Short answer. An AI Overview is not a summary of page one. Google decomposes your question into a set of related sub-queries, retrieves separately for each, and assembles an answer from whatever passages best serve them. In the largest public study of the question — Surfer SEO's December 2025 analysis of 10,000 keywords and 173,902 URLs, reported by Search Engine Land — about 68% of cited pages ranked in Google's top 10 for neither the main query nor any of its fan-out queries. Ranking is one way into the candidate pool. It is not the qualifier.
Most advice about AI Overviews still assumes the citation list is a re-ordering of the organic results. It is not, and the gap between the two has widened every quarter since the feature launched. This piece sets out how the selection mechanism works, what the measured evidence says about who gets cited, and what a citation is actually worth once it arrives.
How the selection actually works
Google's AI Overview does not read your page and decide whether it is good. It runs a retrieval pipeline, and your page is only ever in contention for the passages it holds.
- The query is decomposed. One typed question becomes a set of related sub-queries — commonly called query fan-out — including narrower specifications, canonical rephrasings, translations and clarifications of the original.
- Each sub-query retrieves separately. Google runs retrieval for each branch and collects candidates. A page can enter the pool through a sub-query it was never written for.
- Passages are ranked, not pages. The system chunks candidates and scores the chunks. It never quotes a page; it quotes a paragraph, a table row or a list item.
- The answer is synthesised and attributed. One answer is written from the selected passages, and the sources drawn on are linked. Attribution is per-passage, which is why cited pages often look unrelated to the ranked results below.
Ranking gets you into one pool. Fan-out decides how many pools you are in.
Why ranking in the top 10 is not the qualifier
The Surfer SEO study extracted 33,000 fan-out queries with Gemini and then checked where the cited pages actually ranked. About 68% of pages cited in AI Overviews ranked in the top 10 for neither the main query nor any fan-out query. That is the headline, but the second half of the finding is the actionable one: pages ranking for the main query and at least one fan-out query were 161% more likely to be cited, and accounted for 51% of all citations. The study reported a Spearman correlation of 0.77 between the number of fan-out rankings a page held and its citation probability.
Breadth across the question's neighbourhood beats height on the question itself.
The trend line agrees. Overlap between AI Overview citations and the organic top 10 has collapsed over eighteen months, though the trackers disagree on the level:
| Measure | Mid-2024 | February 2026 |
|---|---|---|
| ALM Corp, via Omnibound | ~76% | 38% |
| BrightEdge, via Omnibound | ~76% | 17% |
Two trackers, two answers. The direction is agreed and large; the level is not, so the honest quotation is the range — 62–83% of citations now come from outside the top 10 — rather than either endpoint.
What this changes about what you publish
If citations are won at the sub-query level, a page's job changes. It is no longer to be the best answer to one keyword; it is to hold defensible passages on several adjacent questions at once.
- Cover the question's neighbourhood on one page. Definition, mechanism, exceptions, comparison, cost, limits — each under its own heading, each answerable without the others.
- Write headings as the sub-questions themselves. A fan-out query is a question a person would type. A heading phrased that way matches it directly; a heading like "Our approach" matches nothing.
- Front-load. Omnibound's 2026 AEO statistics compilation found 55% of sampled AI Overview citations came from the first 30% of the cited page.
- Keep passages self-contained. A paragraph opening with "it" or "this approach" cannot be lifted without its neighbour, so it is not lifted at all.
- Put structure where the facts are. Tables and lists are cleanly extractable units. Prose with the numbers buried mid-sentence is not.
There is an awkward implication for reporting. A page can gain citations while losing rank, and lose citations while holding rank. If AI visibility is reported against position tracking, the two will drift apart and the report will describe a system that no longer exists. Measure citations directly against a fixed question set, or do not claim to be measuring AI visibility at all.
Why recently updated pages are favoured
The second-clearest signal after structure is recency, and it is unusually strong on commercial questions. Omnibound's compilation found that for commercial and evaluation-stage queries, 83% of AI citations came from pages updated within the previous 12 months, and over 60% from pages refreshed within six months.
