Google AI Overviews: 46% of Cited Sites Had No Top-10 Page

In our 15-day study of 660 US-English search queries, 46% of sites cited in the day-one AI Overviews had no page in the organic top 10 for the same query. Across 9,675 total search runs, citation sets changed substantially from day to day. Rankings still mattered: the first organic result was cited more often than the tenth.

If your AI visibility report says “not cited” after one search, that is a data point, not a verdict. The same query can produce a different source list tomorrow, and the sites Google cites are not simply a copy of the first page of results.

That is the practical takeaway from our study: measure citation presence repeatedly, separate it from organic rank, and make the page’s answer easy to find. Do not treat a top-10 ranking as a citation guarantee, or a single missed citation as proof that your content is invisible.

What we measured

We ran 660 distinct US-English Google search queries across six commercial categories: B2B software, finance, health, local services, online shopping, and travel. The set included 100 generic queries per category and 60 brand-review queries. On the first day, we recorded each query’s organic top 10, AI Overview, and AI Mode answer. We then re-ran the 601 queries that showed an AI Overview daily for 15 days, producing 9,675 total search runs, including the initial 660.

For the daily citation analysis, we counted a site once per answer, even if the answer cited multiple pages from that site. For the day-one ranking comparison, we downloaded 3,972 page-one pages and compared cited pages with uncited pages at the same organic position. Separately, we traced 1,702 cited passages to where they appeared on their source pages.

The design matters. Our repeated measurements cover AI Overviews; the organic-ranking comparison and AI Mode comparison are snapshots from day one. The searches were commercial, in US English, and limited to six categories. The results describe this sample, not every query, country, or search experience.

Rankings matter, but they do not explain every citation

In the day-one ranking comparison, 46% of the sites cited by AI Overviews had no page in Google’s top 10 for that same search. In 10% of searches, none of the cited sites ranked on page one. Only 32% of page-one results were cited in the corresponding AI Overview. The first organic result was cited in 49% of cases; the tenth was cited in 22%.

Those findings do not mean SEO is irrelevant. The citation rate was higher for the first result than for the tenth. They mean organic position alone is an incomplete way to predict which sources an AI Overview will cite.

A separate Ahrefs analysis of 863,000 keyword SERPs and four million cited URLs also found that citations were not confined to the top 10. In its blue-link-only comparison, 37.1% of cited URLs ranked in the top 10, 26.2% ranked between 11 and 100, and 36.7% did not rank in the top 100 for the original query. The studies use different units and methods: Ahrefs counted URLs, while our day-one comparison counted cited sites and checked whether a site had a page in the top 10. The figures should not be compared as if they were the same metric. Together, though, they show why “ranked on page one” and “was cited” should be tracked separately. [3]

One plausible part of the explanation is query expansion. Google says AI Overviews and AI Mode may use “query fan-out”: exploring related searches across subtopics and sources when assembling an answer. That creates more routes for a relevant page to be discovered than the original query’s results page alone. Our study did not observe Google’s internal retrieval path for each citation, so we cannot say that fan-out caused any specific citation. [1]

The sources changed more than a one-time check suggests

Only 11% of AI Overviews in our repeated sample cited exactly the same set of sites as the previous day. The first-listed source changed in 42% of day-to-day comparisons. For a typical query, an Overview cited about seven sites on a given day, but 18 different sites across the 15-day window.

The variation persisted even for searches that consistently showed an Overview. We tracked 418 searches with an AI Overview on all 15 days; not one kept exactly the same cited-site set every day. Only 14% kept the same first-listed source throughout the full period.

This is why “Are we cited?” is a weak measurement question when it is asked once. A more useful question is “How often are we cited across repeated observations of the same search set?” Track the share of runs where your brand or site appears, whether it is named or linked, which URL is cited, and how those measures move over time. Keep citation frequency distinct from organic rankings and from referral traffic: they describe different outcomes.

