Field Notes · GEO
Google AI Overviews: Who Gets Cited
AI Overviews used to mirror the top ten results, but a fast-widening gap between ranking and citation has rewritten the rules of who Google surfaces and why.
Ranking first in Google used to come with a quiet guarantee: if AI Overviews answered the query, you would be one of the sources it named. In 2026 that guarantee is gone. Ahrefs found that 76.10 percent of pages cited inside AI Overviews ranked in the organic top ten in its July 2025 study; by early 2026, later Ahrefs data put that overlap at roughly 38 percent. The retrieval layer that writes Google's answers has quietly decoupled from the ten blue links beneath it.
That single shift reframes everything else. AI Overviews are no longer a tidy re-skin of the rankings you already understand. They are a separate retrieval system that pulls passages, synthesises them into an answer, and credits a handful of sources, and the rules for being one of those sources are not the rules of classic SEO. This Field Note walks through what is actually measured, what is contested, and what you can do about it, with an honest line drawn between evidence and speculation.
What exactly is an AI Overview?
An AI Overview is the generated summary Google places at the top of a results page, stitched from multiple web sources and shown with links to some of them. Google launched it for US users at I/O on 14 May 2024, expanded to six more countries including the UK, India and Japan that August, and reported reaching more than 100 countries and over a billion monthly users by October 2024. Early stumbles, the now-infamous suggestions to eat rocks and glue cheese to pizza, forced Google to scale back how aggressively the feature triggered.
Mechanically, the important detail is that an AI Overview does not pick one winning page. It assembles an answer from several and attributes a subset. Reporting on the citation data suggests 88 percent of AI Overviews cite three or more sources and only around 1 percent cite a single source, which points to a system designed to favour source diversity rather than reward one dominant result.
How often do AI Overviews actually appear?
Honestly, nobody can give you one trustworthy number, and anyone who quotes a single universal figure is overstating it. Credible measurements range from roughly 15 percent to around 50 percent of searches depending on the keyword set, geography, device and detection method. Semrush tracked coverage rising from 6.49 percent of queries in January 2025 to a peak near 25 percent in July, then falling to 15.69 percent by November after Google recalibrated. Conductor's Q1 2026 benchmark across 21.9 million queries landed at 25.11 percent, while a March 2026 secondary aggregation reported about 48 percent and Google's own disclosure says roughly half.
- The spread is real, not noise: it reflects genuinely different methodologies, not one true figure waiting to be found.
- Coverage is also a moving target Google adjusts deliberately, as the 2025 peak-then-pullback shows.
- For planning, treat AI Overviews as common on a meaningful share of your queries and rising, not as a fixed percentage.
Which queries trigger them, and which sources get cited?
AI Overviews skew heavily towards informational, longer-tail queries, though that mix is broadening. In Semrush's study of more than 10 million keywords, 88.1 percent of AI Overview triggers were informational, and the queries themselves ran longer, averaging around four words against two for a standard search. That informational share was already falling through 2025, from 91.3 percent in January to 57.1 percent by October, as commercial and transactional overviews grew. The feature started as an answer machine for how and why questions and is steadily moving towards comparison and purchase intent.
On who gets cited, ranking still helps but no longer decides. In the July 2025 Ahrefs data the median organic rank of a primary citation was position two, the second citation position four, and the third position five, a moderate correlation rather than a deterministic one. The strongest off-site signal was striking: brand mentions on YouTube correlated more closely with AI Overview visibility than any other factor in Ahrefs' analysis of 75,000 brands. Being talked about, on the platform Google owns, mattered more than any single on-page tweak.
Why ranking number one no longer guarantees a citation
Ranking number one no longer guarantees a citation because AI Overviews now draw heavily from pages that do not rank in the top ten. The overlap fell from 76.10 percent in July 2025 to roughly 38 percent in early 2026, with 86 percent of citations still coming from somewhere in the top 100. The mechanism credited for this is query fan-out: Google splits one question into several sub-queries, retrieves sources for each independently, and assembles the answer from across all of them. A page can win a sub-question it is uniquely good at without ranking highly for the headline term.
Different vendors measure the overlap differently, which is worth knowing before you anchor on any one figure. Originality.ai has reported a top-ten overlap closer to 52 percent, and seoClarity and BrightEdge land elsewhere again, because each uses a different method. The direction of travel is the robust finding: citation and ranking are pulling apart, and an entity-dense, self-contained passage that answers a narrow sub-question is now a distinct asset from a page that ranks.
What do AI Overviews do to click-through?
