Why AI citation studies contradict each other

Rank-citation overlap is the share of AI Overview citations that also appear in the classic organic results for the same query, and it is currently the most misquoted number in generative engine optimization.

You can find a credible study saying the overlap is 38% (Ahrefs), another saying it is around 54% and rising (BrightEdge), and a third saying it is 90% (seoClarity). All three are competent analyses of large datasets. None of them is lying. They answer different questions, and the industry has been quoting them as if they answered the same one.

#Three numbers, three studies

Ahrefs analyzed 863,000 keyword SERPs and roughly 4 million AI Overview URLs, and found that 37.9% of cited pages also ranked in the top 10 for the same query (Ahrefs). The rest split almost evenly: 31.2% came from positions 11 to 100, and 31.0% came from pages ranking beyond position 100 (Ahrefs). In the July 2025 version of the same study, the top-10 figure was about 76% (Ahrefs). Read alone, that looks like a collapse.

BrightEdge tracked AI Overview citations over 16 months with its own parser and reported the opposite direction of travel: overlap with organic rankings rising from 32.3% to 54.5% (BrightEdge). Same surface, same broad question, trend pointing the other way.

seoClarity looked at 362,000 desktop queries in the United States and 5.1 million citations, and reported that 90% of AI Overviews cite at least one URL from the organic top 10, rising to 94% for the top 20 (seoClarity). In the same dataset, 56% of individual citations came from top-20 rankings and 44% came from outside them (seoClarity).

Look closely at that last pair. seoClarity's own data contains both a very high number and a middling one, drawn from the same crawl. That is the tell.

#The denominator is doing all the work

There are two different metrics hiding under the phrase "rank overlap."

The first is per-answer coverage: of all AI Overviews, how many cite at least one page that ranks in the top 10? That is a question about whether ranked pages show up at all. Because a typical AI Overview cites several sources, and because you only need one of them to be a top-10 result for the answer to count, this metric runs high by construction. seoClarity's 90% is this measurement (seoClarity).

The second is per-citation share: of all cited URLs, what fraction rank in the top 10? Every citation counts separately, so obscure sources drag the number down instead of being absorbed by a single qualifying result. Ahrefs' 37.9% is this measurement (Ahrefs), and so is seoClarity's 56% for the top 20 (seoClarity).

These are not competing estimates of one quantity. They are two quantities. "Nearly every AI answer includes a page that ranks" and "well under half of cited pages rank" are both true at once, and each supports a completely different strategy slide. The first says rankings are still the ticket to entry. The second says ranking is not close to sufficient. Anyone who quotes one without the other is selling something.

#Better parsing can manufacture a trend

The second trap is subtler, and Ahrefs flagged it themselves. Between the two versions of that study, they improved their parsing so they could detect more of the citations appearing in AI Overviews, and the URL sample doubled (Ahrefs).

Consider what that does to a per-citation ratio. The citations a parser misses first are the awkward ones: collapsed link groups, sources buried in expandable sections, formats it did not previously recognize. Those newly visible citations are disproportionately unlikely to be top-10 organic results, because well-ranked pages were the easiest to catch in the first place. Add them to the denominator and the top-10 share falls, even if Google's selection behavior never changed at all.

This is not a criticism of Ahrefs, who were transparent about the methodology change. It is a warning about the year-over-year comparison that everyone else built on top of it. The drop from 76% to 38% (Ahrefs) is being cited across the industry as evidence that Google decoupled AI Overviews from search rankings. Part of that drop is real. Part of it is a measurement instrument getting better at its job. Nobody outside Ahrefs can currently say what the split is, and honest reporting should say so.

We have made this argument before in a different context: an AI visibility figure is a distribution, not a number, and it moves with sampling and instrumentation as much as with reality.

#What the studies actually agree on

Strip out the denominator confusion and a consistent picture survives.

A large minority of AI Overview citations come from pages that do not rank on page one. Ahrefs puts roughly 62% of citations outside the top 10 (Ahrefs), and seoClarity puts 44% outside the top 20 (seoClarity). Those two figures use different cutoffs and different parsers, and they still tell the same story: pages that rank poorly, or not at all, get cited constantly.

Ranking well also still helps a lot. In seoClarity's data, position one appeared in AI Overviews 43% of the time, declining steadily to 7% at position 20 (seoClarity). Rank is a strong signal. It is just not a gate.

The mechanism explains the tension. Ahrefs' Louise Linehan attributes the shift to AI Overviews relying less on the direct result set and more on the sources that surface in fan-out sub-query SERPs (Ahrefs). A page that ranks 40th for the query you care about can rank 2nd for a sub-question the model generated on its way to an answer. That is query fan-out, and it is why "what do I rank for" is the wrong unit of analysis. The right unit is the set of questions the engine decomposes yours into.

#How to read the next study you see

Three questions, before the number goes in a deck.

What is the denominator: answers or citations? If the study reports a number above four in five, it is almost certainly per-answer coverage, and it does not mean your unranked competitor is being ignored.

Did the parser change between measurement periods? If it did, the trend line is partly an artifact, and the level is more trustworthy than the direction.

Which engine, which query mix, which country? Every figure above describes Google AI Overviews on a particular slice of queries. It says nothing directly about ChatGPT, Perplexity, or Gemini, and nothing about your category.

That last point is the practical one. Industry averages are useful for orienting, and useless for deciding. The overlap that matters is the overlap on your own query set: for the questions your buyers actually ask, how often does an answer cite a page of yours, and how often does that page also rank? Measure that on your own prompts, on a schedule, and you will stop needing to arbitrate between other people's headline percentages.

They were never measuring your queries anyway.