. Premium subscribers get the whole brief.

Subscribe now

This week, we’re looking at what a $10 experiment with 60,000 AI-generated pages revealed about how GPTBot actually behaves, new citation data from 730,000 ChatGPT conversations, and what GPT-5.3 Instant’s web-blending update means for content that depends on being retrieved.

I’ll also connect the dots on what this all means for you.

GPTBot crawls everything - and cites almost nothing.

Here’s what happened:

Metehan Yesilyurt built a 60K-page statistics website using GPT-4.1-nano for under $10, then tracked what happened next. He didn’t expect Googlebot… he expected nothing.

I did not expect GPTBot to crawl a brand-new, zero-backlink domain at the scale it did. That was the real discovery.

GPTBot showed up within minutes of deployment. In the first 12 hours, it made 29,000+ requests to a site with zero backlinks, zero social shares, and no Search Console submission. Googlebot made 11 requests in the same window. That’s a 470x difference in crawl intensity.

The site - stateglobe.com - was entirely AI-generated and deliberately thin. It was built as an experiment, not a real publishing effort. None of that stopped GPTBot from consuming it at roughly 1 request per second.

Yesilyurt also caught a detail most site owners miss: Without server-side tracking, you can’t see any of this. Bot traffic made up ~98% of all requests, but client-side analytics tools captured none of it. He verified individual bot identities against OpenAI’s published IP ranges, ruling out user-agent spoofing.

Why this news matters:

Metehan’s experiment proves the barrier to getting AI crawlers onto your site is effectively zero. GPTBot will find you whether you optimize for it or not - it found a brand new domain with no authority in minutes, and then for some reason stayed for hours.

His experiment results raise an important question: What’s the point of optimizing for AI crawlers? Does that access lead to anything?

A site crawled 30,000 times in 12 hours still gets deindexed by Google and almost certainly won’t appear in ChatGPT answers, because ingestion vs. citation operates on entirely different logic.

My take on this:

High crawl volume to your site from GPTBot is not evidence of AI visibility. It’s simply evidence that OpenAI is building its index, which will include a lot of content across the web that will never surface in a citation.

The experiment shows how hungry ChatGPT is for fresh and new content. While Google is very picky, probably to keep its index clean, ChatGPT can’t get enough. Important to note: ChatGPT crawled 78k times with the GPTBot user agent, but only 642 times with the ChatGPT-User user agent (as of March 8th). In other words, ChatGPT mostly crawls this site for model training. Not for showing it in answers.

Since the site is data-heavy (statistics), it’s prime steak for hungry AI crawlers. Now, the downside is that content used for training gets barely cited, so the question we have to ask ourselves is how much we want to be part of the training data as opposed to beachfront property for live web retrieval.

Here’s what to do:

1. Audit your logs, not your analytics: Stop relying on client-side tools like GA4 to measure AI crawler interest. You need server-side log analysis to differentiate between GPTBot (ingestion/training) and ChatGPT-User (live web retrieval).

2. Draw the line between training and retrieval: Decide if your proprietary data is something you want the model to train on. If you only want to be surfaced in live conversational answers, use your robots.txt to block GPTBot while keeping ChatGPT-User fully permitted.

3. Advance past crawl metrics: A massive crawl spike from an AI bot is not a KPI. Shift your content team’s focus from “are we being crawled?” to “are we providing answers worth citing?”

Is the citation economy winner-take-all?

Read more

This Memo was sent to 25,163 subscribers. Welcome to +159 new readers! Get the free memo weekly. Upgrade to Premium for the full archive, research, frameworks, and templates.

Subscribe now

For a while now, SERP Features have made SEO for e-commerce distinctly different from other verticals like B2B or local. And yet, most teams still measure success against classic search results. Without third-party tools, it’s hard to get a full picture from the data that Google reports. As a result, some retailers fall significantly behind while thinking they’re ahead of the game.

And so the problem remains remarkably hidden. To make it visible, I used e-commerce tracking platform Audience Key to analyze 4,000+ keywords and almost 40,000 product grids over 9 months. (Plus, premium subscribers get my playbook for product grid optimization, including lesser-known tactics).

The last satisfying mile in content ops

Most content and SEO teams have figured out how to generate output with AI. The bottleneck has moved. It’s now in the space between “draft ready” and “published”: matching CMS fields, syncing statuses across tools, and fixing formatting that breaks every time content moves between systems.

That gap is where teams lose hours per page and where ROI gets hard to prove, even when the work is solid.

AirOps is running a live session on March 18th (2 PM EST) that gets specific about closing that gap. Justina Flamme (Product Lead, AirOps) and Josephine Cahill (Web Lead, Oyster) will walk through how to map structured content directly into your CMS, what to automate first between your AI workflows and PM tools, and a simple measurement framework for tracking minutes saved per item, error reduction, and approval cycles.

If your team is producing more content than ever but publishing speed hasn’t kept up, this is the session to attend.

📅 March 18, 2026

⏰ 2:00 PM EST

📍 Live on Zoom

👉 SAVE A SPOT

Product grids get higher CTRs than organic results

Product grids aren’t just another SERP feature to track in your dashboard. They represent Google’s final transformation from search engine to shopping marketplace - a change I’ve been documenting since E-commerce shifts, where I showed how Google merged web results and shopping tabs.

There’s evidence that product grid CTRs are higher than classic search results. AWR data shows product grids cut the CTR on organic results in half.

The quantitative data confirms observations from Brodie Clark, who reports an example of up to 58% CTR on product grids. I see the same with my clients.

When I analyzed the state of e-commerce SERP features, images and product listings were already becoming dominant. But product grids take this visual transformation to its logical conclusion: They push traditional blue links so far down the page that they become secondary navigation options, not primary discovery mechanisms.

Product grids are:

• Filterable: Users can narrow by price, brand, condition, and features without leaving the SERP

• Visual-first: High-quality product images take center stage, not meta descriptions

• Dynamic: Content updates as Google crawls your Merchant Center feed, not when they re-index your page

• Commercial: They only appear for queries with shopping intent, creating a two-tier search system

Product grid placements grew 82% in 9 months

Google is allocating significantly more prime real estate to visual product feeds over traditional text results.

In May 2025, there were 1,825 total grid placements among these brands.

By February 2026, that number skyrocketed to 3,321 - an 82% increase in just 9 months.

In fact, 96% of all SERPs in this dataset show product grids!

• 40% of SERPs show only one product grid

• 32% show 2 grids

• 22% show 3

• 6% show 4 or more

And surprisingly, the number of more than one product grid in a single SERP declined by -4% over the last 9 months.

Case study: 4 brands fighting for first place

The product grid takeover is a great opportunity to show how brands can get left behind when they miss the train. The laptop query space on Google is a great example, with a case study of 4 refurbished computer hardware brands.

Discount Computer Depot is the traditional SEO powerhouse. In early 2026, they held over 87% (4.6m/5.2m) standard organic rankings in the top 3 positions.

Yet, their product grid presence is virtually non-existent, at just 2.4% (80/3,321).

Back Market has a mere 1.7% of top 3 rankings but owns 59% of the visual product grids (see chart below).

Back Market saw massive growth in grid placements, jumping from 745 in May 2025 to 1,960 in February 2026 and overtaking Newegg in late 2025.

Interestingly, their data perfectly shows the inverse correlation between legacy rank and actual modern visibility (notice that product grids are on a 2nd y-axis in the chart below and have a much lower occurrence than classic search results).

Other players, like PC Liquidation, have been able to grow their organic top 3 keyword rankings, but product grids are not following suit. The 2 types of search results can run completely independently of each other.

Other players, like Newegg, see similar trends as Back Market: Classic organic rankings decline while product grid placements grow.

The Back Market vs. Discount Computer Depot comparison reveals the new competitive landscape.

Back Market didn’t win by playing the old game better. They won by recognizing the game had changed.

Organic rank and grid presence are independent systems

Here’s what separates product grid winners from traditional SEO winners:

1/ Feed quality over content quality: Your product descriptions still matter, but your Merchant Center feed quality matters more. Clean, complete, structured data beats beautifully written prose. Google doesn’t need to parse your HTML anymore - they want machine-readable attributes.

2/ Visual assets over backlinks: A single high-quality product image on a white background can generate more clicks than a dozen referring domains. Classic SEO authority (think high-quality backlinks) will still help you rank in traditional results, but they're invisible in product grids where image quality, price competitiveness, and merchant ratings dominate.

3/ Price competitiveness over domain authority: When users can compare prices at a glance, your 15-year-old domain and DR 70 profile mean nothing. The lowest price (with acceptable shipping terms) wins. This commoditizes traffic in a way traditional SEO never did.

4/ Merchant Center optimization over on-page SEO: Product title templates, GTIN accuracy, and feed error rates are the new meta descriptions and header tags. Many e-commerce SEOs don’t even have access to their company’s Merchant Center account.

The implications extend beyond individual tactics. As I noted in How to compensate eroded traffic, when you can’t win with product keywords, you need to think horizontally: different page types, new categories, and editorial content that addresses users earlier in their journey.

Category pages win grids; product pages rarely qualify

Even when focusing my analysis only on actively ranking URLs, Google shows a massive preference for category and listing pages (e.g., /categories/apple-refurbished.html) over individual product pages.

• Category Pages: 3,367 instances (~97% of the filtered data).

• Product Pages: 90 instances (~3% of the filtered data).

While rare, Product pages (PDPs) are significantly more likely to rank for keywords that specify technical details or specific hardware models.

• Keywords like “17 inch desktop monitor,” “19 computer screen,” and “20in computer monitor” are among the top terms that successfully trigger specific product pages.

• Searches for specific builds, such as “computer dell optiplex” or “computer desktop hp i7” also lead to individual product pages rather than broad categories.

There is a strong correlation between “price-intent” keywords and the density of the product grids column in the search results.

• Maximum grid density: Keywords such as “gaming desktop price,” “lenovo laptops prices,” and “cheap laptop” consistently trigger a maximum of 4 product grids.

• Visual competition: These keywords represent high-intent comparison shoppers, and Google responds by filling the SERP with visual product grids to facilitate quick price comparisons.

The measurement gap makes the problem invisible

For years now, e-commerce SEO has split into 2 distinct disciplines:

1. Traditional SEO will increasingly focus on informational content, brand queries, and long-tail discovery - areas where product grids don’t dominate. This is where authority, content depth, and technical optimization still matter.

2. Merchant Center optimization becomes its own specialization, focused on feed quality, product data accuracy, competitive pricing strategy, and visual asset production. This looks more like marketplace management than SEO.

The most frustrating part of this transformation? We can’t properly measure it without expensive third-party tools.

Google Search Console reports on traditional organic results. Merchant Center provides product grid analytics. But there’s no unified view. You can’t answer basic questions like:

• What percentage of my search visibility comes from product grids vs. traditional results?

