Google AI Overviews optimisation: what Google documents, what the studies show, and a page-level checklist.

The short answer.

Getting into AI Overviews is mostly ordinary search work done thoroughly, plus one new habit: covering the follow-up questions your buyer would ask next. There is no AI Overview sign-up, special markup or AI file to add.

That does not mean nothing changes. Google builds these answers with a technique it calls query fan-out, running several related searches behind the one you typed. In Ahrefs’ March 2026 data, only 37.9% of AI Overview citations also ranked in the top 10 for the original query. A page can rank first and not be cited, while a page on page three is cited for answering a sub-question well.

So the work has three layers, in this order:

1. Eligibility. Indexed, crawlable, snippet-eligible, not excluded. Fully documented.
2. Retrieval. Rank for the question and its sub-questions. Principle documented, detail not.
3. Selection. Be the clearest, most original source on the point.

The first layer is quick to check. The second and third are content and evidence work, which a service for getting named in AI answers covers alongside the technical checks.

What AI Overviews are, briefly.

AI Overviews are the AI-written summaries that appear above or among Google’s normal results, with links to the pages used. Google says they “help people get to the gist of a complicated topic or question more quickly” and are “only shown when our systems determine that it is additive to classic Search.”

Three dates set the context. On 14 May 2024 Google said AI Overviews would begin “rolling out to everyone in the U.S.” On 27 January 2026 it said it was “making Gemini 3 the new default model for AI Overviews globally.” On 3 June 2026 Google announced a reporting view and an opt-out control for site owners in Search Console, first for a subset of sites; Google says both reached all websites worldwide by 31 August 2026.

In its 3 June 2026 announcement, Google said AI Overviews have “over 2.5 billion monthly active users”. That is Google’s own figure, not an independent measurement.

How this differs from AI Mode

AI Mode is the separate, conversational tab. Google says the two “may use different models and techniques, so the set of responses and links they show will vary.” Our guide to Google AI Mode for B2B covers AI Mode and how the two differ; this article covers the Overview on the normal results page.

Google’s eligibility rules.

Google’s documentation sets four conditions, all checkable in an afternoon. If one fails, nothing else in this guide matters for that page.

Google adds that eligibility is not the same as being shown: indexing and serving are not assured for any page.

The CDN line is the one most often missed. Our play on checking AI crawler access has a short test, and the AI crawler log file check shows how to confirm Googlebot is actually fetching your key pages.

How Google says sources are chosen.

Google says its AI features are “rooted in our core Search ranking and quality systems,” and that they use query fan-out to find supporting pages. It does not publish how a single link is picked from the candidates, and anyone who claims to know the exact formula is guessing.

What is documented: both features “may use a ‘query fan-out’ technique,” issuing “multiple related searches across subtopics and data sources” to “identify more supporting web pages” while the response is written. Google’s generative AI guide sums up the consequence: this work is “still SEO.”

Fan-out is why the query you track is not the only one that matters. Google does not show which sub-queries it ran, so infer them from what the answer covers.

What Google says you can ignore.

Google’s generative AI guide also lists tactics that do not help in its AI features, including llms.txt and other AI text files, special AI schema, chunking pages into tiny pieces, writing in a special style for AI and seeking inauthentic mentions. Our guide to Google AI Mode sets out the tactics Google says you can ignore, with what to do instead, so we do not repeat the list here. If a supplier sells any of them as AI Overview work, ask for their evidence.

On llms.txt, our guide to llms.txt and AI crawlers sets out what each AI company has documented. Monitoring tools can track which pages are cited, but Google says “No third-party tool has access to our internal ranking or AI systems”; our comparison of AI visibility tools explains what each kind measures. What Google does recommend, from crawlable text to “unique, non-commodity content” (John Mueller, Search Central blog, 21 May 2025), is covered in what Google says about pages it quotes.

What the studies show about rankings.

Independent studies agree that a top-10 ranking helps but no longer decides who is cited. Since mid-2025 the measured overlap has fallen, and recent studies find most citations come from outside the top 10. Each figure below is the company’s own published data, measuring different things.

Ahrefs’ explanation for its own drop is that AI Overviews “are relying less on the direct search results and more on the sources showing up in fan out query SERPs,” possibly because Gemini 3 expands queries more broadly. It also improved its parsing between studies, so part of the fall may be measurement.

For a B2B site: ranking for your main query is still the best starting point, but a page that targets only that query leaves most citation chances to pages that answer the sub-questions.

Rank position and citation rate.

Higher rankings raise the chance of a citation, but even the top result is usually not cited. In Conductor’s US data for 15 June to 15 August 2026, the page in position 1 was cited in an AI Overview 24.9% of the time and the page in position 10 just 10.2% of the time.

In Conductor’s words: “Three out of four times, the page Google ranks first isn’t the one its own AI Overviews reach for.” Its authors add that they “didn’t set out to prove ranking causes citation.”

So the fix is rarely to push the ranking higher. It is usually to read what the Overview says, which sub-questions it answers, and whether your page answers them in plain text.

A quick way to see your own gap

Take ten real buyer questions. For each, record whether an Overview appears, which pages it links and where your page ranks. Sort them into four groups: ranking and cited, ranking but not cited, cited but not ranking, and neither. “Ranking but not cited” is where page-level work pays back fastest. Our guide to checking AI search visibility gives a fuller method across all the engines.

Who else gets cited.

AI Overviews cite many sources per answer, and a large share go to forums and video rather than company websites.

