How Google AI Overviews Choose Which Sources to Cite
Amit Malvi
Founder, SearchAxe. Former Senior SEO Manager, Invideo AI
Published Updated 11 min read
Quick answer
How Google AI Overviews choose which sources to cite comes down to three stages. First, eligibility: only indexed pages that can show a snippet qualify. Second, retrieval: Google's model runs several related searches through its core ranking systems, a step Google calls query fan-out. Third, selection: it links the pages that support a specific part of the answer. That is why Ahrefs found only 37.9% of cited pages ranked in the top 10 for the original query.
On this page
- How Google AI Overviews choose which sources to cite
- What is query fan-out and why does it change which pages get cited?
- How much do AI Overview citations overlap with organic rankings?
- Why does Google skip some pages that rank on page one?
- Does Google choose passages or whole pages?
- What do the studies agree on, and where do they disagree?
- What does this mean for a SaaS content plan?
Key takeaways
- Google documents the entry conditions: the page is indexed, eligible for a snippet, and on a site included in Search Console's Search generative AI control. Retrieval then runs through its core ranking systems.
- Query fan-out means the overview searches related subtopics, so pages ranking for a sub-question get cited for queries they never ranked for.
- Ahrefs' March 2026 data shows 37.9% of cited URLs ranked in the top 10 for the same query, down from about 76% in its July 2025 study.
- Ranking pages get skipped when they lack a passage that supports one part of the answer, restrict snippets, repeat commodity information or simply lose a churn cycle.
- Citations are unstable: Ahrefs found 45.5% of cited sources changed between consecutive observations of the same overview.
If you lead marketing at an AI or SaaS startup and you rank on page one for a question but the AI Overview above you cites someone else, this explains why. Google AI Overviews cite sources in three stages: eligibility, retrieval through several related searches, and selection of pages that support specific parts of the generated answer. Google documents the first stage clearly and the other two only in outline, so the published studies fill the gaps. For the practical steps that follow from this, see our guide on how to rank in Google AI Overviews.
How Google AI Overviews choose which sources to cite
Google AI Overviews choose sources by searching first and writing second. Google's guide to generative AI features describes retrieval-augmented generation, or grounding, as relying on its core Search ranking systems to retrieve relevant, up-to-date pages, and says the features are rooted in those ranking and quality systems. The overview then links to pages that support what it says. Google does not publish how it weighs one eligible page against another.
| Stage | What Google documents | What studies have measured |
|---|---|---|
| Eligibility | The page must be indexed and eligible to show with a snippet, with no extra requirements | Not measured, but follows from Google's rules: noindex, nosnippet or an exclude setting in the Search generative AI control removes a page here |
| Retrieval | Core ranking systems retrieve pages, and fan-out issues related searches across subtopics | 37.9% of cited URLs ranked in the top 10 for the original query (Ahrefs, March 2026) |
| Selection | Systems can identify the relevant part of a page that covers several topics | 88% of AI summaries cited three or more sources, with a median length of 67 words (Pew, July 2025) |
| Regeneration | Not documented | 45.5% of cited sources changed between consecutive observations (Ahrefs, November 2025) |
What is query fan-out and why does it change which pages get cited?
Query fan-out turns one question into several searches, so the pool of candidate pages is wider than the results for the query you typed. Google's AI features documentation says both AI Overviews and AI Mode may issue multiple related searches across subtopics and data sources to develop a response. A page that ranks well for one of those sub-searches can be cited even if it is nowhere near page one for the original query.
Google's March 2025 AI Mode announcement gives the clearest description: searches issued concurrently across subtopics and multiple data sources, then combined into one response. Its example was a question comparing sleep tracking on a smart ring, a smartwatch and a tracking mat, where the system makes a plan, searches, and adjusts the plan based on what it finds.
