Blog
By Haobo Zhang AI SEO Jul 3, 2026 10 min read

Do AI Overview citations come from the top 10? Every study, reconciled

Published figures run from 12% to over 99%. Nearly all of that spread is denominators and counting methods, not disagreement — every major study, reconciled.

No — you don’t need a top-10 ranking to be cited in AI Overviews. Published figures run from 12% to over 99%, and almost none of that spread is disagreement about reality — it’s different denominators, counting units, keyword sets, and dates. Count individual citations: roughly 17–38% today. Count whole AI Overviews: about 90% still touch page one.

If you’ve researched this question, you’ve met the whole zoo: 99%, 76%, 52%, 38%, 17%. They circulate as if they answer the same question. They don’t — and I wanted one page that lines every study up and explains exactly why each number is what it is. (For the broader numbers landscape, see our AI search statistics roundup.)

What does every study actually find?

Here are the studies in one table — read the “what it counts” column before the headline number, because that column explains the spread.

StudyDateSampleWhat it countsHeadline number
AhrefsJul 21, 20251M AIOs / 1.9M citationsTop-3 most visible citations per AIO only, URL-level76.1% of cited pages in top 10; median rank of primary citation: 2
Ahrefs updateMar 2, 2026863K SERPs / 4M cited URLsAll parsed citations, URL-level37.9% in top-10 blocks; 31.2% at 11–100; 31.0% beyond 100
Ahrefs cross-assistantAug 11, 202515K long-tail promptsChatGPT/Gemini/Copilot links vs Google top 10 — assistants, not AIOs12% on average (Perplexity separately: 28.6%)
Originality.AINov 18, 2025~29K YMYL queries”Organic citations” only — citations that also appear in the top-100 organic results52% of those from top 10 (≈25% of all citations — see below)
BrightEdgeSep 18, 20259 industries, May 2024–Sep 2025Per-citation overlap with tracked rankingsOverlap anywhere grew 32.3% → 54.5%, but only 16.7% of citations from top 10
SurferNov 30, 202510K keywords (76% had AIOs)All citations vs top 10 for the query or any fan-out query67.82% NOT in top 10; top-3 visible only: 45.86% not in top 10
seoClarityOct 12, 2025362K US desktop keywords / 5.1M citationsBoth units in one study90% of AIOs cite ≥1 top-10 page (94% for top 20) — but only 56% of citations come from the top 20
seoClarity, earlier trackingpage updated Oct 23, 2025not stated for this figurePer-AIO: share of AIOs with ≥1 top-10 source”>99% of the instances of AI Overviews were sourced from the top 10 web results”
seoClarity AI ModeOct 2, 20251K transactional US queriesPer-citation — AI Mode, not AIOs19% of citations from top 20; rank 1 cited only 25% of the time

Three footnotes the table can’t hold. Ahrefs’ 2026 “top-10 blocks” include SERP features; restricted to classic blue links it’s 37.1%. Surfer used Gemini-generated fan-out queries as a proxy and didn’t pull rankings beyond position 10, so some “not in top 10” pages rank 11+ — Surfer flags both. And BrightEdge doesn’t publish whether it matches at URL or domain level. All of these datasets skew US, where the keyword set is disclosed at all.

Why do the studies disagree so much?

They mostly don’t disagree — they count different things. Four axes explain nearly the entire 12%-to-99% spread.

  1. Per-citation vs per-AIO vs per-domain. “56% of citations come from the top 20” and “94% of AIOs contain at least one top-20 citation” are both true in the same seoClarity study. One AI Overview citing one ranked page satisfies the per-AIO measure while its other seven citations come from nowhere. BrightEdge’s report holds a rising 54.5% any-rank overlap and a 16.7% top-10 share on the same page.

  2. All citations vs the top-3 visible ones. Google front-loads ranked sources. Ahrefs’ 76.1% counted only the three most visible citations; Surfer measured the gap directly — 45.86% of top-3-visible citations weren’t in the top 10, vs 67.82% of all citations. Same data, 22-point swing.

  3. Rankings for which query? AI systems retrieve for related sub-queries, not just your literal query — that’s query fan-out. Measured against the original prompt alone, Ahrefs found only 12% of ChatGPT/Gemini/Copilot citations in Google’s top 10 (that study covers assistants, not AI Overviews — it gets misquoted as an AIO stat constantly).

  4. Keyword set and date. Originality.AI sampled YMYL queries, where Google leans hardest on ranked pages; seoClarity’s AI Mode study used transactional queries on a different surface entirely. And the date matters, because the share is genuinely moving.