• How do grid placements correlate with conversion rates?

• Am I losing traditional rankings because I’m gaining grid placements, or despite it?

Even the comparison of product snippets and merchant listings in Search Console does not allow you to compare product grids against classic web results.

Premium: Your new product grid playbook

Every month your team optimizes pages without touching the Merchant Center feed, you’re investing in a surface that product grids are pushing below the fold.

The playbook below covers the specific feed attributes, structured data requirements, and merchant signals that decide grid position, plus I’ll cover the Glue ranking system Google uses to score clicks inside SERP features.

Read more

This Memo was sent to 24,956 subscribers. Welcome to +149 new readers! Get the free memo weekly. Upgrade to Premium for the full archive, research, frameworks, and templates.

Subscribe now

Most AEO “strategies” are tactic lists dressed up as long-term direction. They often break the first time a platform changes or leadership asks questions. A real AI SEO strategy starts with the business problem, builds on your brand’s unique advantages, and lets tactics come last.

This week, we’re covering:

• How to identify your actual AI SEO challenge (it’s a business problem, not a channel problem)

• A 3-part strategy document structure that survives leadership scrutiny and platform shifts

• How to present AI SEO investment using scenario planning instead of traffic forecasts

Premium subscribers also get an interactive strategy builder tool to create your AEO strategy document.

Which brands are winning in AI visibility?

Semrush’s AI Visibility Awards reveal the brands that are leading their industries in AI visibility. We’re recognizing those with the highest visibility overall, the fastest risers, and the fresh names who are overachieving.

Get insights into what sets brands like Patagonia, Anthropic, and Nothing Technology apart from their industry competitors. Then dive into the complete AI Visibility Index for more learnings, refine your strategy, and power your climb up the rankings.

Explore the winners

1. Tactics without a strategy waste quarters of work

Strategy as a concept is even more misunderstood in the AI SEO era than it was in traditional SEO. Most “AEO/GEO strategies” I see are actually just tactics: optimize for long-tail queries, add structured data, create FAQ content. These might be part of your execution, but they’re not your strategy.

The result? Teams chase citations in ChatGPT without understanding if that’s a solution to an actual business problem. They optimize for Perplexity when the real challenge is protecting branded search volume. They copy competitor tactics instead of building on their unique advantages.

When you set out to build (or repair) your AI SEO strategy, distinction matters because a tactic list can’t answer the one question strategy exists to answer: What problem are we solving?

2. Start with your brand’s unique challenge

Your strategy must answer one question first: What business problem are we solving?

This sounds obvious. Most teams skip it. They see “AI search is growing” and immediately jump to “we need to rank in ChatGPT” and start trying new tactics. That’s a reaction, not a clear strategy.

Use the same approach I outlined in creating an SEO strategy from scratch: identify your actual challenge through research, then build your approach around solving it.

Common AI SEO challenges I see:

• Brand visibility erosion. Branded queries get answered by AI without attribution, bleeding awareness over time.

• Pipeline protection. Qualified traffic is shifting to AI Mode, but your brand is invisible in those results.

• Category definition. AI models cite competitors as the category solution. Your brand doesn’t appear.

• Conversion influence decay. Users research in ChatGPT, arrive at your site decision-ready, or don’t arrive at all. The pre-site journey now happens inside an AI interface - and you can’t see your target audience’s detailed behaviors via analytics.

These are business problems, not channel problems. Your challenge should connect directly to revenue, market share, or competitive position. If it doesn’t, you’re optimizing for a metric that can’t survive a budget review.

3. Do your research first to kill your own incorrect assumptions.

You can’t build an AI SEO strategy on assumptions. What works varies by industry, query type, and user intent… and the platforms are moving and shifting fast.

Your research phase should answer 4 questions:

1/ Where is your audience using AI search? Don’t assume. Survey customers, analyze referral data, review session recordings. ChatGPT usage patterns differ from Perplexity and Google AI Overview usage. Our AI Mode user behavior study showed that 250 sessions of real behavior look nothing like what most teams expect.

2/ Which queries drive the pipeline? Map the queries that connect to revenue, not just site visits from AI Mode, Gemini, or ChatGPT & Co. In zero-click environments, you need to understand which visibility opportunities actually influence buying decisions. Start with pain points your sales team hears on calls. Turn those into the questions buyers type into ChatGPT or Google. Then check which of those questions generate AI answers where your brand does or doesn’t appear. That’s your revenue-connected query set.

3/ What kind of site content or external third-party mentions drive visibility in your category? Test which internal content structures (like types of blog posts and landing pages) and external-third party sites that mention your brand (like reddit and G2) earn citations in your category for revenue-connected queries. For your internal content that you have more control over, the ski-ramp data from The science of how AI pays attention shows 44% of citations pull from the first 30% of a page, which means front-loading claims, definitions, and data changes citation rates more than adding depth at the end. Run one test: rewrite the first 3 paragraphs of your top 10 pages to lead with the answer, not the context.

4/ What’s your citation baseline? Use tools like AirOps, Profound, or SearchGPT to map where you currently appear. Track competitors. Measure the gap.

Compare your current performance against where you need to be. Use 5x Why analysis to identify root causes. If you’re not being cited, the problem could be content depth, authority signals, or technical accessibility. Each requires a different approach.

4. Your strategy document has 3 parts

An AI SEO strategy document should include 3 components. No more.

Part 1: The challenge. State the core business problem in one sentence. Example: “Our brand is invisible in AI-generated answers for category-defining queries, allowing competitors to own mindshare with buyers before they reach a search engine.”

Part 2: The approach. Explain how you’ll address the challenge. This is where your unique advantages matter. Your approach should be something only your brand can do, or something you do better than competitors.

Example approaches:

• Authority multiplication. Leverage your executive team’s expertise through strategic bylines, podcast appearances, and research publications that AI models pick up as authoritative sources. Third-party authority signals influence brand mentions and citation selection.

• Product-led content. Use your product data to create depth that competitors can’t replicate. Apply product-led SEO principles to AI SEO by building content assets that only your data can produce.

• Community signal amplification. Build visibility through customer stories, case studies, and user-generated content that demonstrates applied expertise. Personas built from real customer data sharpen this work because they tell you which community signals actually match how your buyers search.

Part 3: The actions. Now - and only now - list your tactics. These should flow directly from your approach:

• Create conversational-query content (or update existing content) that addresses hyper-specific buyer contexts

• Optimize technical accessibility for LLM crawlers

• Build systematic digital PR to drive third-party citations

• Develop persona-specific content that matches AI search patterns (using synthetic personas to scale prompt tracking)

• Reinforce internal linking as entity maps, not just crawl paths

Include resource allocation: What percentage of capacity goes to each action area? Include success metrics tied to business outcomes, not just “track citations.” Read Budget for capacity, not output to learn more about how to do this.

At the end of this memo, premium subscribers get deeper guidance on how to define your brand’s challenge, what approach to take, and what actions to pair with your brand’s unique needs.

5. Scenario planning sells AI SEO to leadership

Here’s where AI SEO strategy gets difficult. You’re asking for investment in a channel that’s still forming, with metrics leadership doesn’t yet understand.

Don’t present traffic forecasts. They’re fiction in AI search. Use scenario planning instead.

Frame it like this: “If we allocate 30% of capacity to authority building and 20% to conversational content, we expect citation increases of 40-60% within 6 months, which should influence 15-20% of assisted conversions based on current attribution data.”

Include stage gates. Make the investment reversible. Executives are more likely to approve experiments with clear decision points than open-ended commitments.

Present 3 scenarios: conservative, moderate, and aggressive. Show what resources each requires and what outcomes they might produce. Let leadership choose.

The strategy document from section 4 gives you the structure to do this. The challenge statement defines the goal. The approach defines the bet.

6. Review your strategy quarterly or it goes stale

Your AI SEO strategy is not a one-time document. The platforms change, and user behavior is shifting fast. Your own test outcomes should also change your tactics.

Build quarterly strategy reviews into your plan. Each review should answer 4 questions:

• What changed in AI search since our last review?

• What did we learn from our tests?

• Do our tactics still serve our approach?

• Is our approach still solving the right challenge?

Your AI SEO strategy should be a decision-making tool, not a task list. Most teams fail at AI SEO because they treat it like traditional SEO with a different name and a slight shift in tactics.

Start with the business challenge. Build an approach around what only your brand can do… let your tactics flow from there.

And make the whole thing reversible and adaptable, because we’re all still learning what works.

Build your AI SEO strategy with the Growth Memo library

Once your strategy document is set, these past Growth Memo posts cover the execution layer. Each addresses a specific capability your AI SEO approach will need.

Plus, this week, premium subscribers get an AI SEO Strategy Builder tool at the bottom of this memo.

First, know your audience

Personas are critical for AI search covers how to turn in-house data into personas that shape briefs, prompts, and content decisions.

Making SEO personas actionable across teams moves personas from a planning artifact into day-to-day workflows across content, product, and SEO teams.

Synthetic personas for better prompt tracking solves the cold-start problem in prompt tracking by simulating search behavior across segments at 85% accuracy.

Second, understand user behavior in AI search

The first-ever UX study of Google’s AI Overviews tracked 70 users across 8 tasks to map what “visibility” means when AI answers sit above organic results.

What our AI Mode user behavior study reveals analyzes 250 sessions of AI Mode behavior to show how users actually interact with Google’s AI interface.

Google’s AI Mode SEO impact is the second part of that study, covering what’s measurable, what’s guesswork, and what visibility means in AI Mode.

Third, create content that builds long-term topical and brand authority

Topic-first SEO explains why keyword-first SEO creates surface-level content and cannibalization, and how topic-first thinking fixes both problems.

Operationalizing your topic-first SEO strategy is the execution blueprint for running topic-first across your team.

How to measure topical authority offers a method to quantify topical authority using Google leak signals and competitive benchmarks.

How you can track brand authority for AI search covers the difference between topical and brand authority, and how to measure brand authority with real numbers.

SEOzempic explains how less is more: less low-quality, thin pages, and more sharply targeted website content around the key topics that matter to your brand’s target audience.

And understand how AI reads and cites your content - so it influences how you create it

The science of how AI pays attention is an analysis of 1.2M search results showing exactly where AI pulls citations from and why content structure determines selection.

Internal linking grows up reframes internal linking as an entity reinforcement tool, which directly affects how AI systems understand your site’s authority.

How AI really weighs your links analyzes 35,000 datapoints on backlinks and AI visibility, with findings that should reshape your link building priorities.

The science of how AI pays attention provides data-backed insights for how your content should be written and structured to increase chances of citation.

For premium subscribers: Building your AI SEO strategy doc

Below, you’ll get the process for completing each part along with a light AI SEO Strategy Builder tool that will guide you through building a strategy document you can share with your leadership or clients.