Ahrefs' August 2025 overlap study offers a likely explanation for why consistency across sub-searches matters. It suggests results from fan-out queries are merged with methods like reciprocal rank fusion, so a page ranking around sixth for three related queries can beat a page ranking first for only one. That is Ahrefs' hypothesis about AI systems in general, not something Google has confirmed for AI Overviews, but it matches what the citation data shows.
Find out which AI engines recommend you
A free consultation on where you appear today across AI engines and search.
Schedule a Free ConsultationHow much do AI Overview citations overlap with organic rankings?
Less than they used to. In Ahrefs' March 2026 study of 863,000 keyword SERPs and 4 million AI Overview URLs, 37.9% of cited URLs also appeared in the first 10 result blocks for the same query. The rest split almost evenly between positions 11 to 100 (31.2%) and beyond the top 100 (31.0%). In its July 2025 study the top 10 share was about 76%.
Ahrefs suggests the drop may be linked to AI Overviews moving to a newer model in January 2026, with fan-out queries playing a bigger role in source selection. Two other details matter for a SaaS team. When Ahrefs counted standard blue links only, 36.7% of cited pages did not rank in the top 100 at all, which means some citations come from SERP features rather than organic listings. And among cited pages that did not rank in the top 100, 18.2% were YouTube URLs.
The source mix also differs from the blue links. Pew Research Center's July 2025 analysis of searches made in March 2025 found Wikipedia, YouTube and Reddit made up 15% of sources in AI summaries and 17% in standard results. Government sites were 6% of AI summary sources against 2% of standard results, while news sites were 5% in both.
Why does Google skip some pages that rank on page one?
A ranking page gets skipped when it does not support any specific part of the answer the model wrote, when its snippet is restricted, or when it simply loses a regeneration cycle. Ranking makes a page a candidate. It does not make it the best support for a sentence about pricing, setup or limitations. These are the causes we check, in the order we check them.
- No passage answers the sub-question. The page ranks for the head term, but the overview's sentences are about setup, cost or alternatives, and the page only covers features. Google's passage ranking system identifies individual sections of a page, so a section that answers plainly is what gets matched.
- The answer is buried. The facts exist but sit below a slogan, inside an image of text or deep in a long page. Google asks you to keep important content in text form.
- Snippet controls are too tight. Google's May 2025 guidance says nosnippet, data-nosnippet, max-snippet and noindex limit how content appears in AI formats, and that more restrictive settings limit it further.
- The content is commodity. Google's guide asks for non-commodity content and warns against recycling what others have already said. If ten pages say the same thing, the overview needs only one of them.
- Nobody else corroborates you. Ahrefs' May 2025 study of 75,000 brands found branded web mentions correlated 0.664 with AI Overview brand visibility, against 0.218 for backlinks.
- The overview regenerated. Ahrefs' November 2025 analysis found overview content changed every 2.15 days on average and 45.5% of cited sources were new from one observation to the next.
- The site opted out. Search Console's Search generative AI control can exclude a property from AI Overviews and AI Mode, and child properties inherit it, so check the setting on the domain property too.
Some overviews cite nobody. In Ahrefs' December 2025 comparison of AI Overviews and AI Mode, 11% of AI Overviews had no cited sources, mostly simple calculations, sensitive topics, help-center redirects and unsupported languages. If your query falls into one of those groups, there is no citation to win.
Does Google choose passages or whole pages?
Google links to pages, but the evidence suggests it matches on passages. Its generative AI guide says its systems can understand the nuance of multiple topics on a page and show the relevant piece to users, and its ranking systems guide lists a passage ranking system that identifies individual sections. Google does not document a passage-level citation mechanism for AI Overviews specifically, so treat the passage view as a well-supported inference rather than a confirmed rule.
The shape of the overviews points the same way. With a median summary of 67 words and 88% citing three or more sources in Pew's data, each cited page is usually supporting one or two sentences. A long guide is typically cited for the support one paragraph gives, which is why the structure advice in how to structure a web page so AI engines can quote it applies directly here.
What do the studies agree on, and where do they disagree?