Is the top-10 share genuinely falling?

Yes — with an asterisk Ahrefs itself put there. Ahrefs’ own re-run of its own study fell from 76.1% in July 2025 to 37.9% in March 2026. Part of that is methodology: the update says Ahrefs “improved our parsing methodology” to capture far more citations per AIO than the old top-3-visible approach, so the two datasets aren’t directly comparable. How much of the drop is real versus parsing is not separable from the published data. Medium

But Ahrefs also asserts a real shift, in its own words: “AI Overviews are relying less on the direct search results and more on the sources showing up in fan out query SERPs” — and notes AI Overviews have been powered by Gemini 3 since January 2026. That direction matches Originality.AI’s finding that only 48% of all AIO citations overlap the top-100 organic results at all — which is also why its “52% from the top 10” needs unpacking: 52% of the overlapping half is roughly 0.52 × 0.48 ≈ 25% of all citations. Almost nobody who quotes the 52% does that arithmetic.

How does “>99% of AI Overviews” become “99% of citations”?

The most famous number in this debate is a real statistic wearing the wrong denominator. seoClarity’s early tracking reported, verbatim: “>99% of the instances of AI Overviews were sourced from the top 10 web results.” That’s a per-AIO measure — the share of AI Overviews containing at least one top-10 source. The page gives no sample size or date for it; seoClarity’s tracking began in August 2024, so it likely reflects 2024-era AI Overviews.

Retold, “instances sourced from” becomes “citations come from,” and suddenly 99% of citations come from the top 10 — a claim every per-citation dataset (16.7%, ~25%, 37.9%, 56% top-20) contradicts.

We’re not writing this from a pedestal. Our own AI Overviews guide repeated the “over 99% of citations” version, citing a third-party roundup that, on inspection, contains no 99% figure anywhere — its nearest stat is a 92.36% per-AIO, domain-level number attributed to an unnamed study. We corrected that guide in the same release that published this post. The lesson generalizes: before trusting any AIO stat, find the primary source and check the denominator.

Do you still need to rank, then?

Ranking is no longer the gate, but it’s still the strongest single lever you control. Google’s official position is that there are “no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary” — a page needs only to be indexed and snippet-eligible (Google Search Central). The same doc says fan-out exists to “display a wider and more diverse set of helpful links… than with a classic web search.” Falling overlap is working as designed.

Yet within that design, rank pays. Position-1 pages get cited 43% of the time vs about 7% at position 20 (seoClarity), and 57.91% at rank 1 in Originality.AI’s YMYL set. The sharpest practical finding: pages ranking for fan-out queries are 161% more likely to be cited than pages ranking only for the main query, and main-plus-fan-out rankers took 51% of citations (Surfer data via Search Engine Land, Spearman 0.77). Cover the query’s neighborhood, not just the head term.

As a sanity check, Cyrus Shepard’s May 2026 meta-analysis of 54 studies scores URL accessibility 9.5, search rank 9.4, and fan-out rank 9.3 as the top AI-citation factors. The studies above, reconciled, say the same thing.

Frequently asked questions

Do you need to rank in the top 10 to appear in AI Overviews?
No. In Ahrefs' March 2026 data only 37.9% of cited URLs ranked in the top-10 result blocks, with the rest split roughly evenly between positions 11–100 and beyond the top 100. Ranking still helps a lot, though — seoClarity found position-1 pages cited 43% of the time.
What percentage of AI Overview citations come from page 1?
It depends on the counting method and date: 76.1% (Ahrefs, July 2025, top-3 visible citations only), 37.9% (Ahrefs, March 2026, all citations), 16.7% (BrightEdge, September 2025), and 56% from the top 20 (seoClarity, October 2025). There is no single true number.
Is it true that 99% of AI Overview citations come from top-10 pages?
No. That figure traces to seoClarity's finding that over 99% of AI Overview instances contained at least one top-10 source — a per-AIO measure, not a per-citation one. Per-citation studies in 2025–26 put the top-10 share at roughly 17–38%.
Does Google say you have to rank to be cited in AI Overviews?
No. Google's documentation says there are no additional requirements beyond being indexed and eligible for a snippet, and that query fan-out deliberately surfaces a wider and more diverse set of links than a classic search.
Why is the top-10 share of AI Overview citations falling?
Ahrefs attributes it to AI Overviews relying more on fan-out query results, and notes AIOs have run on Gemini 3 since January 2026. Part of the 76%-to-38% drop is also methodological: Ahrefs improved its citation parsing between the two studies, so the datasets aren't directly comparable.

Sources