Read more

This Memo was sent to 24,738 subscribers. Welcome to +119 new readers! Get the free memo weekly. Upgrade to Premium for the full archive, research, frameworks, and templates.

Subscribe now

AI-SEO transformation will fail at the alignment layer, not the tactics layer. 25 years of transformation research - spanning 10,800+ participants across industries - reveals that the gap between successful and failed initiatives isn’t technical skill. It’s organizational readiness.

What you’ll get:

• Why AI SEO implementation challenges are people and process problems, not technical ones

• The specific alignment failures that kill AI-SEO initiatives before tactics ever get tested

• A sequenced approach that transforms you from channel executor to organizational translator

Premium subscribers also get an AI SEO change management checklist - your 45-day plan in leading your brand into the next era of SEO - and a supportive stakeholder slide deck to educate your team.

Feature your brand in all the right answers

Discover how Semrush Enterprise helps brands like Square 10X their productivity to leave competitors behind in SEO and AI search. Our clickthrough demos let you preview:

• Advanced automations that scale work across thousands of pages

• Prompt tracking, citation analysis, and content optimization for AI search

• Enterprise reporting with customizable dashboards for different stakeholders

• Competitive intelligence powered by Semrush’s market-leading database

Consolidate your stack and be first everywhere customers search, with Semrush Enterprise.

Preview Semrush Enterprise

The underlying infrastructure of AI SEO - retrieval-augmented generation, citation selection, answer synthesis - operates on different principles than the crawl-index-rank paradigm SEO teams previously mastered. And unlike past shifts, the old playbook doesn’t bend to fit the new reality.

AI SEO is different. It’s not just an algorithm update: This is a search product change and a user behavior movement.

Our classic instinct is to respond with tactics: prompt optimization, entity markup increase, LLM-specific structured data, citation acquisition strategies.

These aren’t wrong. But long-term, it’s likely AI SEO strategies will fail, and the reason isn’t tactical incompetence or lack of staying up-to-date and flexible. It’s internal organizational misalignment.

1. AI SEO transformation fails at the alignment layer: Here’s what it looks like

Your marketing team - and your executive team - is being asked to transform their understanding of SEO during a period of unprecedented change fatigue. Those who have survived 2 decades of algorithm updates are expertly adaptable, but reeducation is required because LLMs are a new product, not just another layer of search.

And this, of course, is the alignment layer fail.

In AI SEO, misalignment has specific symptoms:

1. Conflicting definitions of success: One stakeholder wants “rankings in ChatGPT.” Another wants brand mentions. A third wants citation links. A fourth wants traffic recovery. Every experiment gets judged against a different standard, and no one has agreed which matters most or how they’ll be measured. (Although our AI Overview and AI Mode studies confirm brand mentions are more valuable than citations.)

2. Metrics mismatch with leadership expectations: Executives ask for increased traffic in a growing zero-click environment. Classic SEO reports on influence metrics; leadership sees declining sessions and questions the investment. In our December 2025 Growth Memo reader survey, 84% of respondents said they feel their current LLM visibility measurement approach is inaccurate. Teams can’t prove value because no one has agreed on how value would be proven.

3. Turf fragmentation: AI SEO touches SEO, content, brand, product, PR, and (at times) legal. Without explicit ownership and a baseline, agreed-upon understanding of your brand’s AI SEO approach, each team runs experiments in its silo. No one synthesizes learnings. Conflicting tactics cancel each other out.

4. Premature tactics without a shared foundation: This looks like “Let’s test prompts” without agreeing what success means; “Let’s scale AI content to mitigate click loss” without understanding AI-assisted versus AI-generated content limits; “Let SEO handle AI” while product, PR, and legal stay uninvolved.

5. Panic-testing instead of strategic reorientation: Teams deploy short-term tactics reactively rather than reorienting the whole ship for better long-term outcomes.

This is classic change management failure: unclear mandate, fragmented ownership, mismatched incentives. No amount of tactical excellence or smart strategy pivots can fix it.

Layering AI SEO tactics + tools on top without structured change management compounds fatigue and accelerates burnout. The “scrappy resilience” that has carried the industry in the past can’t be assumed to instantly apply to this new channel without a strategic transition.

2. Org transformation failure rates are stubbornly high - AI SEO isn’t an exception

The playbook for dramatically improving AI SEO strategy transformation success rates during a big landscape shift already exists: It’s called change management. (Lucky us - no additional guesswork needed).

A baseline understanding of organizational change management matters in the AI SEO era… because most organizational transformations fail or underperform.

Your AI-SEO initiative is no different, even if changes in SEO seem contained to your marketing and product teams and stakeholders, rather than the larger organization or brand as a whole.

I’d argue that AI SEO falls into the category of industry transformation that affects your brand and org. And from decades of research, failure and underperformance are the statistical norm for these big transitions - seasoned leaders know this already. No wonder they’re skeptical of your AI SEO plans.

One McKinsey survey found fewer than ⅓ of teams succeed at both improving performance and sustaining improvements during significant shifts. BCG’s forensic analysis of 825 executives across 70 companies found transformation success at 30%.

Multiple major consulting firms’ independent research shows that most change transformations underperform.

Assuming that tactical excellence alone will carry you - without strategic reeducation and thoughtful change management as our industry shifts - is assuming you’re the exception to the rule.

3. Structured AI-SEO change management can create an 8x success multiplier

The most powerful evidence for intentional change management for any initiatives - SEO, AI, or otherwise - comes from Prosci’s 12th Edition benchmarking study. (Prosci’s study is the largest body of change management research, covering 2K+ practitioners, 400K+ data cells, over 25 years of accumulated findings.)

The correlation between the quality of managing a big shift and your project’s success is dramatic:

The gap between excellent and poor represents a nearly 8x improvement. Even the jump from poor to fair quadruples success rates.

BCG’s 2020 analysis reinforces this from a different angle, noting 6 critical factors that increase successful transformation odds from 30% to 80%:

• Integrated strategy with clear goals: This is where a carefully crafted AI SEO strategy comes in, one that not only outlines growth goals, but also clear testing and what successful outcomes look like

• Leadership commitment from CEO through middle management: If you’re a consultant or agency, this step can’t be skipped, especially if they have an in-house team assisting in executing the strategy

• High-caliber talent deployment: Or I would argue, high-quality reeducation of existing talent - make sure all operators have a baseline shared understanding of what has changed about SEO, how LLM outputs work, what the brand’s goals are, and how it will be executed.

• Flexible, agile governance: Teams should have the ability to deal with individual challenges without losing sight of the broader goals, including removing barriers quickly.

• Effective monitoring: Establish core, agreed-upon KPIs to measure what winning would look like, and note what actions were taken when.

• Modern/updated technology: Your SEO team needs the right tools to succeed, but they also need to know how to use them effectively. Don’t skip allotting time for integration of new workflows and AI monitoring systems.

Marketing teams that treat AI-SEO simply as a technical project to execute or tactics to update are leaving an 8× multiplier on the table.

With that in mind, this week, premium subscribers get a done-for-you, undesigned slide deck that will [guide you in a plan to help your marketing team and stakeholders manage this huge shift.]

4. Your internal “people-process” problems are bigger than your “tools-tactics” tension

The most counterintuitive change-management findings for the tool-obsessed: The tech you use to carry out organizational-level changes matters far less than the people using it.

• BCG’s 2024 AI implementation study found that roughly 70% of change implementation hurdles relate to people and processes. Only about 10% of challenges were purely technical.

• A 2024 Kyndryl survey found that while 95% of senior executives reported investing in AI, only 14% felt they had successfully aligned workforce strategies.

Your brand’s ability to test, update tactics, learn AI workflows, implement structured data, and optimize for LLM retrieval is not the bottleneck you need to be concerned about.

The real concern is whether your team - leadership, cross-functional team partners, and frontline executors/operators - is aligned on what AI SEO means, why and how you’re making changes from your classic SEO approach, what success looks like, and who owns outcomes.

5. Executive sponsorship, financial stakes, and what alignment actually looks like

Active and visible executive sponsorship is the #1 contributor to change success, cited 3-to-1 more frequently than any other factor, according to 25 years of benchmarking research by Prosci. Your first step as the person leading the AI SEO charge for your brand (or across your clients) is to earn executive buy-in.

But the Head of SEO cannot transform a brand’s understanding and approach of AI SEO alone. Bain’s 2024 research emphasized that successful transformations “drive change from the middle of the organization out.”

Keep in mind, financial benefits can compound quickly: One research analysis of 600 organizations found “change accelerators” experience greater revenue growth than companies with below-average change effectiveness.

Alignment isn’t just a feeling - it’s observable. You’ll know when you get there:

• Stakeholders can talk through AI SEO without hyperfocusing on tools

• Teams agree on what to stop prioritizing (not just what to start)

• Cross-functional partners have explicit ownership stakes

Alignment isn’t happening when:

• Everyone is good with “experimenting with” or “investing in” LLM visibility but no one owns outcomes

• Success gets retroactively defined, or

• Leadership asks “what happened to traffic” when you report influence metrics

6. Your new role: From channel expert to change agent

AI-SEO elevates the SEO professional’s role from channel executor to something more challenging: organizational translator.

Noah Greenberg outlined this pretty clearly in a recent LinkedIn post - step 0 in your AI SEO transformation is to become the expert.

New responsibilities:

• Translating new, confusing AI-based search concepts into plain language (see this clever LinkedIn post by Lilly Ray as a perfect illustration)

• Educating stakeholders on the structural differences between classic search engines and LLM retrieval - guiding teams to explain why your CEO doesn’t see the same LLM output when they look up the brand vs. what you’re reporting

• Explaining the tradeoffs, not just opportunities

• Setting expectations executives won’t like at first but need to hear (traffic loss or slower growth than in years prior)

This is uncomfortable. Less direct control. More indirect influence. Higher stakes.

Your mindset - as the change agent for your clients or org - centers on three principles:

1. Honesty over confidence. What we don’t know: the precise value of an AI mention. What we do know: your brand not appearing for related topics is a measurable miss.

2. Progress over perfection. Alignment doesn’t require certainty. It requires shared uncertainty - agreeing on what you’re testing and how you’ll learn.

3. Translation over broadcasting. The same strategic message needs adaptation for ICs (how their work changes), managers (how they report success), and executives (how budgets should shift). Uniform communication fails; translated communication scales.

7. The bottom line and what to do next: A practical sequence

AI SEO is a product-level change forced on the industry - and on organic searchers - that requires a fundamentally different organizational response.

Do this in order:

1. Write the one-sentence AI SEO mandate for your org. If you can’t explain AI SEO in one sentence to leadership, you’re not ready to execute.

