AI Halftime Report: H1 2026
AI's impact kept growing in H1 2026, but the ability to measure it kept falling behind.
Every 6 months, I take a sharp look at the latest developments in AI and Search. Based on how quickly things are developing, I would almost need to increase that cadence to monthly. (And maybe I will.) But first, here’s what happened in H1 2026.
H1 2026 moved money, traffic, jobs, and market cap before anyone could prove how much value AI created. Search behavior changed, token budgets exploded, software stocks sold off, and companies blamed layoffs on AI.
In this memo:
Trust is the obvious new “ranking factor”
Software fell 30% on perceived AI disruption, not measured performance
Meta engineers burned 73.7 trillion tokens in 30 days with no ROI anyone could name
Every major AI story in H1 2026 was really a story about attribution.
We’re still figuring out how to improve accuracy in measuring AI visibility.
Companies spent billions on inference, but there is no obvious answer to “What’s the ROI?”
Public markets punished software companies, but were investors reacting to actual or perceived disruption?
We’re hearing “AI caused layoffs,” but when you look deeper, the real causes aren’t connected.
It’s clear where the traffic loss is coming from, but it’s not clear what content marketplace model will replace it.
The common thread: AI’s economic impact is expanding faster than our ability to attribute it.
How does AI search see US retail?
Refresh your ecommerce strategy for the remaining summer shopping season with data from 126M+ US prompts. Explore how AI search platforms influence brand visibility, recommendations, and discovery across the customer journey.
Build your winning strategy based on the trends that are driving AI visibility for this category’s leading brands.
AI Search
In the H1 2025 report, I predicted Google would roll AI Mode out further, and that turned out to be true. AI Mode is now a click away from AI Overviews, which means it’s just two clicks away from regular search results.
AI Mode reached 1 billion MAU with queries around 3x longer than classic search.
Google called it the biggest search box upgrade in 25 years at I/O 2026.
Gemini 3 auto browse is shipping inside Chrome.
Nick Fox says Google’s AI Mode sends out billions of clicks to the open web.
To understand the impact of this shift, I published a series of user behavior studies and data analyses in the first half of 2026.
The through line of Growth Memo research in H1 2026:
Tracking the presence of a brand in AI Search is severely complex. You need to factor in the engine, personalization, reasoning levels, model updates, stochastic variability, etc. Measurement is fragmented. The citation/mention overlap between engines is minuscule: 91% of citations appear in only one of ChatGPT, Perplexity, or AI Overviews. Prompt tracking should be closer to polling and focus groups than SEO rank tracking.
Brand mentions in AI answers are more impactful on business outcomes than citations. Don’t get me wrong, citations shape the answer, and for some businesses, that’s most important. But the majority of vendors and merchants need to pay attention to how often they show up in a panel of prompts, in which context, what sentiment and whether they’re recommended or not (ahead of competitors).
Trust is the highest currency in AI search. Close to 75% of consumers pick the number one result in an AI shortlist. But if they see a trusted brand anywhere on that list, they will pick it.
The average US adult has high trust in AI recommendations. 88% of the time, users accepted AI Mode product recommendations as best there is. In AI Overviews, on the other hand, users click, evaluate, and compare a lot. That’s classic search behavior, but it doesn’t carry over to AI chatbots.
Writing style, token budget, and technical factors matter for agents. If you want to be present in AI Search and agentic recommendations, you need to lead with unique information, write in a direct, easy-to-understand way, remove fluff and keep your site technically fast and easy to access.
AEO/GEO is a brand channel disguised as a performance channel. AI recommendations shape demand (not citations). The unit of optimization is whether AI names, trusts, and recommends your brand.
⚠️ Traditional rank tracking built for Google misses the other AI search surfaces entirely. The AI Citation Audit gives you a repeatable spot-check protocol and the Prompt Tracking Upgrade Guidance teaches you how to track better. (Both are the Growth Memo Premium Resource Library, and it’s built to reduce your AI visibility reporting mistakes before they happen.)
The grAIpes are sour
November 2025 was a pivotal turning point in AI. Claude’s Opus 4.5 was the first model that was perceived as reliable and robust enough for agentic workflows. A month later, Peter Steinberger went viral with Clawdbot, which led to millions of installs, long lines in China, and Nvidia building a Nemoclaw clone.