They agree that rankings help, that fan-out widens the pool, and that citations are unstable. They disagree on which domains dominate, mostly because they measured at different times with different keyword sets. Treat any single study as a snapshot of one sample.
| Study | Sample | Finding | Caveat |
|---|---|---|---|
| Ahrefs, March 2026 | 863,000 keyword SERPs, 4 million cited URLs | 37.9% of cited URLs in the top 10 for the same query | Single snapshot; earlier study used a different parser |
| Ahrefs, November 2025 | 43,000 keywords over a month | Content changes every 2.15 days; 45.5% of citations change | Checks were not daily, so churn may be higher |
| Ahrefs, May 2025 | 75,000 brands | Web mentions correlate 0.664 with AI Overview brand visibility | Correlation, not causation |
| Ahrefs, July 2025 | 17 million citations across platforms | AI Overviews cite content about as old as organic results | Age based on first crawl date |
| Pew Research Center, July 2025 | 68,879 searches by 900 US adults | 88% of summaries cite three or more sources | March 2025 searches, results collected in April |
| Semrush, July 2025 | Prompts for 500+ digital marketing and SEO topics | Quora was the most cited domain in AI Overviews | Ahrefs' 2026 data names YouTube instead |
The Semrush study and Ahrefs' 2026 data illustrate the point. Semrush's July 2025 study of 500+ digital marketing and SEO topics found Quora was the most cited site in AI Overviews, while Ahrefs reported in March 2026 that YouTube was the most cited domain and had grown 34% over the previous six months. Both can be right for their samples. What does not change between studies is that a third-party platform, not a brand's own site, sits at the top.
What does this mean for a SaaS content plan?
It means planning around the sub-questions an overview needs rather than the single keyword you rank for. Each product, integration, pricing and help page should answer one buyer question plainly in its opening lines, and the brand should be mentioned on the third-party pages Google already cites. Pages blocked from snippets or buried in tabs should be fixed before any new content is written. The same answer-first pages also rank in Bing, which matters beyond Google because ChatGPT search and Copilot draw on Bing's index.
AI Overviews are one engine among several. ChatGPT, Perplexity, Claude, Gemini, Google AI Overviews, Google AI Mode and Copilot each retrieve differently, and our cross-engine piece on how AI search engines decide which sources to cite compares them. For how this sub-question mapping is scoped across both Google surfaces, see Google AI Overviews and AI Mode optimization.
Frequently asked questions
Can an AI Overview cite a page that is not indexed?
No. Google's documentation says a page must be indexed and eligible to show in Google Search with a snippet to appear as a supporting link in AI Overviews or AI Mode, with no additional technical requirements. A page that is noindexed, blocked from snippets with nosnippet, or on a site excluded through Search Console's Search generative AI control cannot be cited. Use URL Inspection in Search Console to confirm a page is indexed before asking why it is not cited.
Does Google show which fan-out queries an AI Overview used?
No. Google confirms that AI Overviews and AI Mode may use query fan-out, but it does not show the related searches behind a given answer, and Ahrefs notes they vary between runs. You can approximate them by listing the follow-up questions buyers ask about your category and checking which pages rank for each one. Treat any tool that claims to show exact fan-out queries as an estimate.
Why did my page disappear from an AI Overview it used to appear in?
Most often because the overview regenerated. Ahrefs found AI Overview content changed every 2.15 days on average and that 45.5% of cited sources were new from one observation to the next. Check the query several times over a week. If the page is consistently gone, look for changes to its snippet controls, indexing, or the section that used to answer the question, and see what the newly cited page says instead.
Do AI Overviews prefer big brands as sources?
They favour brands that are widely discussed. In Ahrefs' study of 75,000 brands, those in the top quarter for web mentions averaged 169 AI Overview mentions, against 14 for the next quarter, and 26% had none. That is a correlation, not proof that size causes citations. A small company that is mentioned often in reviews, comparisons and community threads in its niche can still be cited for niche questions.