2. Complete a high-level SWOT. Identify where your organization has existing strengths and gaps. The Brand SEO scorecard from The great decoupling will walk you through.

3. Replace or supplement legacy KPIs. Add LLM visibility estimates alongside classic KPIs (rankings, sessions) to start the transition. Reporting both builds the case for the shift without abandoning the old model cold.

4. Name cross-functional owners explicitly. Who owns brand mentions in LLM outputs - SEO, PR, or brand? Who owns citation link acquisition - SEO or content? Ambiguity is the enemy.

5. Provide baseline education at every level. ICs need to understand how LLM retrieval differs from crawl-index-rank. Executives need to understand why slowed organic traffic or zero-click growth doesn’t mean zero impact.

6. Kill one SEO practice without a fight. Success means everyone understands why, and you don’t receive pushback. If you can’t retire one outdated tactic without internal conflict, you haven’t achieved alignment.

7. Only then change workflows and tactics. Tactics deployed on an unaligned organization waste resources and burn credibility. Tactics deployed on an aligned organization compound advantage.

For premium subscribers: Your AI SEO change management plan

Premium subscribers get 2 assets built for immediate use:

• A 45-day implementation checklist, and

• A team education deck (an unbranded, undesigned version and a Growth Memo branded version).

The checklist sequences 14 core actions across 3 phases with impact ratings, effort levels, and additional guidance for consultants or agencies.

The deck runs a ~45-minute session that takes your team from “SEO is keywords, rankings, and traffic” to “here’s how AI actually decides what to cite” - with blank slides for your own brand’s LLM audit data.

Read more

Today’s Growth Intelligence Brief went out to 567 (+33) marketing leaders.

As a free subscriber, you’re getting the first big story. Premium subscribers get the whole brief.

Subscribe now

This week, we’re looking at the infrastructure of the Agentic Web and the inversion of traditional internet economics. We’ll cover:

• Cloudflare’s “Markdown for Agents”: How the barrier to entry for agent readability just dropped to near zero.

• Contentsquare’s 2026 Benchmark: Why the traditional growth playbook of traffic and conversions is breaking.

• WebMCP: How standardizing agent actions is turning technical SEO into tool optimization.

• The SEO/AEO Landscape: New AI prompt data in the SSI and the impact of the recent Gemini 3 update.

I’ll also connect the dots on what this all means for you.

Cloudflare just made every website agent-ready by default

Here’s what happened:

Cloudflare launched “Markdown for Agents,” a feature that automatically converts HTML pages into clean Markdown when an AI agent requests it. Any website using Cloudflare can enable this at the network level, no code changes required.

When an AI system sends a request with a text/Markdown content negotiation header, Cloudflare’s edge network converts the HTML response to Markdown on the fly. Markdown burns fewer tokens than raw HTML, which means agents can consume more of your content within their context windows and process it faster.

Why this news matters:

This is infrastructure-level plumbing for the agentic web, and it matters because Cloudflare powers roughly 20% of all websites. With a single toggle, millions of sites can become dramatically easier for AI agents to read, parse, and act on.

Until now, making your site “agent-friendly” required deliberate technical work: building APIs, cleaning up markup, publishing structured data. Cloudflare just commoditized the first layer of that effort. The barrier to entry for agent readability just dropped to near zero, which means the sites that don’t enable it will stand out for the wrong reasons.

My take on this:

This is a quiet but significant shift in who controls the agent experience layer. Cloudflare is positioning itself as the translation layer between the human web and the agent web, sitting between your origin server and every bot that wants to read your content.

Think about what this means strategically: your CDN provider now influences how AI agents perceive your brand. If Cloudflare’s Markdown conversion strips context, misreads your page hierarchy, or drops critical structured data during the conversion, the agent gets a degraded version of your content and you may never know it.

The bigger implication is competitive. In last week’s brief, we talked about how the agentic web infrastructure is complete: eyes (Clarity), hands (Auto Browse), and wallet (UCP/ACP). Cloudflare just added “translation” to that stack. Sites that serve clean, token-efficient content to agents will get consumed more thoroughly. Sites that serve bloated HTML will get truncated or skipped when the agent hits its context window limit.

We’re entering a world where your CDN configuration is a growth lever.

I welcome the change. I’ve tested Markdown pages with clients and saw clear citation growth. Plus, Markdown is much more token-friendly (and therefore, more environmentally-friendly too). Agents are not incentivized to send traffic to websites anyway, so why not make it easier for them to parse your content?

What I do expect, though, are more safety guards against cloaking. Nothing prevents you from serving a Markdown version of your page that has much more content (and potentially toxic prompt injections).

Here’s what to do:

1. Enable Markdown for Agents today. If you’re on Cloudflare, turn it on. It’s a free, zero-risk way to make your site more consumable by every major LLM and agent.

2. Test the output. Request your own pages with the text/Markdown accept header and audit what the agent actually sees. Check whether your key content, CTAs, product information, and structured data survive the conversion intact.

3. Don’t treat this as “done.” Cloudflare’s auto-conversion is a baseline, not a strategy. The real winners will layer purpose-built agent content (JSON-LD, dedicated API endpoints, tool schemas) on top of this foundation.

4. If you’re not on Cloudflare, talk to your CDN or hosting provider about equivalent capabilities. This is going to become table stakes fast.

The web is getting more expensive and less effective

Read more

This Memo was sent to 24,563 subscribers. Welcome to +120 new readers! Get the free memo weekly. Upgrade to Premium for the full archive, research, frameworks, and templates.

This week, I share my findings from analyzing 1.2 million ChatGPT responses to answer the question of how to improve your chances of getting cited.

Topic Compass: Find the “known unknowns” in your topic

Topic Compass maps what your audience associates with any topic using semantic analysis, so you can uncover adjacent concepts and plan content that builds topical authority.

Run a search > export the map into a brief > ship faster.

3 free searches/day to start.

Growth Memo readers get an exclusive 10% off during launch month: Use MEMO10 in checkout (offer expires March 16th).

Discover new content angles

For 20 years, SEOs have written ‘ultimate guides’ designed to keep humans on the page. We write long intros. We drag insights all along through the draft and into the conclusion. We build suspense to the final CTA.

The data shows that this style of writing is not ideal for AI visibility.

After analyzing 1.2M verified ChatGPT citations, I found a pattern so consistent it has a P-Value of 0.0: the “ski ramp.” ChatGPT pays disproportionate attention to the top 30% of your content. Further, I found 5 clear characteristics of content that gets cited. To win in the AI era, you need to start writing like a journalist.

1/ Which sections of a text are most likely to be cited by ChatGPT?

There isn’t much known about which parts of a text LLMs cite. We analyzed 18,012 citations and found a “ski ramp” distribution.

1. 44.2% of all citations come from the first 30% of text (the intro). The AI reads like a journalist. It grabs the “Who, What, Where” from the top. If your key insight is in the intro, the chances it gets cited are high.

2. 31.1% of citations come from the 30-70% of a text (the middle). If you bury your key product features in paragraph 12 of a 20-paragraph post, the AI is 2.5x less likely to cite it.

3. 24.7% of citations come from the last third of an article (the conclusion). It proves the AI does wake up at the end (much like humans). It skips the actual footer (see the 90-100% drop-off), but it loves the “Summary” or “Conclusion” section right before the footer.

Possible explanations for the ski ramp pattern are training and efficiency:

• LLMs are trained on journalism and academic papers, which follow the “BLUF” (Bottom Line Up Front) structure. The model learns that the most “weighted” information is always at the top.

• While modern models can read up to 1 million tokens for a single interaction (~700-800K words), they aim to establish the frame as fast as possible, then interpret everything else through that frame.

18K out of 1.2M citations gives us all the insight we need. The P-Value of this analysis is 0.0, meaning it’s statistically indisputable. I split the data into batches (randomized validation splits) to demonstrate the stability of the results.

• Batch 1 was slightly flatter but batches 2, 3, and 4 are almost identical.

• Conclusion: Because batches 2, 3, and 4 locked onto the exact same pattern, the data is stable across all 1.2M citations.

While these batches confirm the macro-level stability of where ChatGPT looks across a document, they raise a new question about its granular behavior: Does this top-heavy bias persist even within a single block of text, or does the AI’s focus change when it reads more deeply? Having established that the data is statistically indisputable at scale, I wanted to “zoom in” to the paragraph level.

A deep analysis of 1,000 pieces of content with a high amount of citations shows 53% of citations come from the middle of a paragraph. Only 24.5% come from the first and 22.5% from the last sentence of a paragraph. ChatGPT is not “lazy” and only reads the first sentence of every paragraph. It reads deeply.

Takeaway: You don’t need to force the answer into the first sentence of every paragraph. ChatGPT seeks the sentence with the highest “information gain” (the most complete use of relevant entities and additive, expansive information), regardless of whether that sentence is first, second, or fifth in the paragraph. Combined with the ski ramp pattern, we can conclude that the highest chances for citations come from the paragraphs in the first 20% of the page.

2/ What makes ChatGPT more likely to cite chunks?

We know where in content ChatGPT likes to cite from, but what are the characteristics that influence citation likelihood?

The analysis shows 5 winning characteristics:

1. Definitive language

2. Conversational question-answer structure

3. Entity richness

4. Balanced sentiment

5. Simple writing

Also, at the end of this memo, premium subscribers get 2 tools to operationalize these findings: a training deck your content leads can customize and run in their next team meeting, and a checklist your writers can use on every page they touch.

1. Definitive vs. vague language

Citation winners are almost 2x more likely (36.2% vs 20.2%) to contain definitive language (“is defined as,” “refers to”). The language citation doesn’t have to be a definition verbatim, but the relationships between concepts have to be clear.

Possible explanations for the impact of direct, declarative writing:

• In a vector database, the word “is” acts as a strong bridge connecting a subject to its definition. When a user asks “What is X?”, the model searches for the strongest vector path, which is almost always a direct “X is Y” sentence structure.

• The model tries to answer the user immediately. It prefers a text that allows it to resolve the query in a single sentence (Zero-Shot) rather than synthesizing an answer from 5 paragraphs.

Takeaway: Start your articles with a direct statement.

• Bad: “In this fast-paced world, automation is becoming key...”

• Good: “Demo automation is the process of using software to...”

2. Conversational writing

Text that gets cited is 2x more likely (18% vs. 8.9%) to contain a question mark. When we talk about conversational writing, we mean the interplay between questions and answers.

Start with the user’s query as a question, then answer it immediately. For example:

• Winner Style: “What is Programmatic SEO? It is...”

• Loser Style: “In this article, we will discuss the various nuances of...”

78.4% of citations with questions come from headings. The AI is treating your H2 tag as the user prompt and the paragraph immediately following it as the generated response.