AI Overviews reduce clicks to traditional results, on the strongest independent evidence available, though Google disputes it. Pew Research Center's study, the most defensible to date, found users who saw an AI summary clicked a traditional result on 8 percent of visits, against 15 percent when no summary appeared, close to half. They clicked a link inside the summary on just 1 percent of visits, and ended their session on 26 percent of pages with a summary versus 16 percent without. The sample was 900 US adults across 68,879 unique searches in March 2025, of which 12,593 produced a summary. Google publicly called the methodology flawed and not representative.
Ahrefs measured the same direction from a different angle, recording a 34.5 percent drop in position-one click-through when an AI Overview was present across roughly 300,000 keywords. The takeaway is not that citation is worthless. In-overview clicks are tiny, but a citation is brand visibility whether or not it earns a click, and vendor data suggests cited pages still out-click uncited competitors. The honest framing is fewer clicks, but citation still matters for visibility and brand.
How do you earn a place as a cited source?
You earn citations by writing self-contained, entity-dense passages that answer narrow questions outright, and by building the off-site brand signals that Google's retrieval rewards. The single robustly-sourced optimisation finding is Ahrefs' YouTube result: brand mentions there correlated more strongly with visibility than anything on the page itself. Everything below follows from how fan-out retrieval works, that it pulls discrete passages, prefers diverse sources, and increasingly looks beyond the top ten.
- Answer the heading in the first sentence. Every H2 should be a clear question or claim, with the opening line resolving it in one or two sentences so a passage can be lifted whole.
- Build entity density. Name the products, places, people and specifics a model needs to recognise your passage as the authoritative answer to a sub-question.
- Invest in brand mentions off-site, with YouTube as a priority given its measured correlation with AI Overview visibility.
- Keep ranking, but stop treating it as the finish line. Top-ten placement still raises your odds; it no longer gates them.
- Be quotable, not just rankable: structure content so a single self-contained block stands on its own once removed from the page.
A note on the popular optimisation statistics doing the rounds in 2026: claims that high semantic-completeness scoring makes content several times more likely to be cited, that ideal extracts run to roughly 134 to 167 words, or that combining text, image, video and structured data lifts selection rates by triple digits, all come from vendor analyses, not peer-reviewed work. They are useful as direction, not as fact. Treat them as hypotheses to test on your own pages rather than laws to build a strategy on.
The story of AI Overviews in 2026 is not that search died or that clicks vanished. It is that the cited answer and the ranked page have become two different things to win, governed by different signals. The agencies and teams that internalise that split, and write for passage-level retrieval while still earning rankings and brand mentions, are the ones being named in the answer. That decoupling is the whole game, and it is exactly the work we do at DarkHorse: building pages that get found on Google and cited by AI, measured in numbers rather than narratives.
The cited answer and the ranked page are now two prizes, not one. Win both, and measure it in citations, not vibes. That is the work we do at DarkHorse.
Questions people ask
Do I still need to rank in the top ten to appear in AI Overviews?
Ranking helps but no longer decides. In Ahrefs' July 2025 data the median primary citation sat at organic position two, yet by early 2026 only around 38 percent of citations came from the top ten, down from 76 percent. Top-ten placement raises your odds of being cited; it is now a strong signal rather than a gate, largely because query fan-out lets pages win narrow sub-questions without ranking highly for the headline term.
How often do AI Overviews appear in search results?
There is no single reliable figure, and credible measurements range from roughly 15 to 50 percent depending on methodology. Semrush tracked coverage falling to 15.69 percent by November 2025 after a July peak near 25 percent, Conductor's Q1 2026 benchmark recorded 25.11 percent, and Google's own disclosure says about half. The honest answer is a range, because keyword set, geography, device and detection method all move the number.
Is it worth being cited if hardly anyone clicks the link?
Yes, because a citation is brand visibility even without a click. Pew Research found in-overview link clicks happened on just 1 percent of visits, so the direct traffic is small, but appearing as the named source still puts your brand in front of the user inside the answer they trusted. Vendor data also suggests cited pages out-click uncited competitors, so framing citation as pointless because of low click-through misreads what it delivers.
What is the single most effective thing I can do to get cited?
Make every section answer its heading outright in the first sentence, so a self-contained passage can be lifted whole into an answer. Beyond on-page work, the strongest measured off-site signal is brand mentions on YouTube, which correlated with AI Overview visibility more closely than any other factor in Ahrefs' study of 75,000 brands. Combine quotable, entity-dense passages with genuine off-site brand presence rather than relying on rankings alone.