Get the app
AI visibility and answer engines

How does Google AI Overviews pick sources?

What is known about how AI Overviews selects what it quotes, what follows from it, and what nobody outside Google actually knows.

3 min read
Short answer

AI Overviews draws on Google's existing index, so ordinary search visibility is the prerequisite. Beyond that, it favours pages that answer the specific sub-question directly, come from sources it treats as reliable, and can be quoted in a sentence or two. Google has not published the selection mechanism, so anything more precise than that is inference.

Start from what is documented

Google states that AI Overviews are grounded in its search index and links to supporting pages. Two things follow with confidence:

  • If you are not indexed, you cannot appear. Indexation is a precondition, not an optimisation.
  • Blocking Googlebot removes you. Google-Extended is a separate control governing Gemini training and grounding, and Google documents that it does not affect inclusion or ranking in Google Search — so blocking it does not remove you from AI Overviews eligibility, which follows ordinary Search eligibility.

What the observable pattern suggests

Watching which pages get cited across many queries produces a consistent picture. This is inference from observation, not documentation:

  • Sub-question matching. An overview decomposes a query into parts and cites a different source for each. You can be cited for one clause of a question you do not rank highly for overall.
  • Passage extractability. Cited passages tend to be short, direct and self-contained — a definition, a number, a step.
  • Source treatment varies by topic. For health, finance and other consequential subjects, institutional and established sources dominate heavily. For practical or technical questions the field is much more open.
  • Freshness where it is relevant. Time-sensitive queries favour recent pages. Stable topics do not show the same effect.

What to do about it

  1. Fix indexation first. Confirm the page is in Google's index at all — nothing else matters until it is.
  2. Break broad pages into the specific questions they answer, each with its own heading.
  3. Put a direct one- or two-sentence answer immediately under each heading.
  4. Add real attribution: named author, organisation, date, sources.
  5. Use structured data that fits — FAQPage, HowTo, Article — where it genuinely describes the page.
The uncomfortable part, stated plainly: AI Overviews reduce clicks for informational queries, including for the sites they cite. If your traffic depends entirely on answering questions people can now have answered in place, that is a strategic problem and no amount of optimisation solves it. The durable responses are content that cannot be summarised away — tools, data, community, depth — rather than trying to be summarised more favourably.

Ignore anyone claiming certainty

Nobody outside Google knows the ranking mechanism for AI Overviews, and it changes. Advice presented with precise weightings and guaranteed methods is invented. The defensible strategy is the boring one: be indexed, be specific, be attributable, be genuinely useful — which is also what survives the next change.

What our audit reports about this

Every item below is measured directly, not inferred. Run it against your own site and the result names the exact rule or header responsible.

  • Whether Googlebot and Google-Extended are allowed, reported separately with the rule for each.
  • Whether the page carries valid structured data of an appropriate type.
  • Whether headings decompose the topic into answerable questions.
  • Whether authorship, organisation and dates are present and machine-readable.

For agents and scripts, the same measurement is at /api/v1/seo?url=yoursite.com — see the API documentation.

Sources

Claims about crawler behaviour on this page come from the operators' own documentation. Vendor documentation last checked . Where a vendor does not state something, this page says so rather than filling the gap.

Related questions

Does blocking Google-Extended remove me from AI Overviews?

No. Google documents Google-Extended as controlling Gemini training and grounding, and states that it does not affect inclusion or ranking in Google Search. Google also documents that AI Overviews and AI Mode have no additional technical requirements beyond ordinary Search eligibility, and that no AI-specific file is needed. Googlebot access is what matters.

Do AI Overviews reduce my traffic?

For informational queries, generally yes, including for cited sites. Being cited is better than being absent, but it does not restore the click-through of the previous era.

Can I opt out of AI Overviews but stay in search?

The nosnippet and max-snippet directives limit what can be shown, which reduces overview eligibility — and also removes your ordinary search snippet, which usually costs more than it saves.

Read next

What is generative engine optimization (GEO)?

Optimising to be quoted inside an AI answer rather than ranked in a list of links — what genuinely differs from SEO, and what is the same work under a new name.

Read

Why is my website not showing on Google?

Work through it in order — not indexed, indexed but not ranking, or ranking but not for anything you noticed.

Read

What is structured data and do I need it?

Machine-readable facts about your page. Worth adding where it genuinely describes the content, worthless — and occasionally harmful — everywhere else.

Read

How do I add an FAQ section that search engines actually use?

Real questions, visible answers, correct markup — in that order. Schema on questions nobody asked achieves nothing.

Read

All 100 guides · How every check works · API for agents