For most organisations that makes updating cheaper than publishing. A thorough page from 2023 that nobody has touched competes badly against a thinner page revised last month, and the cheapest available win is usually the pages that already answer buyer questions, reviewed and genuinely changed. Re-dating a page you did not edit is a different thing entirely, and answer engines are not the only readers it misleads.
How often does an AI Overview even appear?
Before optimising for AI Overviews it is worth knowing how much of your query set triggers one. The published estimates disagree enormously:
| Source | Trigger rate | Basis |
|---|---|---|
| Semrush, November 2025 | 15.69% of queries | — |
| Conductor, Q1 2026 | 25.11% of queries | 21.9 million queries |
| BrightEdge, February 2026 | ~48% | Commercial verticals |
| Google's own statement | ~50% | US queries |
The spread is a sampling artefact rather than a contradiction: trigger rate depends heavily on query mix, country and device, and informational question-shaped queries trigger far more often than navigational ones. The only figure that supports a business case is the rate across the questions your own buyers ask, measured on your own list.
What is a citation worth in traffic?
Two things are true at once, and most vendor material carries only the convenient one.
Seer Interactive's tracking, compiled by Omnibound, put click-through rate on queries showing an AI Overview at 1.76% in June 2024, falling to 0.61% by September 2025, then recovering to 2.4% by February 2026 — against 3.8% on queries with no AI Overview. Separately, the Pew Research Center's study of 68,000 real search queries, reported by Search Engine Land, found users clicked a traditional result on 8% of searches where an AI summary appeared, versus 15% where none did — a drop of roughly 47%.
The recovery matters as much as the fall. Anyone quoting the September 2025 floor as the current state is quoting a snapshot of a moving system; anyone quoting the recovery without the gap to non-AIO queries is doing the same in the other direction. The honest framing is that an AI Overview citation is worth much less in clicks than a blue link used to be, and worth something real in influence, because it is what the buyer actually reads. Build the case on the second.
What nobody can promise you
- No one controls the citation. There is no submission, no index request and no paid placement. Anyone selling a guaranteed AI Overview citation is selling something they do not own.
- The fan-out set is itself unstable. Surfer SEO found only about 27% of extracted fan-out queries stayed consistent across repeated runs. Optimising for a specific sub-query list is building on sand; optimising for topical breadth is not.
- Your own domain is one lever of several. Omnibound's compilation puts roughly 85% of AI references on third-party sites rather than the brand's own domain.
- Coverage moves under you. Trigger rates have been recalibrated before and will be again, so programme design should survive it happening.
The sibling surfaces behave differently again: Google AI Mode is a separate source list, ChatGPT retrieves from its own index, and Perplexity is markedly more stable. None of them can cite a page their crawler cannot fetch, which is the subject of AI crawler access.
How Lifewood approaches this
Lifewood treats AI Overview visibility as a measurement problem before a content problem. The instrument is a fixed question set per market, run repeatedly, with citations logged directly rather than inferred from rank tracking — because a page can gain citations while losing position, and a blended report hides both. Content work then targets the fan-out neighbourhood rather than a single keyword: one page holding self-contained, evidence-carrying passages on definition, mechanism, comparison and limits.
The multilingual dimension is where the delivery model matters, since a fan-out query in Vietnamese is rarely the English question translated. 50+ languages and 40+ delivery centres across 30+ countries mean the question set and the content are both produced in-market. See AEO services, GEO services and how the three disciplines divide.
Sources and further reading
- Surfer SEO, AI Overview fan-out rankings boost citation odds, December 2025, reported by Search Engine Land — the 10,000-keyword, 173,902-URL study behind the 68% and 161% figures.
- ALM Corp and BrightEdge AI Overview citation overlap trackers, February 2026, compiled by Omnibound.
- Semrush, Conductor, BrightEdge and Google AI Overview prevalence figures, 2025–2026, compiled by Omnibound.
- Seer Interactive click-through rate tracking on AI Overview queries, June 2024 – February 2026, compiled by Omnibound.
- Omnibound, Answer Engine Optimization statistics 2026 — position-in-page, freshness and third-party citation share.
- Pew Research Center, click behaviour study of 68,000 queries, reported by Search Engine Land.