Example loser structure:

• <h2>The History of SEO</h2> (Abstract Topic)

• <p>It began in the early 90s...</p>

Example winner structure (The 78%):

• <h2>When did SEO start?</h2> (Literal Query)

• <p>SEO started in...</p> (Direct Answer)

The reason that specific example wins is because of what I call “entity echoing:” The header asks about SEO, and the very first word of the answer is SEO.

3. Entity richness

Normal English text has an “entity density” (that is, contains proper nouns like brands, tools, people) of ~5-8%. Heavily cited text has an entity density of 20.6%!

• The 5-8% figure is a linguistic benchmark derived from standard corpora like the Brown Corpus (1 million words of representative English text) and the Penn Treebank (Wall Street Journal text).

Example:

• Loser sentence: “There are many good tools for this task.” (0% Density)

• Winner sentence: “Top tools include Salesforce, HubSpot, and Pipedrive.” (30% Density)

LLMs are probabilistic. Generic advice (”choose a good tool”) is risky and vague, but a specific entity (”choose Salesforce”) is grounded and verifiable. The model prioritizes sentences that contain “anchors” (entities) because they lower the perplexity (confusion) of the answer.

A sentence with 3 entities carries more “bits” of information than a sentence with 0 entities. So, don’t be afraid of namedropping (yes, even your competitors).

4. Balanced sentiment

In my analysis, the cited text has a balanced Subjectivity score of 0.47. The Subjectivity Score is a standard metric in Natural Language Processing (NLP) that measures the amount of personal opinion, emotion, or judgment in a piece of text.

The score runs on a scale from 0.0 to 1.0:

• 0.0 (Pure Objectivity): The text contains only verifiable facts. No adjectives, no feelings. Example: “The iPhone 15 was released in September 2023.”

• 1.0 (Pure Subjectivity): The text contains only personal opinions, emotions, or intense descriptors. Example: “The iPhone 15 is an absolutely stunning masterpiece that I love.”

AI doesn’t want dry Wikipedia text (0.1), nor does it want unhinged opinion (0.9). It wants the “analyst voice.” It prefers sentences that explain how a fact applies, rather than just stating the stat alone.

The “winning” tone looks like this (Score ~0.5): “While the iPhone 15 features a standard A16 chip (Fact), its performance in low-light photography makes it a superior choice for content creators (Analysis/Opinion).“

5. Business-grade writing

Business-grade writing (think The Economist or HBR) gets more citations. “Winners” have a Flesch-Kincaid score of 16 (college level) compared to the “losers” with 19.1 (Academic / PhD level).

Even for complex topics, complexity can hurt. A grade 19 score means sentences are long, winding, and filled with multisyllable jargon. The AI prefers simple subject-verb-object structures with short to moderately long sentences, because they are easier to extract facts from.

Conclusion

The “ski ramp” pattern quantifies a misalignment between narrative writing and information retrieval. The algorithm interprets the slow reveal as a lack of confidence. It prioritizes the immediate classification of entities and facts.

High-visibility content functions more like a structured briefing than a story.

This imposes a “clarity tax” on the writer. The winners in this dataset rely on business-grade vocabulary and high entity density, disproving the theory that AI rewards “dumbing down” content (with exceptions).

We’re not only writing robots… yet. But the gap between human preferences and machine constraints is closing. In business writing, humans scan for insights. By front-loading the conclusion, we satisfy the algorithm’s architecture and the human reader’s scarcity of time.

Methodology

To understand exactly where and why AI cites content, we analyzed the code.

All data in this research comes from Gauge.

• Gauge provided roughly 3 million AI answers from ChatGPT, alongside 30 million citations. Each citation URL’s web content was scraped at the time of answer to provide direct correlation between the true web content and the answer itself. Both raw HTML and plaintext were scraped.

• Gauge is extending a one-time 75% discount for Growth Memo subscribers to help them grow their AI presence. Gauge will pull a custom assessment of your brand if you book a live demo. Mention Growth Memo during the call or use GROWTHMEMO at checkout to redeem.

1. The Dataset

We started with a universe of 1.2 million search results and AI-generated answers. From this, we isolated 18,012 verified citations for positional analysis and 11,022 citations for “linguistic DNA” analysis.

• Significance: This sample size is large enough to produce a P-Value of 0.0 (p < 0.0001), meaning the patterns we found are statistically indisputable.

2. The “Harvester” Engine

To find exactly which sentence the AI was quoting, we used semantic embeddings (a Neural Network approach).

• The Model: We used all-MiniLM-L6-v2, a sentence-transformer model that understands meaning, not just keywords.

• The Process: We converted every AI answer and every sentence of the source text into 384-dimensional vectors. We then matched them using cosine similarity.

• The Filter: We applied a strict similarity threshold (0.55) to discard weak matches or hallucinations, ensuring we only analyzed high-confidence citations.

3. The Metrics

Once we found the exact match, we measured 2 things:

• Positional Depth: We calculated exactly where the cited text appeared in the HTML (e.g., at the 10% mark vs. the 90% mark).

• Linguistic DNA: We compared “winners” (cited intros) vs. “losers” (skipped intros) using Natural Language Processing (NLP) to measure:

  • Definition Rate: Presence of definitive verbs (is, are, refers to).

  • Entity Density: Frequency of proper nouns (brands, tools, people).

  • Subjectivity: A sentiment score from 0.0 (Fact) to 1.0 (Opinion).

For premium subscribers: Your implementation kit is below.

Reading this memo won’t change your citations or writing practices. But training your writers will.

Most teams will read the ski ramp data, nod, and keep writing the same way.

Premium subscribers get the tools to break that pattern: a ready-for-you-slide training deck built for team meetings and a checklist that turns this research into a repeatable content audit.

Read more

This Memo was sent to 24,402 subscribers. Welcome to +107 new readers! Get the free memo weekly. Upgrade to Premium for the full archive, research, frameworks, and templates.

We all know prompt tracking is directional. The most effective way to reduce noise is to track prompts based on personas.

This week, I’m covering:

• Why AI personalization makes traditional “track the SERP” models incomplete and how synthetic personas fill the gap

• The Stanford validation data showing 85% accuracy at 1/3 the cost and how Bain cut research time by 50-70%

• The 5-field persona card structure and how to generate 15-30 trackable prompts per segment across intent levels

Premium subscribers get a production-ready script to build synthetic personas from your support tickets, CRM data, and reviews.

AI is becoming the new first page of Google

AI tools like ChatGPT, Gemini, and Perplexity don’t invent authority. They repeat what credible websites already say.

That’s why we launched Branded Web Mentions for AI visibility.

A new service from dofollow.com that earns authoritative, contextual brand mentions on real SaaS + industry publications — so you show up more often in AI-generated recommendations and comparisons.

• Earned brand mentions on credible publications

• Context that reinforces your category + topical expertise

• Long-term brand recognition inside AI-generated answers and comparisons

Trusted by Surfshark, Pitch, and Experian.

Book a call.

A big difference between classic and AI search is that the latter delivers highly personalized results.

• Every user gets different answers based on their context, history, and inferred intent.

• The average AI prompt is ~5x longer than classic search keywords (23 words vs. 4.2 words), conveying much richer intent signals that AI models use for personalization.

• Personalization creates a tracking problem: You can’t monitor “the” AI response anymore because each prompt is essentially unique, shaped by individual user context.

Traditional persona research solves this - you map different user segments and track responses for each - but it creates new problems. It takes weeks to conduct interviews and synthesize findings.

By the time you finish, the AI models have changed. Personas become stale documentation that never gets used for actual prompt tracking.

Synthetic personas fill the gap by building user profiles from behavioral and profiling data: analytics, CRM records, support tickets, review sites. You can spin up hundreds of micro-segment variants and interact with them in natural language to test how they’d phrase questions.

Most importantly: They are the key to more accurate prompt tracking because they simulate actual information needs and constraints.

The shift: Traditional personas are descriptive (who the user is), synthetic personas are predictive (how the user behaves). One documents a segment, the other simulates it.

Example: Enterprise IT buyer persona with job-to-be-done “evaluate security compliance” and constraint “need audit trail for procurement” will prompt differently than an individual user with the job “find cheapest option” and constraint “need decision in 24 hours.”

• First prompt: “enterprise project management tools SOC 2 compliance audit logs”

• Second prompt: “best free project management app”

• Same product category, completely different prompts. You need both personas to track both prompt patterns.

Build personas with 85% accuracy for 1/3 of the price

Stanford and Google DeepMind trained synthetic personas on 2-hour interview transcripts, then tested whether the AI personas could predict how those same real people would answer survey questions later.

• The method: Researchers conducted follow-up surveys with the original interview participants, asking them new questions. The synthetic personas answered the same questions.

• Result: 85% accuracy. The synthetic personas replicated what the actual study participants said.

• For context, that’s comparable to human test-retest consistency. If you ask the same person the same question 2 weeks apart, they’re about 85% consistent with themselves.

The Stanford study also measured how well synthetic personas predicted social behavior patterns in controlled experiments - things like who would cooperate in trust games, who would follow social norms and would share resources fairly.

The correlation between synthetic persona predictions and actual participant behavior was 98%. This means the AI personas didn’t just memorize interview answers; they captured underlying behavioral tendencies that predicted how people would act in new situations.

Bain & Company ran a separate pilot that showed comparable insight quality at 1/3 the cost and 1/2 the time of traditional research methods. Their findings: 50-70% time reduction (days instead of weeks) and 60-70% cost savings (no recruiting fees, incentives, transcription services).

The catch: These results depend entirely on input data quality. The Stanford study used rich, 2-hour interview transcripts. If you train on shallow data (just pageviews or basic demographics), you get shallow personas. Garbage in, garbage out.

How to build synthetic personas for better prompt tracking

Building a synthetic persona has 3 parts:

1. Feed it with data from multiple sources about your real users: call transcripts, interviews, message logs, organic search data.

2. Fill out the Persona Card - the 5 fields that capture how someone thinks and searches.

3. Add metadata to track the persona’s quality and when it needs updating.

The mistake most teams make: trying to build personas from prompts. This is circular logic - you need personas to understand what prompts to track, but you’re using prompts to build personas. Instead, start with user information needs, then let the persona translate those needs into likely prompts.

Data sources to feed synthetic personas:

The goal is to understand what users are trying to accomplish and the language they naturally use:

1. Support tickets and community forums: Exact language customers use when describing problems. Unfiltered, high-intent signal.

2. CRM and sales call transcripts: Questions they ask, objections they raise, use cases that close deals. Shows decision-making process.

