In Brief
- Core Answer: You cannot rank in AI Overviews as a separate objective. Overviews are generated from Google's existing Search index, so the levers are conventional — be indexed, rank well for the query, and write passages that directly answer it in extractable form.
- Why It Matters: A large amount of paid advice sells a distinct "AI Overview optimisation" that does not exist as a separate system. Knowing that redirects effort to the things that do work.
- Best For: SEO owners deciding what, specifically, to change on pages that are losing clicks to AI Overviews.
If you want to know how to rank in AI Overviews, start with the mechanism: Google AI Overviews are generated from the same Search index as organic results, using the same ranking and quality systems. Google's own guidance states there is no separate optimisation for them. What does influence inclusion is being retrievable, ranking for the query, and writing passages that resolve a question completely enough to be lifted.
That sounds deflationary. It is also actionable, because it tells you exactly which work is real.
How an AI Overview Is Built
The mechanism, as far as it is publicly documented and observable:
- A query arrives that Google's systems judge suitable for a generated answer — typically informational, multi-part, or one where synthesis genuinely helps.
- The query is expanded into related sub-queries. This "query fan-out" means a single search runs several underlying retrievals, which is why Overviews cite pages that do not rank for the original phrasing.
- Candidate documents are retrieved from the Search index.
- Passages are selected that support each claim the answer will make.
- The answer is composed and links are attached to the sentences they support.
Two consequences follow directly. First, the unit of selection is a passage, not a page — which is why a page ranking eighth can be cited while the page ranking first is not. Second, fan-out means the query you optimise for is not necessarily the query being retrieved against, so covering the adjacent sub-questions on the same page materially widens your surface.
How to Rank in AI Overviews: What Actually Influences Inclusion
1. Being in the Index and Ranking Reasonably
Overviews draw on Search. Pages excluded from the index cannot appear. Pages ranking on page five rarely do. This makes conventional SEO the precondition, not the alternative — and it makes the fastest AI Overview win an ordinary ranking improvement on a query that already triggers one.
2. Answering the Sub-Question Completely in One Passage
The selected passage must stand alone. A paragraph that begins "This is why the second approach usually wins" cannot be lifted, because the subject is elsewhere. A paragraph that begins "The second approach wins in most cases because..." can.
Forty to sixty words, declarative, no anaphora, one claim. Under every heading. This is the single highest-return change and it is editorial, not technical.
3. Covering the Fan-Out
Because one query becomes many, a page that answers the primary question and four adjacent ones has five chances of being retrieved. Practically: after your main answer, add short sections for the obvious follow-ups — the definitional variant, the cost variant, the comparison variant, the how-long variant. Each gets its own question-shaped heading and its own standalone answer.
4. Specific, Attributed Claims
Generated answers are assembled from claims, and citations attach to the passages that supplied them. Vague statements are not claims. A dated, sourced, numerical statement is quotable in a way that "many organisations find" is not. This also protects you when the model paraphrases, because there is less room to drift.
5. Freshness Where Freshness Is Relevant
For queries where recency matters, honest dateModified values and genuinely updated content are favoured. Bumping the date without changing content is a credibility cost with no upside.
What Does Not Influence It
Claim | Reality |
|---|---|
A special AI Overview markup exists | It does not. Standard structured data is all there is |
llms.txt improves AI Overview inclusion | Google has said it does not use llms.txt for search |
Blocking Google-Extended removes you | It governs Gemini training, not Search inclusion |
You can opt out of Overviews specifically | Only via snippet controls that also affect ordinary results |
On that last row: nosnippet, max-snippet and data-nosnippet do limit what can be shown, and Google has indicated they apply to AI-generated experiences too. But they apply to featured snippets and result descriptions as well, so opting out of one means opting out of the others. For almost every commercial site the trade is not worth making.
Measuring It Honestly
Search Console does not break out AI Overview impressions separately. What you can do:
- Segment question-shaped queries in the performance report and watch impressions against clicks. A widening gap on those queries is the Overview signature.
- Sample manually. Take your thirty priority questions, run them in a clean session, and record whether an Overview appeared and whether you were cited. Repeat monthly with the same set.
- Watch branded search. Uncredited exposure inside an Overview frequently shows up later as branded queries rather than clicks.
- Track position within the citation list. First-cited sources receive disproportionate attention.
