How to appear in Google AI Overviews, what makes a page eligible

No. Google states plainly that there are no additional requirements and no special optimizations needed to appear in AI Overviews or AI Mode. The documentation adds that you do not need to create machine readable files, AI text files, or markup, and that no dedicated schema.org structured data exists for these surfaces. Three technical conditions make a page eligible, being indexed, being eligible to be shown with a snippet, and meeting the Search technical requirements. Google then adds that meeting every requirement guarantees neither crawling, nor indexing, nor serving. On top of that, the documentation says AI Overviews only show when Google systems decide they add value over classic Search, which is why the very same page can be quoted one week and absent the next. Appearing in an AI Overview stays a probability you improve, never an outcome a tag buys.

That distinction shapes this whole guide. We keep what Google documents officially separate from what the market assumes. Official recommendations are verifiable and dated. GEO hypotheses can be useful, they remain hypotheses until Google confirms them, and we flag them as such.

We are publishing this guide now because the French rollout finally landed. Google switched AI Overviews and AI Mode on for google.fr on 22 July 2026, after two years during which France stayed the last large market on the sidelines. That is what makes the screenshots below possible, and it moves the subject from theory to daily operations for French-speaking sites. The eligibility rules themselves are the same everywhere, so this guide holds whatever market you sell to.

• No tag, no file and no content format forces an appearance in AI Overviews. Any method sold as guaranteed is false.
• A page can be perfectly optimized and never appear, because Google only shows an AI Overview when its systems decide it adds value over classic Search.
• Google explicitly warns against third-party tools that promise rankings or claim to use internal metrics. No third-party tool has access to its ranking or AI systems.
• AI Overviews and AI Mode may use different models, so they can return different answers and different links for the same query.

Google Search Central, AI features and your website | Google, guide to optimizing for generative AI features

AI Overviews appear above the blue links, on a results page that stays a results page. AI Mode is a separate conversational surface, built for questions that need reasoning or complex comparisons. Classic results keep running underneath both. Since 22 July 2026 both surfaces are live in France, the last large market to receive them.

The screenshot below comes from a real search on google.fr. The AI Overview splits its answer into three themed blocks and attaches a clickable source chip to each one, backed by a card panel on the right. Citation surfaces are therefore spread along the text, sentence by sentence, rather than gathered in one link at the bottom.

Google Search Central, AI features and your website | Blog du Moderateur, AI Overviews and AI Mode rollout in France

A page becomes eligible for AI Overviews once it meets three cumulative conditions. It must be indexed in Google Search. It must be eligible to show with a snippet, so it cannot carry nosnippet, max-snippet set to zero, or noindex. It must meet the Search technical requirements, meaning Googlebot is not blocked, the page returns an HTTP 200 status code, and the page has indexable content. Google adds a fourth condition specific to generative surfaces, the site must stay included in the Search generative AI control in Search Console. That control defaults to include, it is inherited from the closest parent property, and an exclusion takes effect within one to two days. It is not used as a ranking signal elsewhere in Search.

Inheritance is where teams get caught. An exclusion set on a domain property propagates to child properties, and nobody remembers it six months later. Before blaming your content, open the setting and read its actual value.

• Confirm the page is indexed with URL Inspection in Search Console, and that Google fetched the version you intend to serve.
• Inspect the HTML Googlebot received and look for robots meta tags. A nosnippet or max-snippet:0 value makes the page ineligible for AI Overviews while leaving it indexed.
• Check that data-nosnippet does not wrap the very passage you want quoted.
• Open Search Console Settings, then Search generative AI, and confirm the Include value, parent property included.
• Verify that robots.txt, the CDN and the firewall let Googlebot through. An infrastructure block produces the same symptoms as a markup mistake.
• Test rendering if your page depends on JavaScript, since content injected after load still has to be crawlable.

Google, Search technical requirements | Search Console, Search generative AI control

Google states one simple and demanding rule, create non-commodity content that is helpful, reliable and people-first. Its documentation contrasts two telling examples. A piece titled "7 Tips for First-Time Homebuyers" is common knowledge and adds no value of its own. A piece titled "Why We Waived the Inspection & Saved Money, A Look Inside the Sewer Line" carries experience nobody else can copy.

The second lever is form. Important text has to exist as text, not only inside an image, a video or a client-rendered component. Headings and sections serve human reading, and that same structure helps systems isolate the passage that answers a precise question.

• Answer the main question in the first lines, then expand. A self-contained passage is easier to quote than a conclusion that depends on three earlier paragraphs.
• Bring a first-hand point of view, a test you ran, a number you measured, a method you actually apply.
• Keep important content as visible text in the served HTML, without depending on client rendering.
• Use headings that restate real questions, and keep paragraphs short.
• Make the page reachable through internal links from your strong pages, otherwise it stays hard to discover.
• Support the text with relevant images and videos, following Google image SEO best practices.
• Reduce duplicate content, which wastes crawl budget and dilutes the signal.

