How Google AI Overviews choose sources
AI Overviews break your query into sub-questions, then cite whichever pages answer each part cleanly — length and top-10 rank barely matter.
Google's AI Overviews choose sources by splitting a query into several smaller sub-questions, retrieving results for each one the way normal Search would, and pulling the passage that answers each part most directly — often from several different pages, none of which need to rank first for the original query.
That fan-out step is the part most explanations of AI search optimization skip, and it explains outcomes that otherwise look random: a page ranking eighth gets cited while the page ranking first does not, because the eighth-ranked page happens to answer one specific sub-question in one clean sentence.
Why does ranking first not guarantee a citation?
Because the citation is decided at the sub-question level, not the query level. AI Overviews run something closer to several small searches behind one visible answer, and each of those smaller searches has its own ranking, separate from the one you can check by typing your target keyword into Google [1].
Spearman r ≈ 0.04
A page can rank on page two for the exact phrase you optimised it for and still be cited, because the citation came from a sentence in that page that happened to answer a sub-question the system generated on its own. This is why chasing the number-one spot for a head term is a weaker lever here than writing the individual sub-answers clearly.
How much overlap is there with the normal top 10?
Not much, and less than most SEO advice assumes. Ahrefs cross-referenced 15,000 long-tail queries across Google, Bing, ChatGPT, Gemini, Copilot and Perplexity and found that only around 12% of the URLs cited by an AI assistant also appeared in Google's own top 10 results for that identical query [3].
| Weak or no predictor | What actually matters instead |
|---|---|
| Ranking #1 for the head keyword | Answering the specific sub-question the fan-out generates |
| Total word count on the page | One passage that states a fact in 2–3 self-contained sentences |
| Publishing more pages on the topic | One page that answers a narrow question completely |
| Keyword density in the body copy | A clear heading matching the question, directly above the answer |
This is uncomfortable for teams used to treating rank position as the whole game, but it also means a page ranking on page two is not out of contention the way it would be for a normal click. If it answers a sub-question no other page states as cleanly, it can still be the one an AI Overview pulls into the answer.
Perplexity is the one exception worth knowing about: its citations overlap with Google's own top 10 far more than any other assistant's do, which makes a strong ranking a genuinely stronger signal there than it is for an AI Overview specifically. How Perplexity ranks its sources explains why that engine behaves differently from the rest.
What does this mean for how you should write a page?
Write each section so it stands alone as the answer to the heading above it. A fan-out system extracts a passage, not a page, so a passage that needs the paragraph before it to make sense is a passage that cannot be lifted cleanly, no matter how well the page as a whole ranks.
Put the plainest version of the answer first in the section, then the detail and the caveats after it. The local business questions buyers ask are a good test case: "does this clinic take walk-ins" needs a direct yes-or-no-and-when in the first sentence, not three paragraphs of context before the answer arrives.
Server-rendered, reachable pages matter here as much as they do for any other engine — a sub-question cannot be answered from a page Google's crawler could not read in the first place, so the fan-out mechanism does not change the reachability requirement underneath it, it just changes what you are optimising for once a page is reachable.
How do you check whether it is working?
Through Search Console's AI features report, which Google rolled out worldwide alongside its August 2026 core update and which sits inside the ordinary Performance report rather than as a separate tool [1]. It shows impressions and clicks attributable to AI Overviews and AI Mode specifically, broken out from standard blue-link results.
Watch the trend over weeks rather than reacting to a single day, since which sub-questions a fan-out generates for a given query can shift between requests. A page that gained AI-feature impressions after a rewrite is a stronger signal than any one screenshot of an answer. Our AI Overviews and AI Mode service is built around mapping that fan-out for a real query set rather than guessing at it.
Questions people also ask
Do I need to rank in Google's top 10 to be cited in an AI Overview?
No. Ahrefs found that only about 12% of URLs cited across AI assistants also sit in Google's own top 10 for the identical query, so ranking helps but is not the mechanism AI Overviews use to pick a source.
Does a longer page get cited more often?
No. Across 174,048 pages Ahrefs found in AI Overview citations, word count had almost no relationship to citation frequency — short pages, long guides and everything between all appeared.
Why does an AI Overview cite three different domains for one query?
Because the query is usually broken into several smaller questions behind the scenes, and each one is answered from whichever page states that particular fact most directly, not necessarily from one page that covers everything.
Can I see which of my pages got cited?
Search Console's AI features report, part of the Search generative AI performance data Google rolled out in 2026, shows impressions and clicks from AI Overviews and AI Mode alongside normal search results.
Sources
- AI features in Search — Google Search Central (checked 23 September 2026)
- We analyzed 174,048 pages cited in AI Overviews: short vs. long content — Ahrefs (checked 23 September 2026)
- Only 12% of AI cited URLs rank in Google's top 10 for the original prompt — Ahrefs (checked 23 September 2026)
Find out which of your pages Google's AI features are already citing
Book a free 30-minute consultation with Ibrahem and Jeremy. We look at your site while we talk, and you leave with a prioritised plan.
Who wrote this
Jeremy ZavaCo-founder
Leads growth and client partnerships.
Works on: CRO, SEO
Reviewed before publishing by Ibrahem Helal, Co-founder.
- AI search optimization: the 2026 guideHow to get cited by ChatGPT, Gemini, Perplexity, Claude, Copilot and Google's AI features: crawler access, citable pages, third-party proof and measurement.
- Get ChatGPT to Recommend Your BusinessChatGPT builds local recommendations from what it reads about you online, not a paid listing, so accurate, consistent hours and address matter most.
- Cloudflare AI bot blocking: what it doesCloudflare's AI Crawl Control toggle blocks AI crawlers site-wide in one click, including some that let you get cited. What it blocks, and how to fix it.
- How Perplexity picks and ranks its sourcesPerplexity crawls its own index, fetches pages live, and cites more sources per answer than any rival — with the closest overlap to Google's rankings.