Google also says AI Overviews and AI Mode can use different models and techniques, so their answers and links may vary. That matched what we observed: on the same searches on day one, AI Mode and AI Overviews shared an average of 33% of cited sites, and 10% of searches had no site in common. AI Mode gave an answer without citing a site in 10.5% of cases in our sample. These are day-one comparisons, not a 15-day AI Mode trend. [1]

For a practical measurement cadence and a fuller discussion of repeat runs, see our guide to how often to run AI visibility analysis (https://www.dodoxhq.com/blog/ai-visibility-analysis-frequency) and our framework for measuring AEO success (https://www.dodoxhq.com/blog/how-to-measure-aeo-success).

The answer’s position on the page may help, but word count is not a shortcut

In 1,702 passages we traced to source pages, 67% came from the top 30% of the page; 56% came from the first fifth. For AI Mode, 62% of traced passages came from the top 30% of the page.

That is a useful editorial signal, not a published Google rule. Put the direct answer and the most important qualification near the top, then use the rest of the page to explain the evidence, method, exceptions, and implications. This makes the page easier for a person to scan and gives a retrieval system a clear, self-contained passage to work with. Our observations do not prove that moving a paragraph upward will cause a citation.

Length alone did not distinguish cited from uncited pages in our comparison. Pages under 500 words were more common among ignored pages than cited pages (25% versus 17%). Pages of 3,000 words or more appeared at roughly equal rates in both groups. This is not evidence for a word-count target. It is evidence against assuming that a long page is automatically useful, or that a short page cannot be complete.

Google’s published guidance supports that distinction: there are no special AI Overview or AI Mode optimizations or required special markup. A page must be indexed and eligible to appear with a snippet in Search. Google recommends the same fundamentals it recommends elsewhere: useful, reliable, people-first content, accessible text, crawlable pages, internal links, and structured data that matches what users can see. [1] [2]

Source choice depends on the question

The sites that appeared most often varied by source type and search intent. Across 8,295 AI Overviews in our dataset, YouTube appeared in 63%, Reddit in 24%, Google-owned pages in 18%, and NerdWallet in 10%. Those shares are not mutually exclusive: one answer could cite several sources. Specialists also appeared within their subject areas, including NerdWallet in finance and Mayo Clinic in health.

Shopping queries showed an especially sharp intent split. For product-selection searches such as “air fryer best rated” or “toaster oven vs air fryer,” 84% of AI Overviews cited Google’s own product pages. For how-to searches, that figure was 7%. Retailers accounted for 18% of page-one shopping results but 7% of cited sites; a retailer already on page one was cited 22% of the time, compared with 33% for other page-one results.

These patterns are a reason to study the actual result for the query you care about, not to copy the source mix from another category. A buyer comparing products, a person asking for instructions, and someone evaluating a healthcare claim are asking different questions. The most useful sources can differ accordingly.

What to do with the findings

Start with the searcher’s question and give a precise, direct answer before the supporting detail. Show how you know: include original observations, first-hand experience, named sources, dates, definitions, and limitations where they matter. If a claim depends on a particular version, market, or use case, say so next to the claim.

Then make the page easy to discover and interpret. Keep important content in crawlable text. Link to it from relevant pages using descriptive anchor text. Make sure the page can be indexed and that its structured data reflects the visible content. Use Search Console to monitor search performance and analytics to understand visits and conversions; Google reports AI-feature traffic within the overall “Web” search type in the Performance report, rather than as a separate AI Overview report. [1]

Finally, build a repeatable prompt set and measure it over time. Include the real questions buyers ask, not just a handful of brand-name prompts. Record the engine and date for every run. Report visibility as a trend across runs, and separate Google AI Overviews from AI Mode and other answer engines. When the cited sources change, inspect what changed in the query, source mix, and answer, not just your rank.

The goal is not to reverse-engineer one fluctuating answer. It is to publish a source that is demonstrably useful for the question, technically available to the search system, and worth citing, and then measure whether it appears consistently enough to matter.

References

• [1] AI features and your website | Google Search Central — https://developers.google.com/search/docs/appearance/ai-features
• [2] Creating helpful, reliable, people-first content | Google Search Central — https://developers.google.com/search/docs/fundamentals/creating-helpful-content
• [3] Update: 38% of AI Overview citations pull from the top 10 | Ahrefs — https://ahrefs.com/blog/ai-overview-citations-top-10/