1 billion data points.
Ten findings that rewrite the AEO playbook.
Ahrefs analyzed over a billion data points across 14 studies of AI search. We pulled the ten findings that actually change what you should do — and what each one means for your visibility.
By Lynceus Research Team
The source
The closest thing we have to a map of AI search.
Over six months, Ahrefs analyzed more than 1 billion data points across 14 studies of AI search, summarized by Tim Soulo in June 2026. It's the most data-dense look at how AI assistants choose what to cite that anyone has published. Below are the ten findings that actually change what you should do — with our read on each. The numbers are Ahrefs'; the interpretation is ours.
Finding 1 — format
"Best X" listicles dominate citations.
43.8% of the page types ChatGPT cites are "best X" blog listicles — the single most prominent format. Most are third-party roundups, not brand-owned pages. The implication: getting onto other people's lists is as important as publishing your own. We unpack the tactics in the listicle playbook.
Finding 2 — influence
Two-thirds of citations are sources you can't touch.
67% of ChatGPT's top 1,000 citations come from sources marketers can't influence directly: Wikipedia (29.7%), homepages (23.8%), and app stores (6.6%). Only 32.3% is influenceable content — educational pages, reviews, news, blog posts. The lesson isn't "give up," it's "concentrate." Pour effort into the third you can move, and earn your way into the two-thirds you can't (a Wikipedia-worthy reputation, strong app-store presence) rather than spreading thin.
Finding 3 — the separate layer
Nearly a third of cited pages don't rank in Google at all.
28.3% of ChatGPT's most-cited pages have zero Google organic visibility. They get cited repeatedly despite not ranking. This is the most strategically important number in the set: AI search is a separate discovery layer with its own winners, not Google with a filter on top. If your AI plan is "rank well and hope it carries over," you're optimizing the wrong index.
Finding 4 — retrieval vs citation
Being retrieved isn't being cited.
ChatGPT cites only about 50% of the URLs it retrieves. It fetches dozens of pages per query and uses half as silent background context with no attribution. So "the model read my page" and "the model credited my page" are different outcomes. The work isn't just getting fetched — it's being the source clean and quotable enough to earn the citation. See how ChatGPT decides what to cite.
Finding 5 — schema
Schema markup didn't move citations.
Adding schema had no meaningful effect on AI citations in the study: AI Overviews dipped 4.6%, while AI Mode (+2.4%) and ChatGPT (+2.2%) showed changes indistinguishable from zero. Read this carefully — it means schema is not a citation lever, not that schema is worthless. Structured data still aids retrieval and clean extraction; it just won't manufacture citations by itself. A lot of AEO advice being sold right now over-promises on this exact point.
Finding 6 — the strongest signal
YouTube mentions beat every conventional SEO metric.
YouTube mentions had the highest correlation with AI brand visibility of anything Ahrefs measured: 0.737 — above backlinks, Domain Rating, and page count, and it held for both Google-owned and OpenAI products. Correlation isn't causation, but it's a strong enough signal to act on. The thing that gets you cited isn't a link; it's being talked about where the models trained. More in the YouTube playbook.
Finding 7 — the disappearing click
AI Overviews cut clicks to #1 by 58%.
Clicks to the top organic result drop 58% when an AI Overview is present — up from 34.5% just ten months earlier. The trend is accelerating. This is why click-through is becoming the wrong success metric and why measuring visibility without clicks is now a discipline of its own.
Finding 8 — intent
AI Overviews are almost entirely an informational play.
99.9% of AI Overviews appear on informational queries. Transactional, navigational, and local searches are nearly AIO-free, and shopping triggers an AIO just 3.2% of the time. Translation: AI search is reshaping the top of the funnel — research, comparison, "how do I" — far more than the moment of purchase. Defend your informational content first.
Finding 9 — same answer, different sources
AI Mode and AI Overviews agree — but cite different pages.
For a given query, Google's AI Mode and AI Overviews reach the same conclusion 86% of the time, yet their citations overlap only 13.7%. Being cited on one Google surface does not mean you're cited on the other. You have to track them as distinct surfaces, not assume coverage transfers.
Finding 10 — volatility
The words churn; the meaning barely moves.
AI Overviews change every 2.15 days on average, with 70% of content differing between consecutive observations — but semantic similarity holds at 0.95. The sources, phrasings, and entities reshuffle constantly while the underlying answer stays stable. So don't panic over day-to-day citation flicker; track the trend over weeks, and measure whether the meaning of the answer favors you, not whether your exact sentence survived today's reshuffle.
The throughline
AI search rewards being known, not being optimized.
Put the ten together and one picture emerges: you can't schema-hack your way into the answer, your Google rankings don't entitle you to citations, and the strongest signal is how much the internet talks about you. AI search rewards brands that are known — quoted, listed, discussed — over brands that are merely technically optimized. You earn the mention, and the citation follows.
Which is also why you can't manage this surface with tools built for Google's index. Different layer, different winners, different scoreboard. The first step is seeing your own scoreboard — what AI visibility is, and what measuring it looks like.
Frequently asked questions
Where do these AI search statistics come from?
From Ahrefs. In a June 2026 summary, Tim Soulo reported that over the previous six months Ahrefs analyzed more than 1 billion data points across 14 separate studies of AI search. The findings below are Ahrefs' data; the interpretation and the 'what to do' framing are ours. Treat the numbers as directional signals from one rigorous dataset, not universal laws.
Does schema markup help with AI citations?
According to Ahrefs' study, adding schema markup had no meaningful impact on AI citations — AI Overviews citations actually dipped 4.6%, while AI Mode (+2.4%) and ChatGPT (+2.2%) changed in ways indistinguishable from zero. The nuance: schema may still help retrieval and clean extraction even if it doesn't move citation counts on its own. It's not a citation lever; don't treat it as one.
What content format gets cited most by ChatGPT?
'Best X' listicles. Ahrefs found they make up 43.8% of all page types ChatGPT cites — the single most prominent format. Most of those are third-party roundups, not brands' own pages, which is why earning placement on other people's lists matters as much as publishing your own.
What correlates most strongly with AI brand visibility?
YouTube mentions, at a 0.737 correlation — higher than backlinks, Domain Rating, page count, or any conventional SEO metric Ahrefs studied, and it held for both Google-owned and OpenAI products. Correlation isn't causation, but the signal is strong enough to take seriously: the models reward brands the internet talks about, especially on video.