3. Customer interviews and surveys: Direct voice-of-customer on information needs and research behavior.

4. Review sites (G2, Trustpilot, etc.): What they wish they’d known before buying. Gap between expectation and reality.

5. Search Console query data: Questions they ask Google. Use regex to filter for question-type queries: (?i)^(who|what|why|how|when|where|which|can|does|is|are|should|guide|tutorial|course|learn|examples?|definition|meaning|checklist|framework|template|tips?|ideas?|best|top|lists?|comparison|vs|difference|benefits|advantages|alternatives)\b.* (I like to use the last 28 days, segment by target country)

Persona card structure (5 fields only - more creates maintenance debt):

These 5 fields capture everything needed to simulate how someone would prompt an AI system. They’re minimal by design. You can always add more later, but starting simple keeps personas maintainable.

1. Job-to-be-done: What’s the real-world task they’re trying to accomplish? Not “learn about X” but “decide whether to buy X” or “fix problem Y.”

2. Constraints: What are their time pressures, risk tolerance levels, compliance requirements, budget limits, and tooling restrictions? These shape how they search and what proof they need.

3. Success metric: How do they judge “good enough?” Executives want directional confidence. Engineers want reproducible specifics.

4. Decision criteria: What proof, structure, and level of detail do they require before they trust information and act on it?

5. Vocabulary: What are the terms and phrases they naturally use? Not “churn mitigation” but “keeping customers.” Not “UX optimization” but “making the site easier to use.”

Specification requirements:

This is the metadata that makes synthetic personas trustworthy; it prevents the “black box” problem.

When someone questions a persona’s outputs, you can trace back to the evidence.

These requirements form the backbone of continuous persona development. They keep track of changes, sources, and confidence in the weighting.

• Provenance: Which data sources, date ranges, and sample sizes were used (e.g., “Q3 2024 Support Tickets + G2 Reviews”).

• Confidence score per field: A High/Medium/Low rating for each of the 5 Persona Card fields, backed by evidence counts. (e.g., “Decision Criteria: HIGH confidence, based on 47 sales calls vs. Vocabulary: LOW confidence, based on 3 internal emails”).

• Coverage notes: Explicitly state what the data misses (e.g., “Overrepresents enterprise buyers, completely misses users who churned before contacting support”).

• Validation benchmarks: 3-5 reality checks against known business truths to spot hallucinations. (e.g., “If the persona claims ‘price’ is the top constraint, does that match our actual deal cycle data?”).

• Regeneration triggers: Pre-defined signals that it’s time to re-run the script and refresh the persona (e.g., a new competitor enters the market, or vocabulary in support tickets shifts significantly).

Where synthetic personas work best

Before you build synthetic personas, understand where they add value and where they fall short.

High-value use cases:

• Prompt design for AI tracking: Simulate how different user segments would phrase questions to AI search engines (the core use case covered in this article)

• Early-stage concept testing: Test 20 messaging variations, narrow to top 5 before spending money on real research

• Micro-segment exploration: Understand behavior across dozens of different user job functions (enterprise admin vs. individual contributor vs. executive buyer) or use cases without interviewing each one

• Hard-to-reach segments: Test ideas with executive buyers or technical evaluators without needing their time

• Continuous iteration: Update personas as new support tickets, reviews, and sales calls come in

Crucial limitations of synthetic personas you need to understand:

• Sycophancy bias: AI personas are overly positive. Real users say “I started the course but didn’t finish.” Synthetic personas say “I completed the course.” They want to please.

• Missing friction: They’re more rational and consistent than real people. If your training data includes support tickets describing frustrations or reviews mentioning pain points, the persona can reference these patterns when asked - it just won’t spontaneously experience new friction you haven’t seen before.

• Shallow prioritization: Ask what matters and they’ll list 10 factors as equally important. Real users have a clear hierarchy (price matters 10x more than UI color).

• Inherited bias: Training data biases flow through. If your CRM underrepresents small business buyers, your personas will too.

• False confidence risk: The biggest danger. Synthetic personas always have coherent answers. This makes teams overconfident and skip real validation.

Operating rule: Use synthetic personas for exploration and filtering, not for final decisions. They narrow your option set. Real users make the final call.

Solving the cold start problem for prompt tracking

Synthetic personas are a filter tool, not a decision tool. They narrow your option set from 20 ideas to 5 finalists. Then you validate those 5 with real users before shipping.

For AI prompt tracking specifically, synthetic personas solve the cold-start problem. You can’t wait to accumulate 6 months of real prompt volume before you start optimizing. Synthetic personas let you simulate prompt behavior across user segments immediately, then refine as real data comes in.

Where they’ll cause you to fail is if you use them as an excuse to skip real validation. Teams love synthetic personas because they’re fast and always give answers. That’s also what makes them dangerous. Don’t skip the validation step with real customers.

Premium: Synthetic persona script

Premium subscribers get a production-ready script that builds synthetic personas from your actual user data, along with detailed guidance on how to generate prompts from the persona cards, as well as tracking structure for prompt monitoring.

The script handles 3 things for you: data synthesis (feed it your data!), persona card generation, and prompt generation.

Run the script in Google Colab. Update the core documents when your data changes, and personas stay current instead of stale documentation.

Upgrade to Premium to get the script + full advisory-level guidance.

Read more

As a free subscriber, you’re getting the first big story. Premium subscribers get the whole brief.

Subscribe now

Today’s Growth Intelligence Brief went out to 534 (+28) marketing leaders.

This week, we’re looking at the infrastructure of the Agentic Web coming together across major Search and Commerce updates from Google and OpenAI, and how Microsoft Clarity finally lets us track AI agent visitors.

I’ll also connect the dots on what this all means for you.

Microsoft Clarity is offering us some clarity

Here’s what happened:

Microsoft Clarity introduced reporting that shows AI bot traffic and activity across websites, offering more transparency into how automated agents crawl and interact with content. (And Growth Memo readers everywhere rejoiced!)

The new “Bot Activity” report is a dashboard that tracks server-side signals to show exactly how AI agents access your site.

• Unlike standard analytics that track human visits, this requires a CDN or server integration to capture the “upstream” activity, meaning the scraping and crawling that happens before a user ever sees an answer.

• The report breaks down traffic by “Bot Operator” (e.g., OpenAI, Anthropic, Google) and, crucially, “Bot Activity” type, distinguishing between an “AI Crawler” (scraping for training data) and an “AI Assistant” (fetching a live answer for a user).

Why this news matters:

AI bots don’t behave like search bots; they don’t just index content, they consume it. What was once hidden in complex server logs is now visible, letting us easily track whether an AI is reading our site 10,000 times a day or ignoring it completely.

This dashboard democratizes the “AI Request Share” metric, allowing us to quantify how much of our infrastructure is serving non-human agents versus actual customers without needing a data science team. It effectively separates “Training” (extractive) from “Inference” (potential visibility).

My take on this:

It’s okay not to get traffic. That’s the new reality we have to accept. The real breakthrough here is that now it’s much easier to know whether an AI actually used your content as the basis for those answers.

Previously, “Zero Click” was a black box; we had to guess if our content was fueling the AI’s response or if we were just being ignored. Now, we have proof. If you see high AI consumption of your content, you know you are winning mindshare and influencing the answer, even if you aren’t getting the click. This metric finally validates the strategy of “feeding the bot” to maintain brand relevance in a world where the user might never leave the chat interface.

Here’s what to do:

• Enable Server-Side Integration: You cannot get this data with just the JavaScript snippet. Connect your CDN (Cloudflare, etc.) to Clarity to see the server logs.

• Audit “Path Requests”: Identify what they are reading. Are they scraping your high-value proprietary data (pricing, JSON endpoints) or your brand-building content (blog, about page)?

• Calculate your “AI Conversion Rate”: Compare your AI Request Volume (from Clarity) to your AI Referral Traffic (from GA4). If the ratio is massively skewed, you need to rethink your content strategy for agents.

From Search Engine to Personal Agent

Read more

This Memo was sent to 24,218 subscribers. Welcome to +124 new readers!

My findings this week show GSC data is about 75% incomplete, making single-source GSC decisions dangerously unreliable.

Premium subscribers get the scripts I used for bot share and a sample rate to apply to their own data.

Show up everywhere your customers search

Semrush Enterprise empowers you to dominate visibility across both search engines and AI search.

It’s one platform that connects all your data, strategy and teams.

Powerful technical SEO capabilities lay the foundation. Real-time tracking and optimization for both SEO and AI visibility are alongside, purpose built for global scales. Semrush Enterprise is how ambitious brands win every layer of discovery.

Plus, it’s all powered by the industry’s leading search database.

Discover the unfair advantages Semrush Enterprise could unlock for your brand.

Explore Semrush Enterprise

1. GSC used to be ground truth

Search Console data used to be the most accurate representation of what happens in the search results. But privacy sampling, bot-inflated impressions, and AI Overview (AIO) distortion suck the reliability out of the data.

Without understanding how your data is filtered and skewed, you risk drawing the wrong conclusions from GSC data.

SEO data has been on a long path of becoming less reliable, starting with Google killing keyword referrer to excluding critical SERP Features from performance results. But 3 key events over the last 12 months topped it off:

• January 2025: Google deploys “SearchGuard,” requiring Javascript and (sophisticated) CAPTCHA for anyone looking at search results (turns out, Google uses a lot of advanced signals to differentiate humans from scrapers).

• March 2025: Google significantly amps up the number of AI Overviews in the SERPs. We’re seeing a significant spike in impressions and drop in clicks.

• September 2025: Google removes num=100 parameter, which SERP scrapers use to parse the search results. The impression spike normalizes, clicks stay down.

On one hand, Google took measures to clean up GSC data. On the other hand, the data still leaves us with more open questions than answers.

2. Privacy sampling hides 75% of queries

Google filters out a significant amount of impressions (and clicks) for “privacy” reasons. One year ago, Patrick Stox analyzed a large dataset and came to the conclusion that almost 50% are filtered out.

I repeated the analysis (10 sites in B2B out of the USA) across ~4 million clicks and ~450 million impressions.

Methodology:

• Google Search Console (GSC) provides data through two API endpoints that reveal its filtering behavior. The aggregate query (no dimensions) returns total clicks and impressions, including all data. The query-level query (with ‘query’ dimension) returns only queries meeting Google’s privacy threshold.

• By comparing these 2 numbers, you can calculate the filter rate.

• For example, if aggregate data shows 4,205 clicks but query-level data only shows 1,937 visible clicks, Google filtered 2,268 clicks (53.94%).

• I analyzed 10 B2B SaaS sites (~4 million clicks, ~450 million impressions), comparing 30-day, 90-day, and 12-month periods against the same analysis from 12 months prior.

My conclusion:

1/ Google filters out ~75% of impressions

• The filter rate on impressions is incredibly high, with ¾ filtered for privacy.

• 12 months ago, the rate was only 2 percentage points higher.

• The range I observed went from 59.3% all the way up to 93.6%.

