Peer-reviewed study finds 29.8% of AI Overview citations don't rank on Google's first page

A peer-reviewed study from Washington University in St. Louis, presented at the October 2026 ACM Internet Measurement Conference, tracked 55,393 Google searches over 40 days and found that 29.8% of the domains cited by AI Overviews never appeared in the first-page organic results shown for the same queries. According to a dev.to write-up of the research, the finding undercuts the widespread assumption that Google's AI answers simply repackage the pages that already rank.

Nearly a third of citations bypass page one

The researchers logged every AI Overview shown between March 13 and April 21, 2026, along with its citations, the first-page results displayed beside it, and the full content of each cited page. Nearly three in ten cited domains were absent from page one entirely, meaning a page can be quoted in Google's AI answer while being invisible in the conventional rankings for that same query.

Independent data points the same way. Ahrefs analyzed 863,000 keywords and 4 million AI Overview URLs and found that only 38% of cited pages also appeared in the organic top 10, down sharply from 76% in July 2025. BrightEdge put the overlap even lower, at 17%. As the dev.to author notes, the link between organic ranking and AI citation is loosening over time, not tightening.

How Google picks AI Overview sources

The likely mechanism is query fan-out, described in Google's own documentation. Rather than answering the original query directly, the system splits it into multiple related sub-queries, retrieves pages for each, and assembles the response from the combined pool. A page that does not rank for the primary query can still be cited if it answers one of the offshoot questions.

When AI Overviews appear

The study also measured how often the feature triggers at all:

• 13.7% of all searches produced an AI Overview.
• Question-form queries triggered one 64.7% of the time, versus 9.5% for non-question queries.
• For non-question searches, activation climbed from 9.9% for one-word queries to 38.7% for queries of six words or more.
• Topic mattered: 46.1% of hobby and leisure searches and 39.9% of science searches got an overview, while politics (7.5%) and law and government (9.6%) rarely did.

The pattern favors natural-language questions over terse keyword strings, which has direct implications for how content gets discovered.

Credibility and the long tail

Across 61,212 scored citations, domains cited by AI Overviews earned higher average credibility ratings than the first-page results shown for the same searches. Credibility was measured with a composite metric (PC1) that combines domain authority, content quality and trust signals, not a single factor a site can toggle with a markup file.

The average overview cited roughly eight sources. Concentration is real but not absolute: the top five hostnames took 20% of citations and the top 50 took 48.3%, leaving a long tail spread across 7,379 hostnames. There is room for sites beyond the usual aggregators, provided the content is specific enough to surface through fan-out retrieval.

One in ten claims lacks support

The researchers decomposed responses into 98,020 atomic claims and checked each against its cited source: 89% were clearly or broadly supported, 7% were not addressed by the cited text at all, and 4% directly contradicted it. Only 41.9% of overviews were fully grounded, with every claim backed by its citation. The authors caution that the roughly 11% error rate is an upper bound, since their pipeline could not fully capture dynamic content such as Reddit threads and YouTube videos.

Why it matters

For publishers and SEO teams, the data reframes the problem. Holding a top position for a head keyword no longer guarantees citation, and lacking one no longer rules it out, as the nearly one-third of cited pages outside page one demonstrate. The second path into an AI Overview runs through fan-out retrieval, which rewards covering a topic's related sub-questions with credible, specific material rather than chasing a single ranking position, and Google's own guidance points the same way, favoring original content with a distinct perspective over commodity pages.

For readers, the accuracy findings are the sharper concern. In a meaningful share of cases the citation displayed does not actually support the claim attached to it, and only about four in ten overviews are fully grounded end to end. Anyone treating an AI Overview's source list as verification should recalibrate accordingly.