Seer Interactive’s study, published 28 May 2026 from a 7–13 May 2026 snapshot of US English queries, found AI Overviews cited a mean of 11.4 sources per query (median 11). Reddit held 20.4% of first-citation slots in its sample. In Ahrefs’ March 2026 data, YouTube made up 18.2% of the cited pages that did not rank in the top 100 for the same keyword; those YouTube URLs alone were 5.6% of all AI Overview URLs in the dataset.

Seer also found AI Overviews on 85% of its 8,500 keywords but only 16.24% once weighted by search volume: they appear far more on specific, lower-volume questions than on broad head terms. Very long “ultimate guides” took only 4.4% of definitional slots. Seer stresses this is one snapshot and behaviour shifts over time.

For a B2B company, a forum thread or video can beat a vague service page. Third-party sources matter too: our guide to the off-site sources AI engines trust covers that side, and our own study of where AI answers get their sources includes Google’s AI Overview alongside ChatGPT and Gemini.

The controls, and what each one does.

Several Google controls affect AI Overviews, and each does something different. Mixing them up is how sites remove themselves by accident.

The Search generative AI control is the newest; Google says it “doesn’t affect AI training” and reached all websites worldwide as of 31 August 2026. The cost is stated plainly: “Sites that opt out will not receive traffic or impressions from our generative AI features.”

What a B2B site should check

Confirm the Search Console setting is on include. Then search your templates for nosnippet and max-snippet: a short max-snippet set years ago for snippet length now also limits what AI Overviews can use. To keep one section out, data-nosnippet on that block is the narrow tool.

The page-level checklist.

Use this on one page at a time, starting with the page for your most valuable buyer question.

Eligibility (Google-documented)

• ✓URL Inspection shows the page as indexed, with the canonical you expect.
• ✓No noindex, nosnippet or short max-snippet in the page head or HTTP headers.
• ✓Googlebot is allowed in robots.txt and is not challenged by the CDN or firewall.
• ✓Search Console’s Search generative AI setting is on include.
• ✓Structured data describes only what is visible on the page (our play on schema that matches the page has a test).

Retrieval (fan-out coverage)

• ✓The page answers the main question in its first two sentences, in text.
• ✓It also answers the four or five follow-up questions a buyer would ask next, each under its own heading.
• ✓Related pages on your site link to it with descriptive anchors.
• ✓Key text is in the HTML the server sends, not only loaded later by script (see the play on server-rendered content).

Selection (being the best source on the point)

• ✓It says something not on ten other pages: your method, data or worked examples.
• ✓Every number has a date and a named source in the same sentence.
• ✓It loads quickly on a phone, with a table, diagram or video where that helps.

How to write the passages themselves is covered in our guide to the page patterns AI engines quote, so we have not repeated it here.

Measuring it in Search Console.

Search Console now shows how often your pages appear in Google’s AI features: impressions, not reasons. Google introduced the Search generative AI performance reports on 3 June 2026, covering AI Overviews and AI Mode, and says that as of 31 August 2026 they reached “all websites worldwide.” How to read them next to the normal Performance report, without double counting, is covered in our review of what to measure as AI Overviews cut clicks.

A simple monthly routine

1. Export the report’s Pages view and compare it with the pages you most want cited. The gap is your work list.
2. For each gap page, read the current Overview for the buyer question: what does it say that your page does not?
3. Track enquiries from those pages in GA4 or your CRM; impressions alone do not pay the bills.

Search Console only covers Google. To see whether ChatGPT and Perplexity name you for the same questions, you can get a free AI visibility report that runs a sample of buyer questions across the main engines.

The AI Overview page self-check.

Pick one page and one buyer question, tick what is true today and choose where the page ranks. The first four items are gates: if one fails, the page cannot be used. Nothing you enter is sent anywhere.

The weights are our judgement from the sources above, not a Google score; the gates are each documented by Google.

Pages that pass the gates but score low usually answer only the head question and say what every competitor says. Stronger service content and original evidence for those gaps are part of the scope of our GEO and AEO work for B2B companies.

Questions owners ask.

How do I get my page into Google AI Overviews?

Make sure it is indexed, snippet-eligible, crawlable and not excluded in Search Console. Then rank for the buyer question and its follow-ups with clear, original answers in text. Google says no special markup or AI file is needed.

Does my page need to rank in the top 10 to be cited?

No, though it helps. In Ahrefs’ March 2026 data only 37.9% of AI Overview citations ranked in the top 10 for the same query, and Conductor’s June to August 2026 data put the figure at 43%. Pages that answer related sub-questions are often cited instead.

Is there schema markup for AI Overviews?

No. Google says there is no special schema.org markup to add for its generative AI features. Normal structured data is still worth keeping if it matches the visible page.

Can I stop Google using my site in AI Overviews?

Yes. The Search generative AI control in Search Console excludes your site from AI Overviews, AI Mode and Discover’s generative features, and Google says it does not affect your other rankings. Page-level nosnippet also stops content being used as a direct input.

Does blocking Google-Extended remove me from AI Overviews?

No. Google says Google-Extended does not affect inclusion in Google Search; it limits AI training and grounding in some other Google systems. The Search generative AI setting and snippet controls govern AI Overviews.

Sources: Google, AI features and your website; Google, generative AI features guide; Google, robots meta tag; Search Console Help, Search generative AI control; Search Central blog, 3 Jun 2026; Google, 3 Jun 2026; Search Central blog, 21 May 2025; Google, 14 May 2024; Google, 27 Jan 2026; Ahrefs, 21 Jul 2025; Ahrefs, 2 Mar 2026; Conductor, updated 25 Sep 2026; BrightEdge, 18 Sep 2025; Seer Interactive, 28 May 2026. Accessed 13 November 2026.