2/ Google filters out ~38% of clicks, but ~5% less than 12 months ago

• Click filtering is not something we talk about a lot, but it seems Google doesn’t report up to one-third of all clicks that happened.

• 12 months ago, Google filtered out over 40% of clicks.

• The range of filtering spans from 6.7% to 88.5%!

The good news is that the filter rate has gone slightly down over the last 12 months, probably as a result of fewer “bot impressions.”

The bad news: The core problem persists. Even with these improvements, 38% click-filtering and 75% impression-filtering remain catastrophically high. A 5% improvement doesn’t make single-source GSC decisions reliable when 3/4 of your impression data is missing.

Premium subscribers get the script I used to do this analysis so they understand how much of their own data is filtered out.

3. 2025 impressions are highly inflated

The last 12 months show a rollercoaster of GSC data:

• In March 2025, Google intensified the rollout of AIOs and showed 58% more for the sites I analyzed.

• In July, impressions grew by 25.3% and by another 54.6% in August. SERP scrapers somehow found a way around SearchGuard (the protection “bot” that Google uses to prevent SERP scrapers) and caused “bot impressions” to capture AIOs.

• In September, Google removed the num=100 parameter, which caused impressions to drop by 30.6%.

Fast forward to today:

• Clicks decreased by 56.6% since March 2025

• Impressions normalized (down -9.2%)

• AIOs reduced by 31.3%

I cannot come to a causative number of reduced clicks from AIOs, but the correlation is strong: 0.608. We know AIOs reduce clicks (makes logical sense), but we don’t know exactly how much. To figure that out, I’d have to measure CTR for queries before and after an AIO shows up.

But how do you know click decline is due to an AIO and not just poor content quality or content decay?

Look for temporal correlation:

• Track when your clicks dropped against Google’s AIO rollout timeline (March 2025 spike). Poor content quality shows gradual decline; AIO impact is sharp and query-specific.

• Cross-reference with position data. If rankings hold steady while clicks drop, that signals AIO cannibalization. Check if the affected queries are informational (AIO-prone) vs. transactional (AIO-resistant). Your 0.608 correlation coefficient between AIO presence and click reduction supports this diagnostic approach.

4. Bot impressions are rising

I have reason to believe that SERP scrapers are coming back. We can measure the amount of impressions likely caused by bots by filtering out GSC data by queries that contain more than 10 words and 2 impressions. The chance that such a long query (prompt) is used by a human twice is close to 0.

The logic of bot impressions:

• Hypothesis: Humans rarely search for the exact same 5+ word query twice in a short window.

• Filter: Identify queries with 10+ words that have >1 impression but 0 clicks.

• Caveat: This method may capture some legitimate zero-click queries, but provides a directional estimate of bot activity.

I compared those queries over the last 30, 90, and 180 days:

• Queries with +10 words and +1 impression grew by 25% over the last 180 days

• The range of bot impressions spans from 0.2% to 6.5% (last 30 days)

Here’s what you can anticipate as a “normal” percentage of bot impressions for a typical SaaS site:

• Based on the 10-site B2B dataset, bot impressions range from 0.2% to 6.5% over 30 days, with queries containing 10+ words and 2+ impressions but 0 clicks.

• For SaaS specifically, expect 1-3% baseline for bot impressions. Sites with extensive documentation, technical guides, or programmatic SEO pages trend higher (4-6%).

• The 25% growth over 180 days suggests scrapers are adapting post-SearchGuard. Monitor your percentile position within this range more than the absolute number.

Bot impressions do not affect your actual rankings - just your reporting by inflating impression counts. The practical impact? Misallocated resources if you optimize for inflated impression queries that humans never search for.

5. The measurement layer is broken

Single-source decisions based on GSC data alone become dangerous:

• 3/4 of impressions are filtered

• Bot impressions generate up to 6.5% of data

• AIOs reduce clicks by over 50%

• User behavior is structurally changing

Your opportunity is in the methodology: Teams that build robust measurement frameworks (sampling rate scripts, bot-share calculations, multi-source triangulation) have a competitive advantage.

Premium: Scripts for data sampling and long query tracking

The scripts below let you quantify how much of your GSC data is noise and pinpoint where GSC impressions stop being reliable. If you’re still forecasting, prioritizing, or making strategic calls off raw GSC impressions, this is how you correct the signal before it derails your decision-making.

Read more

Subscribe now

Sent to 24,068 subscribers. Welcome to +156 new readers! Btw, Graphite published insightful research about why SEO is not dead.

Agentic commerce transforms organic search from a source of cheap traffic into the mandatory gatekeeper of AI verification. Marketing arbitrage dies; product truth wins.

This week, we’re covering:

• Why agentic commerce filters out marketing-first brands and rewards granular product data

• How ChatGPT, Copilot, and Google’s protocols reshape merchant economics and customer relationships

• Which feeds to optimize, which protocols to prioritize, and the implementation sequence that matters

Premium subscribers also get my Agentic Commerce Adoption Playbook - a 70+ item implementation checklist covering Shopify settings, feed tags, and protocol-specific constraints.

Want to improve ecommerce performance in ChatGPT

ChatGPT uses Google Shopping results to form its product recommendations. We ran an experiment to confirm this theory and found the encoded fan-out queries that take place in the background.

We also found that the top product in ChatGPT and Google Shopping overlapped 75% of the time.

If you’re an ecommerce brand, your Google Shopping feed needs to be a focus. Any product pages or listings that fill these must be up to date.

Check out the full findings, then start tracking and improving your product performance across AI search platforms with Semrush Enterprise AIO.

CTA: Read the experiment findings

Agentic commerce acts as a “great filter,” so to speak, for marketing arbitrage, transforming organic search from a source of cheap traffic into the mandatory gatekeeper of AI verification.

The signal is already visible in the noise. During the 2025 holiday season, AI agents powered 20% of retail sales. Even allowing for loose definitions, the era of agentic commerce has arrived.

All major LLMs now offer direct checkout and new commerce protocols:

1. ChatGPT has Instant Checkout with Shopify and Etsy, and ACP (Agentic Commerce Protocol)

2. Microsoft Copilot uses ACP and offers Copilot Checkout with PayPal, Shopify, and Stripe.

3. Google has embedded checkout in AI Mode and Gemini via its Universal Commerce Protocol (UCP).

The infrastructure question is settled, but the strategic question remains: How do you compete when users don’t need to click through to websites to buy?

1. Agentic Commerce has a hole in the middle

The phrasing “agentic commerce” sets the wrong expectation. Autonomous purchasing, where you give an agent a credit card and monthly allowance to buy on your behalf, is not becoming a reality in the near future.

• High-priced purchases like plane tickets or cars are too risky to delegate. You have idiosyncratic preferences (airline seat rules, car features) that no agent can reliably model.

• Low-priced purchases like toilet paper or laundry detergent already have automation via subscription services (Instacart recurring orders, Subscribe & Save). An agent adds no incremental value.

• The middle ground is smaller than the hype suggests. If high-priced resists delegation and low-priced is already “automated,” where does autonomous purchasing actually generate value?

“Conversational commerce” is a better frame. Instead of 100% automating the act of buying, LLMs compress the funnel by offering far superior research to classic search engines and showing products in the user interface.

• Models read expert reviews, product specs, ingredient lists, and actual user feedback rather than ranking by keyword bids and conversion history.

• The value lies in collapsing 14 clicks (Amazon’s disclosed average before purchase) into one or two.

2. Protocols make e-commerce “headless”

The new commerce protocols allow AI agents to directly plug into the backend of your business, instead of crawling your site to show them in a list of search results. Protocols make commerce “headless” and decouple the front from the back-end:

• Websites become less important as destinations and more important as databases.

• The game shifts from optimizing landing page design for human eyes to optimizing data feeds for machine ingestion.

• If your shipping speed, inventory status, or return policy isn’t accessible via API, you are invisible to the agent.

The shift from crawling to protocols collapses the legacy 14-click funnel (search, browse, click, checkout) into just 2 interactions: (1) the model parses intent by matching expert reviews against real-time inventory, and (2) the user executes a single click to buy using stored credentials.

While both protocols, ACP and UCP, enable the same user experience, they offer vastly different terms for the merchant.

OpenAI’s ACP (Agentic Commerce Protocol)

• The Vision: The “Walled Garden.” OpenAI aims to handle the entire transaction within the chat interface, treating merchants effectively as suppliers.

• The Trade-off: Efficiency vs. LTV. You gain access to 700M weekly users, but you lose the direct customer relationship. Because OpenAI currently restricts passing customer emails for marketing, you lose the ability to remarket - effectively killing the 15-20% of Lifetime Value (LTV) that typically comes from post-purchase email flows.

Google’s UCP (Universal Commerce Protocol)

• The Vision: The “Distributed Layer.” Google extends its Shopping Graph into a transactional layer that sits on top of Search, Lens, and Gemini.

• The Trade-off: Ownership vs. Competition. Unlike ACP, Google allows merchants to retain the full customer lifecycle, including email rights and loyalty data. The cost is significantly higher competition intensity: Instead of fighting for 10 blue links, you are fighting for 1 of 3 “slots” in an AI Overview, making the margin for error in your product data effectively zero.

3. Conversational commerce disrupts the whole ecosystem

The shift from search to conversation creates a distinct set of winners, losers, and strategic dilemmas.

Buyers get a dramatically better user experience.

• Discovery: High-consideration purchases (e.g., specific running shoes) shift from clicking through 6 potentially irrelevant product listing ads to receiving top-tier recommendations based on expert reviews.

• Cognitive Load: The model handles the research, collapsing the average 14-click journey into 1-2 interactions.

Merchants face a tradeoff between distribution and control.

• On ChatGPT: You gain access to early adopters, but lose the direct customer relationship and email marketing rights. You have no leverage over commission rates or recommendation logic.

• On Google/Copilot: You retain merchant-of-record status, but as the funnel compresses, on-site ad inventory loses value. While conversion rates may rise, total ad revenue falls.

Affiliates die when LLMs disintermediate the click.

• The Trap: If ChatGPT synthesizes reviews without sending traffic, affiliates stop writing. This creates an “ouroboros” where models train on their own AI-generated output.

• The Pivot: Publishers must paywall premium content or charge merchants directly for reviews.

Amazon dominates on price and speed but faces a business model conflict.

• The Conflict: Retail margins are thin (~1%); profitability comes from the $60B advertising business.

• The Risk: Amazon’s ad machine relies on a 14-click funnel. If conversational commerce compresses this to one click, sponsored product inventory evaporates.

• The Choice: They must either block crawlers to protect ad revenue (current strategy) or participate and cannibalize it. Walmart joining ChatGPT forces their hand.

Google is best positioned to weather the shift.

• Parity: They are already monetizing AI Overviews at parity with legacy search.

