Most AI Overview Citations Do Not Rank in the Top 10

Google's AI Overviews cite pages that do not rank. In an Ahrefs study published March 2, 2026, covering 863,000 search results and 4 million citation URLs, only 37.9 percent of cited pages appeared in Google's top 10 for the same query. About 31 percent came from positions 11 to 100, and another 31 percent from beyond position 100. Ranking is still one path to a citation. It is no longer the gate.

Why does ranking first no longer guarantee a citation?

Because the AI Overview is not answering your keyword. It is answering a set of smaller questions it wrote for itself.

Ahrefs calls the mechanism query fan-out and describes it plainly: "This is when the initial query is split into multiple related sub-queries. The pages that appear most often within those sub-query SERPs then get cited in the AI Overview."

The page that gets cited, then, is not necessarily the page that won the original search. It is the page that kept turning up across the smaller questions the system asked on the way to an answer. A page can sit at position 40 for the phrase you optimized and still be cited, because it was the clearest source on a sub-question nobody on your team is tracking.

Ahrefs author Louise Linehan frames the shift as AI Overviews "relying less on the direct search results and more on the sources showing up in fan out query SERPs."

The distribution follows the mechanism. 37.9 percent of citations came from top 10 pages, 31.2 percent from positions 11 to 100, and 31.0 percent from pages ranking beyond 100 or not ranking at all. Close to two in three citations came from outside the first page of results.

How much of that number is a real change?

This is where most coverage of the study stops being useful, so it is worth being precise.

The same Ahrefs study a year earlier, in July 2025, found 76 percent of citations came from top 10 pages. A fall from 76 to 38 is dramatic, and the tempting reading is that Google changed something enormous. Ahrefs does not make that claim cleanly. The March 2026 article notes that alongside any change in behavior, "we've also improved our parsing methodology in Ahrefs, so that we can see even more of the citations that appear in AI Overviews."

Part of the swing is a better instrument rather than a moving world. Ahrefs measured more of the citations that were always there, and the ones it had been missing skew toward lower-ranking pages.

That caveat does not rescue the old assumption, though. Suppose the true year-over-year change is half the headline. The surviving fact is still the one that matters: across 4 million citation URLs, a clear majority of AI Overview citations come from pages outside the top 10. Treat the direction as solid and the magnitude as approximate. Anyone quoting "76 to 38" without the methodology note is repeating a number they did not read to the end.

That habit is worth naming, because this subject attracts it. A great deal of writing about AI search consists of one figure, stripped of its sample size and its caveats, passed between sites until nobody can say where it came from. The defense is unglamorous: find the study, check what was counted, and check what the authors themselves said about their own confidence.

What actually gets cited?

The domain-level picture is less comfortable than the ranking picture.

In a separate Ahrefs analysis published July 21, 2026, covering more than 3 million United States queries, the most-cited sources in Google AI Overviews were YouTube at 21.1 percent mention share, Reddit at 18.5 percent, Facebook at 10.7 percent, Google's own properties at 7.1 percent, Instagram at 5.8 percent, Quora at 5.0 percent, and Wikipedia at 4.8 percent.

Those seven account for 73 percent of the mention share among the 50 most-cited sources. Six of the seven are platforms nobody in your business owns.

Read the metric carefully before you act on it. Mention share here is a domain's citations as a percentage of the citations earned by the top 50 sources, so it describes concentration at the head of the list, not a share of every citation Google produces. Even read conservatively, the shape holds: the cited surface is crowded with rented properties, and a company site competes for what is left.

What should a small business do with this?

Three conclusions follow, and none of them is to abandon search work.

Stop using position as the only scoreboard. A page at position 12 that gets cited in an AI Overview is doing more for you than a page at position 4 that does not. If your reporting only tracks rank, you cannot see the difference, and you will keep spending on the wrong page. Rank and citation are now two separate measurements and both need a column.

Write to the sub-question, not just the phrase. Fan-out rewards pages that answer a specific question completely in one place. That favors clear definitions, real numbers, stated methods, and named tradeoffs. It penalizes pages built to hold a keyword. This is the same structural discipline behind how AI search changes the way buyers find software partners: the machine is looking for a usable answer, not a match.

Accept that a large share of the surface is rented, and plan accordingly. With YouTube and Reddit alone holding almost 40 percent of top-of-list mention share, some of your visibility sits on platforms whose rules you do not set. Those shares move, and they move faster than a quarterly plan. The citation tracker Promptwatch, monitoring daily citation shares from July 7 to August 17, 2026, recorded Reddit holding a steady 3.8 percent of ChatGPT Search citations from July 18 through August 7, then dropping to a 0.5 percent average across August 14 to 17. That is an 86 percent fall in about a week, on one engine, with no announcement. Be present on those platforms deliberately, and do not book that presence as an asset.

The same measurement makes a second point worth more than the first. Over that identical window, Reddit's share in Google AI Overviews declined 11.3 percent and in Google AI Mode 30.5 percent, while the ChatGPT figure fell 86 percent. This is one tracker measuring one platform family over six weeks, so treat it as an illustration rather than a law, but the direction is clear enough to act on: the engines are not moving together, and a single blended "AI visibility" number will average away the thing you needed to see.

Which is the argument for instrumentation. If you cannot answer "which of our pages were cited last quarter, and by which engine," you are guessing about the channel, and every recommendation you receive about it is unfalsifiable. The work is ordinary: track citations per engine as their own metric, separate from rank, and review it on the same cadence as any other pipeline number. It is the same case we make for executive dashboards and for AI revenue systems, which are only as honest as the measurement underneath them.

None of this makes ranking worthless. A top 10 position still produces the single largest slice of citations, 37.9 percent of them, and it remains the most reliable thing you can influence directly. The change is narrower and more useful than the headlines suggest: ranking has gone from the entry requirement to one input among several, and the businesses that notice first will be the ones measuring the right thing while everyone else is still counting positions.