The finding that reconciles the two halves of this is that ranking and citation have already come apart. Ahrefs looked at 863,000 keywords and 4 million AI Overview citations in March: only 38% of cited pages rank in Google's top 10, down from 76% last July. So a demotion in classic rankings and a 43.8% citation share can coexist without contradiction, because the two systems now select on different things. Rankings reward the page. Citations reward whichever passage is easiest to lift whole and attribute. That is why self-serving listicles keep getting cited while losing rank: the format hands a model a clean, attributable line even when nothing independent sits behind it. The durable version is the same format built on a claim a reader can check, which is also the only version that survives models getting better at spotting the self-ranking pattern.
The gap you can't close is the interesting result here. If self-promotion isn't the variable, one candidate worth checking is editing depth: a separate analysis of roughly 150,000 pages found purely machine-written content ranking about 23% lower than human-written pages, while substantively edited AI drafts landed within about four points of human writing. Authorship and editing depth aren't the same axis. A brand-authored listicle that got heavily reworked by a human would behave like human content regardless of who published it; a lightly-touched publisher listicle would underperform despite not being self-serving at all.
That would also fit the timing. Google's August 19 spam update was the third enforcement action in five months explicitly targeting mass-produced, low-value pages, not brand pages as such. Worth a look in part two: does the vendor/publisher split correlate with editing depth, independent of self-promotion, and does that explain more of the January-to-August decline than authorship does?
Listicles if done right can be useful. But so often it's just slop.
agreed!
The finding that reconciles the two halves of this is that ranking and citation have already come apart. Ahrefs looked at 863,000 keywords and 4 million AI Overview citations in March: only 38% of cited pages rank in Google's top 10, down from 76% last July. So a demotion in classic rankings and a 43.8% citation share can coexist without contradiction, because the two systems now select on different things. Rankings reward the page. Citations reward whichever passage is easiest to lift whole and attribute. That is why self-serving listicles keep getting cited while losing rank: the format hands a model a clean, attributable line even when nothing independent sits behind it. The durable version is the same format built on a claim a reader can check, which is also the only version that survives models getting better at spotting the self-ranking pattern.
The gap you can't close is the interesting result here. If self-promotion isn't the variable, one candidate worth checking is editing depth: a separate analysis of roughly 150,000 pages found purely machine-written content ranking about 23% lower than human-written pages, while substantively edited AI drafts landed within about four points of human writing. Authorship and editing depth aren't the same axis. A brand-authored listicle that got heavily reworked by a human would behave like human content regardless of who published it; a lightly-touched publisher listicle would underperform despite not being self-serving at all.
That would also fit the timing. Google's August 19 spam update was the third enforcement action in five months explicitly targeting mass-produced, low-value pages, not brand pages as such. Worth a look in part two: does the vendor/publisher split correlate with editing depth, independent of self-promotion, and does that explain more of the January-to-August decline than authorship does?