How does Google AI Overviews choose sources?

Someone in your market asks Google a question. Instead of ten blue links, they get a short paragraph with two or three citations underneath it. Your page ranks on page one for that exact phrase. It still is not one of the citations. That is not a bug. It is how AI Overviews actually choose sources, and the mechanics differ from classic SEO enough that ranking well is no longer sufficient on its own.

This is the Google-specific chapter. For the general mechanics that apply across every engine, see how AI decides who to cite. For the same diagnosis worked through for ChatGPT specifically, see why doesn’t ChatGPT recommend my business.

How does Google AI Overviews choose which sources to cite?

It reuses Google’s existing Search ranking and quality systems rather than running a separate algorithm, so being well-ranked, fresh and genuinely helpful still counts. What is new is that Gemini first breaks your question into several related sub-queries and searches all of them at once, then can cite whichever page answers any one of those sub-queries best, not only the page that ranks for the exact words you typed.

Does Google use a separate algorithm for AI Overviews?

No. Google’s own developer documentation states that its generative AI features on Search are "rooted in our core Search ranking and quality systems" and that "there are no additional requirements to appear in AI Overviews or AI Mode" (Google Search Central, 2026). Google runs more than 15 core ranking systems across Search, covering things like PageRank, Helpful Content and Freshness, and AI Overviews draws on that same stack rather than a bespoke one.

What is query fan-out, and why does it matter for citations?

Query fan-out is the step before source selection. At Google I/O 2025, Google’s Head of Search Elizabeth Reid described how, under the hood, Search recognises when a question needs more than a single lookup and calls on Gemini to break it into subtopics, then issues several related searches simultaneously on the user’s behalf (Search Engine Land, 2025). A simple question might generate a handful of sub-queries; a complex or comparative one can generate well over a dozen.

Does ranking number one on Google guarantee an AI Overview citation?

No, and the guarantee has been getting weaker. Across a large-scale study, the share of AI Overview citations that also ranked in the traditional top 10 fell from 76% to 38% within about seven months, meaning a page can now be the cited source without cracking the classic first page at all.

How much has the overlap between rankings and citations changed?

Substantially. Ahrefs’ original analysis of 1.9 million AI Overview citations found 76% also ranked in the top 10, with a median position of 2 (Ahrefs, 2025). A follow-up study across 863,000 keywords and 4 million AI Overview URLs found that figure had dropped to just 38%, with the remaining citations split almost evenly between pages ranking 11 to 100 (31.2%) and pages ranking beyond position 100 (31.0%) (Ahrefs, 2026). Ahrefs notes the shift lines up with Google rolling AI Overviews onto Gemini 3 globally in January 2026.

How is an AI Overview citation different from a normal top-10 result?

The unit Google is scoring shrinks from the page to the passage, and the target multiplies from one query to a cluster of sub-queries. A page ranking 40th for your target keyword can still win the citation if it is the clearest, most current answer to one of the sub-queries fan-out generated, which is precisely the class of page the 62% figure above accounts for.

Does content freshness matter more for AI Overviews than classic search?

Yes, measurably. Ahrefs analysed roughly 17 million citations across AI Overviews and major AI assistants and found AI-cited content is 25.7% fresher than what shows up in ordinary organic results, with roughly half of cited pages published or meaningfully updated within the prior 13 weeks (Ahrefs, 2025). A visible last-updated date and an actual recent edit both help more here than they do for classic rankings.

What if I do not rank in the top 10 at all?

You are not automatically out of the running. Because citations now spread across the full ranking spectrum rather than clustering near position one, the more useful question is not "where do I rank for this keyword" but "which of the sub-queries around this topic do I answer better than anyone else."

What should you actually do to win Google AI Overview citations?

Treat the literal search term as one entry point among several, not the whole target, and put the same effort into extractability and freshness that the research rewards.

• Answer the surrounding questions, not just the target keyword. List the sub-questions a buyer would ask around your topic and give each one a direct, self-contained answer, the same discipline covered in how AI decides who to cite.
• Keep your highest-value pages genuinely current, not just re-dated. Update the facts, not only the timestamp, since freshness is a scored signal here.
• Open every section with the answer in the first sentence or two. Fan-out means Google is matching passages against many sub-queries at once, so a passage that only makes sense in context of the whole page will not survive extraction.
• Check that Google-Extended, the crawler behind Google’s AI features, is not blocked in your robots.txt. The check is the same idea covered for ChatGPT’s crawler in why doesn’t ChatGPT recommend my business, just a different bot name.
• Track citations for the query cluster around your target term, not only the exact phrase, since that cluster is what fan-out is actually scoring you against.

How do you know if it is working?

Google does not give you a chat box to prompt-test the way ChatGPT or Claude does, so proving an AI Overview citation takes a different kind of evidence than asking a model directly. Outercite’s core pipeline actively tests your target prompts across ChatGPT, Claude, Gemini, Perplexity, Grok and DeepSeek, with every candidate citation checked by a judge model and then confirmed or disputed by an independent verifier model before it counts.

For Google specifically, we also watch the other side of the evidence: real visits. Outercite’s CiteTrace detects traffic attributed to Google AI Overviews, via the Google-Extended crawler and AI Overview referral patterns, and binds those visits back to your existing citation records. That turns "we think this page might be cited" into "this citation is sending real people to your site."

Sources

• AI Overviews and AI Mode share Search’s core ranking and quality systems with no additional requirements to appear: Google Search Central, 2026. Ranking systems guide.
• Query fan-out mechanics and Elizabeth Reid’s Google I/O 2025 remarks on breaking questions into subtopics: Search Engine Land, 2025. Guide.
• The original 76% top-10 overlap across 1.9 million AI Overview citations, median position 2: Ahrefs, 2025. Study.
• The drop to 38% top-10 overlap across 863,000 keywords and 4 million AI Overview URLs, with the 31.2%/31.0% position split: Ahrefs, 2026. Study.
• AI-cited content 25.7% fresher than organic results, roughly half of citations under 13 weeks old, across 17 million citations: Ahrefs, 2025. Study.