Then, things got even crazier. Shopify, Uber, Meta, and many other tech giants built token leadership boards that incentivized engineers to light as many tokens on fire as possible.
“Let me give you a thought experiment. Let’s say you have a software engineer or AI researcher, and you pay them $500,000 a year. At the end of the year, I’m going to ask him how much did you spend in tokens. And [if] that person said $5,000, I will go ape something else. If that $500,000 engineer did not consume at least $250,000 worth of tokens, I am going to be deeply alarmed.” — Jensen Huang
A Meta engineer created an internal leaderboard dubbed “Claudeonomics“ that ranked over 85,000 employees by the number of AI tokens they burned. It led to massive waste as employees left AI agents running on idle or useless tasks just to climb the ranks and earn titles like “Token Legend”.
Reportedly, Meta engineers consumed 73.7 trillion tokens in just 30 days. (I wonder how much of OpenAI’s and Anthropic’s growth was ignited just by workslop.)
Good times. But in April, CFOs pumped the brakes after discovering that employees had burned through their annual token budget in 4 months. TokenMAXXXING shifted to valueMAXXXING.
Meta’s leaderboards were shut down in April (along with leaderboards at other companies), but not before the #1 user on the board averaged 281 billion tokens, which would cost over $1.4 million at standard API rates.
But what remains from the all-you-can-eat token craze is a step change in productivity that we cannot yet measure. As I wrote in AI changed my work. And yours, too:
I have used AI to raise my impact by a magnitude.
I designed landing pages end-to-end for a major travel brand that made it into production.
I automated topic prioritization, SEO testing, and SEO reporting for my clients with full-blown apps.
I built an array of useful applications for myself, from automating the SSI (SEO Site Index found in the bimonthly Growth Intelligence Briefs) to Openclaw agents that help me with research and charts.
One outcome of the token tsunami is that a lot more people started using Claude, which in return fired up the growth engines for AI infrastructure companies that Claude recommended, like Supabase.
Source: Supabase Series F
SaaS bloodbath
In February 2026, as a consequence of the Opus 4.6 release and Claude Cowork, the software-as-a-service industry saw a severe market downturn that wiped out hundreds of billions of dollars in market cap over just a few trading sessions. The SaaS vertical saw a steep annual decline of +30% in the markets. Valuations for high-growth software companies collapsed, with revenue multiples shrinking dramatically compared to their pandemic highs.
If you’re working in B2B, you’ve probably felt the impact in the form of lower conversion rates, longer sales cycles and softening brand searches.
The Brand Tax Calculator gives you the dollar figure for the gap between branded paid spend and reported ROAS, and the AI SEO Budget Reallocation Planner turns it into 5 growth-priority scenarios in under 7 minutes. Both live in the Premium Resource Library.
Source: It’s Time to Build, July 17, 2026
It turns out not all software was (or is) forsaken. It’s the bottom quartile of software companies that pulls the market down. Interestingly, the decline is purely associated with how “the market” views a company’s AI robustness. Company valuation drops are not related to performance, but rather to whether people think AI will disrupt their product.
Over the last 30 trading days, both the top quartile and median IGV stock have outperformed the ETF as a whole—it’s the bottom quartile (which include some of the largest companies) that’s pulling down overall performance. —Source: It’s Time to Build
So, if you’re in B2B and either forced or privileged to take a new job, pick a company the market is optimistic about.
AI washing in the labor market
You might have read or heard this recently: “We’re laying off x% of our workforce due to AI.” It’s the narrative du jour. Challenger, Gray & Christmas reported that AI was the leading cited reason for May 2026 job cuts and had been cited in 87,714 cuts year-to-date, equal to 22% of all 2026 announced layoffs through May.
Source: May 2026 Challenger Report
And don’t get me wrong. Layoffs are real:
Tech layoffs up ~66% year over year, toward 150K
Oracle cut 21,000 citing AI
Challenger lists AI as the top cited reason (~87,700 YTD by May)
Block cut 40% and its stock rose (Bloomberg’s “AI washing” framing)
The only problem is that AI is not the cause. I’ve written extensively about the Big Labs using AI replacement as a marketing narrative. But it hasn’t come true yet. A lot of AI layoffs were really to offset capital expenditures, pandemic overhiring, and economic turbulence. We can all imagine a future in which AI is so good that it makes certain roles obsolete, but that time is not now.