• Economics: Higher relevance leads to exploding conversion rates. Advertisers will pay more per click to offset the lower click volume, balancing the ecosystem.

4. SEO shifts from optimizing clicks to optimizing ingestion

We are moving from a world of infinite shelf space (10 blue links, endless pagination) to a world of constrained shelf space (3 recommendation slots in an AI response).

In this environment, SEO shifts from optimizing for clicks to optimizing for ingestion. The goal isn’t to get a human to visit your landing page; it’s to get your product data into the agent’s context window with enough authority that it recommends you.

The New “Technical SEO:” Feed quality in the legacy model meant site speed, mobile responsiveness, and Core Web Vitals. In the protocol era, technical SEO is feed integrity. Agents don’t “browse” your site; they query your API. Your website becomes less of a visual destination and more of a structured database. The winners will be merchants who treat their product feed as their primary storefront.

The New “On-Page SEO:” Legacy SEO often rewarded articles that simply summarized what everyone else was already saying to rank for broad keywords. LLMs, however, are trained on that consensus. To be cited now, you must provide Information Gain, the delta between what the model already knows and the unique value you provide on top of the consensus.

• You cannot “market” your way out of inferior specs. If you claim to be the “best running shoe for flat feet,” the model doesn’t look for adjectives; it validates your arch support measurements against podiatry standards in its training set.

• Your content must shift from general engagement to structured “Product Truth.” LLMs prioritize detailed comparison tables, proprietary test results (e.g., “we dropped this phone 50 times”), and ingredient breakdowns. If your data isn’t structured for easy ingestion/verification, the model will bypass you for a source that is.

The New “Off-Page SEO:” Backlinks still matter, but their function changes. Instead of passing “link juice” for ranking, they now serve as verification sources for reputation synthesis together with reviews and web mentions.

• LLMs scrape third-party sites (e.g., Reddit, specialized forums, expert review sites) to form a consensus. A high volume of verified, specific reviews on trusted third-party platforms is the strongest signal you can send.

• In a world where an AI suggests 3 options, brand familiarity becomes a tie-breaker. Brand advertising and organic brand building returns as a critical lever to ensure users recognize the recommendation the AI provides.

5. The end of “Marketing Brands”

The last decade allowed white-label brands to arbitrage their way to growth via ads, but agentic commerce acts as the quality filter for this model. While humans are swayed by slick branding, LLMs are dispassionate readers of data that will not recommend a “premium” product when the specs prove it is identical to a generic alternative.

The shift to protocols creates a paradox: Models understand long-tail intent perfectly but fulfill it with fat head inventory.

• Safety Bias: Models prefer consensus to avoid hallucinations. A niche brand looks like noise; a Category King looks like truth.

• The RAG Reality: RAG tools typically only scan the top 10-20 search results. Since search engines already favor authority, RAG often just reinforces the incumbents.

The only force that overrides this bias is granular data. Your merchant feed acts as the Claim, but RAG acts as the Trust Layer to verify it.

The market bifurcates:

• The Incumbents win general intent via “trust” (consensus)

• The Specialists win specific intent via “granularity” (specs), but only if they rank in the top search results

If you expose data points the giants ignore (e.g., exact sourcing, chemical analysis), the model’s reasoning engine must select you to fulfill the constraint, but only if you rank on page 1 to be fetched.

Organic search is no longer about the click; it is the prerequisite for agentic verification.

Premium: The Agentic Commerce Playbook

Today’s Memo explains why agentic commerce matters. But knowing the landscape isn’t the same as navigating it.

Premium subscribers get the Agentic Commerce Adoption Playbook: a 70+ item implementation checklist that walks you through every phase of getting your products into ChatGPT, Copilot, and Google’s AI Mode.

It includes:

1. Exact paths for Shopify settings, feed tags, and Merchant Center configs

2. Protocol-specific constraints (like ChatGPT’s email marketing restriction and what it means for your LTV math)

3. A decision matrix for prioritizing protocols based on your product type and business model

4. The pitfalls I’m seeing brands make right now

The window for early-mover advantage is open. If you want the step-by-step guide to capture it, subscribe to premium.

Read more

Subscribe now

Last week, I announced that all posts older than 4 weeks will now be gated. You can upgrade to Premium for the full archive, research, frameworks, and templates.

Growth Memo Premium is trusted by top growth leaders and operators navigating AI search in real time. Sent to 23,862 subscribers. Welcome to +163 new readers.

SEO died as a traffic channel the moment pipeline stopped following pageviews. Traffic is either down for many sites, or its growth nowhere near reflects growth rates of 2019-2022, but demos and pipeline are up for brands that shifted from chasing clicks to building authority.

What you’ll get in today’s memo:

• Why traffic and pipeline decoupled

• What brand strength actually means in AI search

• How to reframe SEO with executives

This week, premium subscribers also get the Brand-SEO Scorecard - a measurement framework to diagnose whether your SEO strategy is positioned for traffic or brand influence - and a 30-day action plan to transition to a brand-SEO focus.

Be first everywhere customers search

For businesses, visibility means more than rankings. Semrush Enterprise empowers brands to own every layer of search.

By unifying SEO and AI search visibility into one platform, you get comprehensive performance tracking, live content scoring, and real-time optimization guidance.

Your teams can swap manual busywork for impactful strategy with advanced automations. And it’s all backed by the market’s leading search database.

It’s how leaders become dominant across both search engines and AI.

Explore Semrush Enterprise

1. We’ve hit peak search volume for traditional queries

Short-head keyword demand is in permanent decline and likely contributing to slowed traffic growth or decline.

An analysis of roughly 10,000 short-head keywords shows that collective search volume grew only 1.2% over the last 12 months and is forecasted to decline by 0.74% over the next 12 months.

Two forces are driving it:

• Fragmentation into long-tail: demand did not disappear, it atomized into thousands of specific queries.

• Bypass behavior: more users start in AI interfaces (AIOs, AI Mode, ChatGPT) instead of classic search.

This shift is irreversible for 4 structural reasons:

1/ AI Overviews are here to stay. Google’s revenue model depends on keeping users inside the SERP. Zero-click search protects Google’s ad business. The company is not reverting to the 10 blue links.

2/ LLM outputs are preferred starting points. Many users have conditioned themselves to expect direct answers. The behavior change is complete.

3/ Zero-click is now the default expectation. Clicking through now feels like friction, not value. If the answer or solution isn’t easily acquired, the search experience failed.

4/ Content supply exploded. There is significantly more content competing for the same queries than 3 years ago. AI-generated articles, Reddit threads, YouTube videos, and newsletters all compete for visibility. Even if visibility or “rankings” hold, CTR collapses under the weight of infinite options.

Optimizing for traffic growth in this environment is like optimizing for fax machine usage in 2010. The channel is structurally shifting - the products that people use to find answers have fundamentally changed.

2. Traffic and pipeline decoupled because AI ate the click

The correlation between organic traffic and pipeline has broken. But it takes a bit more work to convince stakeholders and executives. We’re seeing this across the industry.

In December, Maeva Cifuentes reported traffic growth of 32% for one of her clients, while signups grew 75% over the same 6-month period. Her post was in response to one from Gaetano DiNardi, who found no correlation between traffic and pipeline across multiple B2B SaaS companies he advises. Maeva’s client data shows you can grow pipeline 2.3x faster than traffic. Gaetano’s data shows you can grow pipeline while traffic stays flat or even declines.

The classic SEO model assumed a linear relationship: More rankings meant more clicks, more clicks meant more traffic, more traffic meant more leads.

Alternatively, AI answers queries without sending clicks. The Growth Memo AI Mode Study found that when the search task was informational and non-transactional, the number of external clicks to sources outside the AI Mode output was nearly zero across all user tasks. Users get the information they need - directly in their interface of choice - without ever visiting your site.

But buying intent didn’t disappear with the clicks.

SEO creates influence. It can still shape which brands buyers trust. It just doesn’t deliver the click anymore.

Education happens inside the AI interface. Brand selection happens after. Your traffic vanished, but the demand for your product/services didn’t.

This explains why Maeva noted she has clients whose traffic is declining but demos are growing by double digits month-over-month.

The SEO work didn’t stop working. The measurement broke. Teams optimized for clicks are being judged on a metric that no longer predicts business outcomes.

3. Strong brands still win in AI search, but “brand strength” has a new definition

In AI search, performance depends less on “more pages” and more on whether AI systems can confidently understand, trust, and cite you for a specific audience and context.

Brand strength in AI search has four components:

1. Topical Authority: Complete ownership of the conceptual map (see topic-first SEO), not just keyword coverage.

2. ICP Alignment: Answers tailored to specific buyer questions, prioritizing relevance over volume. Read Personas are critical for AI search to learn more.

3. Third-Party Validation: Citations from category-defining sources matter more than high-DA links (see the data in How AI weighs your links).

4. Positioning Clarity: LLMs must recognize what a brand is known for. Vague positioning gets skipped; sharp positioning gets cited (covered in State of AI Search).

SEO teams that are structured for traffic optimization are now misaligned with business outcomes.

The conversation you need to have is “traffic and pipeline decoupled, here’s the data proving it, and here’s what we’re measuring instead.”

Move from keyword-first workflows to ICP-first workflows. Start with ICP research (what questions do your buyers ask and where do they ask them), positioning (what are you known for), and omnichannel distribution (SEO + Reddit + YouTube + earned media). SEO is no longer a standalone channel. It’s one input in a brand-building system.

Move from traffic reporting to influence reporting. Stop leading stakeholder conversations with sessions, impressions, and rankings. Report on brand lift (are more people searching for you by name?), pipeline influence (what percentage of demos started with organic touchpoints?), and LLM visibility rates (how often do AI systems mention your brand vs cite your content?).

4. The uncomfortable question: If SEO doesn’t drive traffic anymore, what does it do?

Here’s what SEO actually does and always did: It shapes mental availability and brand recognition, builds topic/category authority, frames the problem (and the solution), and reduces buyer uncertainty.

Traffic was a proxy for those things. The click was the observable action, but the trust was the outcome that mattered.

LLM-based search has removed the click but kept the trust-building. Users still learn from your content. It just happens inside an LLM interface instead of on your domain. Your content can still influence which brands buyers trust. Yes, it’s harder to measure because it’s invisible to analytics. But the outcome - buyers choosing your brand when they’re ready to buy - is the same.

SEO influences brand preference within the category. When buyers are in-market and researching solutions, SEO determines whether your brand is in the consideration set and whether AI systems recommend you.

Traffic was never the point. It was just the easiest thing to measure.

5. What to do in the next 30 days

This week, premium subscribers get the playbook for moving from traffic-first to pipeline-first, including:

• The Brand-SEO Scorecard: An 8-metric diagnostic with industry benchmarks.

• A 30-Day Action Plan: How to execute the pivot.

Read more