Expect the pattern that unsettles most reporting: impressions up, click-through down, branded search up. That is the mechanism working, not failing. Reporting it without alarming the board requires separating the retrieval layer from the consequence layer.
A Page-Level Checklist
- Confirm the page is indexed and ranking on page one or two for its target question.
- Rewrite each H2 as the question a person would type.
- Rewrite the paragraph under each H2 as a standalone forty-to-sixty-word answer.
- Add four short sections covering the obvious follow-up questions.
- Replace every vague claim with a specific, dated, sourced one.
- Add one comparison table if the content is comparative.
- Add Article schema with a real author and an honest dateModified.
- Confirm the page renders server-side and is not dependent on client-side hydration.
Eight items, most of them editorial. There is no ninth item involving a secret configuration, and any advice that offers one is describing something that does not exist.
Which Queries Trigger an Overview
Not every query gets one, and knowing the pattern saves effort. Overviews appear most reliably where synthesis genuinely adds something:
- Multi-part informational questions — "how do I measure X and connect it to Y" — where the answer spans several sources.
- Definitional and explanatory queries, particularly in emerging categories where no single canonical source exists.
- Comparison and evaluation queries, where a synthesised summary of several viewpoints is more useful than one opinion.
- Process and how-to queries with sequential steps.
They appear less often on navigational queries, on transactional queries with clear commercial intent, and on topics where Google's systems apply extra caution — health, finance, legal and similar. They also appear less on queries where a single authoritative source obviously exists.
The practical read: your definitional and explanatory content is where Overview exposure concentrates, and your bottom-of-funnel commercial pages are largely unaffected. That shapes where to spend the rewriting effort, and it explains why sites with heavy top-of-funnel content see the impression-versus-click divergence most sharply.
The Position Nuance
Being cited is not one outcome. Three distinct positions carry very different value:
- First-cited on the opening claim. The strongest position. The opening sentence of an Overview carries the definition, and the source attached to it is effectively the authority on the topic for that user.
- Cited mid-answer on a supporting point. Useful, lower attention, and often on a narrower sub-question than the one you targeted.
- Present in the source list without an inline attachment. The weakest, and easily mistaken for success in a tool that only records domain presence.
If you are recording Overview visibility manually, record which of the three applies. A month where you moved from position three to position one on five priority questions is a real improvement that a binary present-or-absent flag would show as no change at all.
What to Do About the Click Loss
Some click loss on fully-resolved queries is not recoverable, and pretending otherwise leads to bad decisions. Three responses that work:
- Make the answer complete and the depth visible. The extractable paragraph should resolve the question while making clear that a harder version exists below — a trade-off, a failure mode, a method. Readers click for the part the summary could not carry.
- Move effort toward queries Overviews do not resolve. Comparison, pricing, evaluation and implementation queries retain clicks because the user needs specifics an Overview will not supply.
- Reframe the target. On resolved queries the win is being the cited authority, and the measurable consequence is branded search rather than a session. That is a real outcome; it is simply recorded somewhere else. Instrumenting that connection is what keeps the work fundable.
Frequently Asked Questions
How do you rank in Google AI Overviews?
By being indexed, ranking well for the query, and writing passages that answer specific sub-questions completely and independently. Google states there is no separate optimisation for AI Overviews because they draw on the standard Search index.
Is there special schema for AI Overviews?
No. Standard structured data — Organization, Article, FAQPage, Product — is all that exists. There is no AI-Overview-specific markup.
Why is my page cited in an AI Overview but ranking low organically?
Because Overviews select passages, not pages, and because query fan-out retrieves against related sub-queries. A page that answers one sub-question exceptionally well can be cited without ranking highly for the original query.
Can I opt out of AI Overviews?
Only through snippet controls such as nosnippet, max-snippet and data-nosnippet, which also restrict featured snippets and result descriptions. There is no Overview-specific opt-out.
Do AI Overviews reduce clicks?
They reduce click-through on queries they fully resolve. Impressions typically rise while clicks stay flat. Judge the effect against branded search volume and pipeline rather than click-through rate alone.
How do I track AI Overview visibility?
Search Console does not separate it. Sample your priority questions manually each month in clean sessions, recording whether an Overview appeared and whether you were cited, and pair that with impression-versus-click trends on question-shaped queries.