Quick answer: Google’s AI Overviews don’t simply pick the top-ranking pages. When you search, Google’s AI breaks your question into many smaller sub-questions (a process it calls “query fan-out”), searches for each one separately, then builds its answer from the most authoritative, well-structured passages it finds across all of them. This is why a page ranking #1 isn’t guaranteed a citation, and why a page ranking far down for a related sub-question can get cited instead. In practice, AI Overviews favor sources that Google already trusts, that cover a topic comprehensively, and that are structured so a clean, self-contained answer can be lifted straight from the page. The big shift: ranking and citation have partly decoupled — recent studies show only around 17–38% of cited pages also appear in the organic top 10, down from roughly 76% a year and a half earlier.
If you’ve ever wondered why your top-ranking page didn’t make it into an AI Overview — while a competitor’s deeper, lesser-known page did — the answer lies in how Google’s AI actually selects its sources. It’s a different process from traditional ranking, and understanding it changes how you think about visibility. This analysis breaks down, in plain language, how AI Overviews pick sources and what the data reveals about getting cited.
What AI Overviews?
AI Overviews are the AI-generated summaries that appear at the top of many Google searches, pulling together information from multiple sources and citing them with links. They’ve moved from experiment to a core part of Google Search, reaching billions of users. For anyone who cares about being found online, the key question is no longer just “how do I rank?” — it’s “how do I get cited in the summary sitting above the rankings?”
To answer that, you have to look under the hood at how sources get chosen.
The core mechanism: query fan-out
The single most important concept to understand is query fan-out, and it’s a process Google has publicly confirmed.
Here’s what happens. When you type a question, Google’s AI doesn’t just evaluate the pages that rank for those exact words. Instead, it decomposes your query into a set of related sub-questions — the smaller questions a complete answer would need to cover — and runs searches for each of them, in parallel. Then it retrieves promising sources from across all those sub-searches and synthesizes them into one answer, citing the sources it drew from.
A simple example: if you search “is oat milk good for you,” the fan-out might quietly generate sub-questions like “oat milk nutrition facts,” “oat milk vs dairy milk,” “oat milk for weight loss,” “is oat milk bad for blood sugar,” and “who should avoid oat milk.” The AI Overview is then assembled from the best sources across that whole set — not just the pages ranking for the original phrase.
This one mechanism explains almost everything else about how AI Overviews behave.
Why your #1 ranking no longer guarantees a citation
Query fan-out has quietly rewritten the relationship between ranking and being cited. The data makes the shift striking.
Studies through early 2026 found that the overlap between AI Overview citations and the traditional top 10 organic results dropped to somewhere in the 17–38% range — down from roughly 76% about eighteen months earlier. In other words, being in the top 10 used to strongly predict being cited; now it predicts it far less. Some research on Google’s AI Mode found only around 14% of cited URLs overlapped with the organic top 10.
There are two big reasons ranking and citation have decoupled:
- Fan-out pulls from sub-queries, not just your main query. A page sitting at position 40 for a closely related sub-question can be cited in an AI Overview it would never rank for directly. Conversely, your #1 page might not get cited if it doesn’t cleanly answer the specific sub-questions the AI fanned out into.
- Selection happens at the passage level, not the page level. The AI isn’t asking “is this a good page?” so much as “does this page contain a clean, self-contained answer to this specific sub-question?” A high-ranking page can still lack an extractable passage for the exact thing being asked.
The practical takeaway is important: ranking still helps (higher-ranked pages are cited more often than lower ones), but it’s no longer the whole game. Covering a topic’s full question-space matters more than winning one keyword. One widely-cited finding: pages that rank across multiple related fan-out queries are around 161% more likely to be cited than pages ranking for just the single visible query.
The signals AI Overviews actually weigh
Google hasn’t published an official specification for how AI Overviews select sources. But between Google’s own confirmations and consistent findings across large citation studies, a clear picture emerges. Four signals stand out.
1. Authority and trust
AI Overviews strongly favor sources Google already treats as authoritative at the domain and page level. Because an AI-generated answer stakes Google’s credibility on what it says, the system leans toward sources it trusts. Notably, studies find that earned coverage in trusted third-party publications accounts for a large share of citations, while brand-owned promotional content is cited less often. Being talked about on credible external sites matters.
2. Topical comprehensiveness
Because of fan-out, sources that thoroughly cover a topic — including its related sub-questions — get cited more consistently than pages that address only a single narrow slice. A page that answers the main question and the natural follow-ups in one well-organized place is far more likely to be pulled across multiple fan-out sub-queries.
3. Freshness
Recency is a meaningful signal. Updated, current content tends to be favored, because AI Overviews aim to give accurate, up-to-date answers. Content that’s kept fresh reads as reliable current evidence; stale content gets left behind.
