We matched 4,722 AI Overviews to the Google results under them
When Proofsource asks Google a buyer question, one search returns both the AI Overview and the regular results below it. So for each answer we can see which pages the Overview cited and which pages ranked on page one for the same question.
The sample is every AI Overview we collected from October 4 to 30, 2026, for an English buyer question that doesn't name the brand, searched from the US on desktop: 4,722 answers to 2,655 questions for 158 brands in 111 accounts. We dropped repeated identical answers. The median question was 17 words long, the way buyers actually type into an AI box.
The No. 1 result was cited 31.4% of the time
Higher results were cited more often, but none came close to a sure thing. Each page-one position, and how often that result was among the Overview's sources:
- Position 1: 1,484 of 4,722 (31.4%).
- Position 3: 1,270 of 4,719 (26.9%).
- Position 5: 1,042 of 4,716 (22.1%).
- Position 9: 567 of 3,262 (17.4%).
73% of the pages the Overviews cited weren't on page one
The Overviews cited 37,208 pages. 10,125 of them (27.2%) were among the results on page one for that question. Count any page from a site that ranked on page one and it rises to 37.9%. 1,280 of the 4,722 Overviews (27.1%) cited nothing from page one at all.
Google says why in its own documentation: AI Overviews may use a "query fan-out" technique, running several related searches across subtopics to build the answer. A 17-word question splits into shorter searches, and the pages cited come from those.
Why this is lower than the 76% you've probably read
Ahrefs looked at 1.9 million citations from 1 million AI Overviews in July 2025 and found 76.1% of cited pages ranked in the top 10. That study counted only the top three citations in each Overview, and its searches were ordinary keywords. Counting only each Overview's first three sources in our data gives 38.8% (5,410 of 13,944).
We don't think either number is wrong. They answer different questions. If you rank for a short keyword, the Overview on that keyword often cites you. A buyer who types a whole question into Google gets an Overview built from searches you may not rank for.
Brands' own sites were on page one in 2% of these searches
For the 4,512 answers where we knew the brand's domain, the brand's own site was on page one 90 times (2.0%), across 36 brands. When it was, the Overview named the brand in 33 of 90 answers (36.7%, 95% interval 27.4% to 47.0%) and cited its site in 49. When it wasn't, the brand was named in 337 of 4,422 (7.6%) and its site cited in 132 (3.0%).
Ranking clearly goes with being named. But it was rare, and even then the brand was left out of most of those Overviews.
What we'd do
- Type your buyer's whole question into Google, not your keyword, and note who ranks and who the Overview cites. They're often different pages.
- Write pages that answer the narrower questions a long one splits into: pricing, who it's for, how it compares, how to set it up.
- Get listed on the pages that keep getting cited for your questions. A roundup that names you counts even when your site doesn't rank.
- Don't report a No. 1 ranking as AI visibility. Check the Overview itself.
How Proofsource handles it
Proofsource records every page each engine cited in each answer, next to the brands it named. You can see which pages Google's AI Overviews cite for your buyers' questions, and whether your own site is one of them.
Where this could be wrong
Each answer is one search, from the US on desktop, at one moment. Overviews change between searches, so a page we counted as left out may be cited on the next try.
We matched pages by address after removing "www", query strings and trailing slashes. Redirects and other copies of the same page will be missed, which is why we also give the site-level figure. We only saw page one, about nine results, so we can't say where the other cited pages ranked.
The own-site comparison rests on 90 answers and isn't an experiment. Brands that rank for a buyer question are likely better known, and that alone could explain much of the gap.
Besides the 178 repeats, we left out 1,119 Google answers with no Overview block, no page-one results or no citations, and four Shopify review test stores. The questions come from categories our users chose to track.