A year ago, in the H1 2025 report, I predicted that AI layoffs were a PR stunt. And I was right: Despite clear productivity gains from AI, the recent waves of layoffs show no indication of AI being the underlying cause. In fact, companies using AI as a reason to lay employees off are rehiring them.
The agent market fragmented
ChatGPT’s market share decreased from 78% in July 2025 to 56% in July 2026, while Gemini went from 15% to 30% and Claude from 2% to 10%.
Google is now the most likely player to win the AI consumer market, and ChatGPT has lost its significance in AI Search to Google:
ChatGPT still has ~1.1B users, but OpenAI refocused on an enterprise pivot and shut down Sora, dropped video in ChatGPT and killed Instant Checkout. Enterprise makes up ~40% of revenue, heading to ~50% ahead of a potential IPO. I’m skeptical that we’ll be able to buy OpenAI stock. According to Ed Zitron, OpenAI spends $2.8 billion USD every month to make $1.1B.
Anthropic’s fight with the Pentagon pushed Claude to No. 1 among free apps on Apple’s U.S. App Store. A few months later, users seem less sure if Anthropic can resist attacks from open-source models.
The Big Labs massively subsidize tokens. SemiAnalysis found a $200 USD plan of the frontier models gives you 8-14K USD worth in tokens.
Lately, open-source models like Kimi K3, GLM 5.2, or Deepseek V4 have put pressure on the big US labs because of those subsidies. Open-weight and open-source models also show that the model itself becomes commoditized, while the harness and application layer gain more importance.
Forward-deployed-engineer job postings rose 800% in the first 9 months of 2025 and were still up 729% year over year in April 2026. Frontier Labs now pays more than $500,000 in total compensation for people who can turn a model demo into a working system inside a customer’s business.
Never have marketers faced a faster evolving environment with more platforms to measure… which is why understanding what platforms matter most for your audience is important. But from an execution perspective, model availability has become a business risk, as we can see at the hand of Fable 5. Having options is good.
Walk into your next leadership meeting with the Growth Memo AI Search Strategy Pitch Deck, built to reframe your search strategy pitch around business priorities, strategic positioning, and learning deliverables.
Publishers and the law
I predicted up to 70% of 2024 organic traffic could be gone by 2026, but that was overeager. In the end, the numbers land at ~33% referral loss year over year, with 68% of Google searches now ending without a click. Still, as a result, publishers are moving the fight to the courts and regulators.
USA Today CEO Mike Reed explained: “We’re getting close now to the point where we’ll block Google as well and abandon the traditional search traffic that we get today.”
Here are just a few lawsuit-related highlights:
A court in Munich ruled Google liable for false statements generated by AI Overviews (Jun 2026).
400 newspapers sued OpenAI and Microsoft over unauthorized content use (Jun 2026).
The UK CMA ordered Google to give publishers greater control and transparency over how content is used in AI search (including opt-outs from AI Overviews, AI Mode, and Discover summaries), along with clearer attribution and improved reporting on how content appears in AI-generated results. Google responded by adding an impression-based AI report in Search Console.
In response to the intense publishing landscape changes, Parallel AI and Cloudflare are both trying to build publishing marketplaces for content owners and AI companies.
“Assume there’s no search. You have to have your businesses planned as if search is zero.” — Conde Nast CEO Roger Lynch
Publishers are trying to replace the content-for-traffic bargain with a content-for-training market. This leaves several open questions:
Who sets the price and how?
How can you prove that an article contributed to an AI agent’s answer or purchase?
Do publishers have negotiation power?
Are smaller publishers even needed by AI labs?
Dig in: Full list of Growth Memo research completed in H1 2026
Dig deeper into the patterns covered in this halftime report below:
User behavior studies:
Data analyses:
As we pass the halfway point in the year, a big thanks to my editorial and research partners, Amanda Johnson and Eric van Buskirk. And an extension of gratitude to my data partners who make these analyses possible: Semrush, AirOps, Gauge, Citation Labs, Siteline, SalaryGuide, and AudienceKey.
Premium: Predictions and planning for H2 2026
H2 2026 will separate intelligence from agency: Intelligence gets cheaper, more abundant, and easier to embed through open-weight models and token competition.
But the authority to act, spend, access data, impersonate someone, or deploy the most capable models becomes more tightly controlled by platforms, governments, payment networks, and users themselves.
