How do AI Overviews pick their sources?

AI Overviews start from Google's core ranking system, then branch out on their own. Google fires off several related searches behind one question and pulls whichever page answers each part best, not just the page sitting at position one. Ranking well still helps, but by 2026 only 38% of cited pages held a top 10 spot, down from 76% a year earlier.

Google's AI Overviews have gone from a novelty at the top of the results page to the thing standing between your site and roughly half of all searches. Google's own tracking puts AI Overview presence at about 48% of queries in February 2026, up from about 31% a year before, and the box itself has grown taller too, averaging around 1,200 pixels, more space than a standard desktop screen shows before anyone scrolls (BrightEdge). If you don't know how it picks who gets quoted in that space, you are optimizing for a result fewer and fewer people ever reach.

Does Google use the same ranking system for AI Overviews?

Yes, as a starting point. Google says plainly that its "generative AI features on Google Search are rooted in our core Search ranking and quality systems," and that a page has to be indexed and eligible to show a normal snippet before it can be considered at all. From there, the system runs its own searches instead of reading straight down the results page.

That eligibility bar is worth sitting with for a second: there is no separate submission process or AI-only sitemap, just the same indexing bar as regular search. Google's developer documentation states there "are no additional requirements to appear in AI Overviews or AI Mode" beyond being indexable and snippet-eligible in ordinary search (Google Search Central). What happens once you clear that bar, which pages actually get picked, is a separate and much less predictable question.

The mechanism behind that picking is called query fan-out. Robby Stein, Google's VP of Product for Search, described it plainly: "If you're asking a question like things to do in Nashville with a group, it may think of a bunch of questions like great restaurants, great bars, things to do if you have kids, and it'll start Googling basically" (Search Engine Journal). Your one search becomes several searches you never see, each pulling its own set of candidate pages, and the summary you get back stitches together whichever passages answered each piece best.

Do you have to rank in the top 10 to get cited?

Not anymore, and less each year. Ahrefs tracked 4 million AI Overview URLs across 863,000 keywords and found only 38% of cited pages also ranked in Google's top 10, down from 76% in a study it ran just eight months earlier. A growing share of citations now come from pages ranking beyond position 100, or not ranking at all.

This is the number that should worry anyone still treating "top 10 in Google" as the finish line. Ahrefs' comparison of its two studies, 1.9 million citations in the July 2025 pass against 4 million in the 2026 follow-up, shows the overlap between AI Overview citations and ordinary blue-link rankings collapsing in under a year (Ahrefs). A separate BrightEdge analysis of one year of AI Overview data found citations from positions 21 to 30 up 400%, with 89% of citations coming from beyond the top 100 organic results entirely (Search Engine Land). Different vendors land on different exact percentages, but they agree on direction: the fan-out searches keep finding pages the ordinary ranking algorithm would never surface for that query.

What actually predicts whether a page gets cited?

Domain-level trust, more than any single on-page trick. A study of 1,000 AI Overviews and 4,243 unique cited URLs found Domain Authority correlated with citation rate at 0.61, the strongest single factor measured, with each 10-point jump in DA raising citation odds by roughly 1.4x. Structured data helped too: schema-marked pages were cited 2.3x more often.

None of this is a guarantee, and the same study flags a competing analysis that found schema's effect disappears once domain authority is controlled for (Digital Applied). Treat these as directional, not settled law. What they do rule out is the comforting idea that a page can format its way to a citation regardless of how established the site behind it is. Domain trust is doing real work in this selection, the kind built over years rather than an afternoon, a theme we cover from the crawler side in do AI crawlers respect robots.txt.

Which sites dominate AI Overview citations?

A small number of large, trusted domains, and the mix shifts hard by topic. Across 36 million AI Overviews and 46 million citations, YouTube appeared in 23.3% of citations overall and Wikipedia in 18.4%, but for gaming queries YouTube alone showed up in 93% of AI Overviews.

That concentration is the uncomfortable part for smaller, newer sites. Surfer SEO's citation study puts YouTube, Wikipedia, and Google's own properties at the top of nearly every category it measured, with the exact mix shifting by topic: NIH and Healthline dominate health queries, LinkedIn also turns up in finance, Reddit sits near the top for gaming, just behind YouTube (Surfer SEO). If your category's incumbents are this entrenched, the realistic path in is not displacing Wikipedia. It's making sure your own pages are the ones the fan-out surfaces for the specific sub-question only you answer well, the same approach we lay out in how to get cited by ChatGPT.

Does getting cited even send you traffic?

Less than a normal search result does, which is worth knowing before you chase citations for their own sake. Pew Research found users clicked a traditional search result in just 8% of visits when an AI summary appeared, versus 15% when it didn't, nearly half the rate.

Pew's numbers come from real browsing behavior, not a survey: 900 U.S. adults, 68,879 searches, with an AI summary showing up on about 18% of them (Pew Research Center). Only 1% of all visits included a click on a link inside the summary itself. Set that beside the 30% year-over-year drop in search clicks that BrightEdge measured while impressions rose 49% (Search Engine Land), and the honest read is that a citation works more like a trust signal than a traffic source. It tells a buyer, and the model, that you are a credible answer, even on the visits where nobody clicks through. That's a real value, just a different one than a blue-link ranking used to deliver, which is the gap GEO vs SEO exists to explain.

So what should you actually check?

Whether your pages clear the eligibility bar (indexed, snippet-eligible), whether the domain carries enough earned trust to compete for fan-out queries, and whether you are cited for the specific sub-questions in your space, tracked over time rather than guessed at once.

You can't verify any of this by staring at your own homepage. You need to run the actual queries your buyers would ask, watch which pages the fan-out pulls in, and repeat that on a schedule, since AI Overviews change from one run to the next. That repeatable check, plus the crawl and content diagnostics underneath it, is what BeCited runs against your domain, scored against the same methodology referenced here. Nothing on that report is presented as a guarantee of a citation. It's a measurement of where you stand today and what changed since the last one. The rest of what we've learned about this shift is on the blog.