4. Structure and extractability
This is the one most sites underestimate. Because selection happens at the passage level, your content needs to contain clean, self-contained answers an AI can lift without ambiguity. Research consistently shows that certain formats get pulled disproportionately often: comparison tables, question-format headings with direct answers beneath them, clear definition blocks, and step-by-step lists. If the AI has to untangle your point from a wandering paragraph, it’s more likely to pick a competitor who stated it cleanly.
Where structured data fits in
A common question: does schema markup make AI Overviews cite you? Here’s the honest, evidence-based answer.
Schema markup is not a magic “cite me” switch, and Google has said no special markup is required to appear in its AI features. But structured data does real, supporting work in this process. It helps Google understand your content and entities clearly, which reduces ambiguity about what your page says and who’s saying it — and clarity is exactly what the passage-selection step rewards. Independent analysis has also found that schema markup, along with engagement signals and content clarity, can help tip source selection in your favor.
Think of it this way: fan-out means the AI is constantly asking “what exactly does this passage say, and can I trust the source?” Structured data helps answer both halves of that question. It’s part of making your content clean and machine-readable — not a shortcut around the authority and comprehensiveness that ultimately drive citations.
What this analysis means for you
Put the mechanism and the signals together, and a clear strategy falls out. To improve your chances of being chosen as an AI Overview source:
- Cover the whole question-space, not one keyword. Answer the main question and its natural sub-questions in one comprehensive, well-organized page. This is how you get pulled across multiple fan-out sub-queries.
- Write extractable, answer-first passages. Lead sections with direct answers. Use question-based headings, comparison tables, and clear lists so an AI can lift a clean, self-contained response.
- Build genuine authority. Earn mentions and coverage from trusted third-party sources. This is one of the strongest citation signals and can’t be shortcut.
- Keep content fresh. Update your key pages regularly and keep facts current.
- Make it machine-readable. Use clean structured data so Google can understand your content and entities without guessing.
- Check crawlability first. None of this works if Google can’t access your content, so make sure you aren’t accidentally blocking crawlers.
Notice how much of this overlaps with simply being genuinely helpful. That’s not a coincidence — Google’s whole aim is to surface trustworthy, clearly-presented answers, which is exactly what its helpful-content guidance has always encouraged.
The bottom line
AI Overviews choose their sources through query fan-out: Google breaks your question into many sub-questions, searches each one, and builds its answer from the most authoritative, comprehensive, fresh, and cleanly-structured passages it finds across all of them. That’s why top rankings no longer guarantee a citation, and why the winners are pages that cover a topic thoroughly and state their answers in a way an AI can lift directly.
You can’t control Google’s algorithm, but you can control whether your content is trustworthy, comprehensive, current, and machine-readable. Get those right and you give your pages the best possible chance of being the source the AI reaches for.
Structured data is the machine-readable layer that helps Google understand exactly what your content says and who your brand is — the clarity that passage-level selection rewards. SchemaEngine AI generates and validates clean schema markup for your WordPress site automatically, so both search engines and AI systems can read your pages without guesswork — no code required. Try the free plugin →
Frequently asked questions
How do AI Overviews pick their sources?
Through a process Google calls query fan-out. When you search, Google’s AI breaks your question into multiple related sub-questions, searches for each independently, and assembles its answer from the most authoritative, comprehensive, fresh, and clearly-structured passages it finds across all of them — then cites those sources. It’s not simply the top-ranking pages for your exact query.
Does ranking #1 guarantee a citation in an AI Overview?
No. Studies show only around 17–38% of AI Overview citations also appear in the organic top 10, down from roughly 76% eighteen months earlier. Higher rankings still help your odds, but query fan-out means pages can be cited for related sub-questions they don’t rank for directly — and a top page can be skipped if it lacks a clean, extractable answer.
What is query fan-out?
Query fan-out is Google’s technique of decomposing a single search into several related sub-queries, retrieving sources for each in parallel, and synthesizing an answer from across all of them. It’s why AI Overviews cite sources beyond those ranking for the main keyword, and why comprehensive content covering a topic’s full question-space gets cited more often.
Does schema markup help you get cited in AI Overviews?
Not as a guarantee — Google says no special markup is required. But structured data helps Google clearly understand your content and entities, which reduces ambiguity during the passage-selection step. Independent analysis suggests schema, along with clarity and engagement, can help tip source selection in your favor. It supports citations rather than directly causing them.
Why does content structure matter so much for AI Overviews?
Because source selection happens at the passage level. The AI looks for a clean, self-contained answer to each sub-question, so content with question-based headings, comparison tables, direct answer blocks, and clear lists gets pulled more often than answers buried in long paragraphs. Extractable structure makes your content easy to lift.
Can I opt out of AI Overviews without hurting my search visibility?
Not cleanly. The available directives that would remove your content from AI Overviews (like nosnippet) also strip your regular search snippets, and there’s currently no AI-Overview-only opt-out. In practice, most sites are better off optimizing to be cited well than trying to opt out.