Google, creating helpful, reliable, people-first content | Google, image SEO best practices

Go further. Internal linking | Topical authority | How generative AI is transforming SEO

A retrieval-augmented generation system picks its sources from the pages Google ranking systems consider relevant and up to date. Topical authority is therefore built exactly where it was already built, through a coherent set of pages that covers a subject in depth rather than a single isolated article. Google stresses a point many content strategies ignore, a unique point of view carries more weight than a summary of what exists elsewhere, because its systems review several sources and look for what stands out. Publishing a synthesis of five competing articles produces content the model can already generate on its own. Publishing a test, a dataset, an applied method or a dated field report produces information the model has to come and get from you.

Originality alone is not enough. Google reminds site owners that its anti-spam systems apply to generative features too, and that seeking inauthentic mentions across the web helps less than it seems. What surfaces are real mentions in blogs, videos and forum discussions, the ones your product or your expertise genuinely triggers.

• Pick a scope you can cover better than anyone else, rather than a broad subject treated on the surface.
• Date and source your claims, and correct pages when reality changes.
• Sign your content with an identifiable author and a page that documents their experience.
• Link pages on the same subject together so the coverage reads as one body of work.

Google, guide to optimizing for generative AI features

Google documents two mechanisms behind its generative features. Retrieval-augmented generation, also called grounding, relies on core Search ranking systems to retrieve relevant, up-to-date pages, then reviews the information on those pages to build the answer and the links shown with it. Query fan-out issues a set of concurrent related queries generated by the model. For "how to fix a lawn that is full of weeds", Google gives "best herbicides for lawns", "remove weeds without chemicals" and "how to prevent weeds in lawn" as example fan-out queries.

The temptation is immediate, ship one page per fan-out query. Google closes that door explicitly. Creating separate content for every variation, in particular by targeting fan-out queries, primarily to manipulate rankings or generative responses, falls under its scaled content abuse spam policy. Google adds that the strategy is ineffective long term, because its systems understand page relevance even without an exact match between query and content.

• Read fan-out as a map of sub-questions to cover inside one solid page, not as a list of URLs to generate.
• No public documentation reveals the fan-out queries actually issued for a given query. Any method claiming to reconstruct them produces a hypothesis, not Google data.
• More pages do not make a site more relevant, Google says so on the same documentation page.

Google, guide to optimizing for generative AI features | Google, scaled content abuse policy

Related resources. Prepare your site for AI agents | Track your citations with Spotlight

Google is clear about structured data. It is not required for generative AI search and there is no special schema.org markup to add. The documentation still sets a consistency requirement, your structured data has to match the visible text on the page. Markup that announces a price, a rating or a date absent from the page becomes an inconsistency, and markup otherwise stays useful for classic rich results.

For local businesses and commerce, the documentation names two concrete surfaces. Generative responses can include product listings, product information and information about local businesses. An up-to-date Merchant Center feed and a correct Business Profile therefore feed those answers directly, which makes them a priority for a retailer and a secondary concern for a software vendor.

• Use the Rich Results Test to check syntax, then reread the page to confirm the markup describes what is actually displayed.
• A type missing from the report is not automatically an error, not every schema.org type produces a rich result.
• Treat Merchant Center and Business Profile as sources of truth, wrong opening hours end up inside a generated answer.
• Follow Google image best practices instead of adding decorative visuals unrelated to the text.

Google, establish your business details on Search | Google Search Central, AI features and your website

Related resources. Glossary, schema markup | Analyzer, page by page audit

Google devotes a full section of its guide to dismantling tactics circulating online. The table below restates its wording and turns it into an operational decision. Mind the scope, these answers apply to Google Search. An llms.txt file can still matter for other systems that read it, Google only says it ignores the file and that its presence neither harms nor helps rankings.

Google, guide to optimizing for generative AI features | Google, evaluating third-party SEO advice

Google points to the generative AI performance report in Search Console to measure its own surfaces. That report gives aggregated impressions and clicks, without separating AI Overviews from AI Mode. We documented how it works and where it goes blind in a dedicated guide, which we do not repeat here.

Two limits are worth stating before you open a dashboard. Search Console only covers Google, so no ChatGPT, no Perplexity, no Claude. And a brand mentioned without a clicked link shows up in no traffic report, even though that mention often triggers the branded search that follows.

Search Console, generative AI performance report

Our measurement guides. Search Console generative AI report | Search Console for AI | Measure AI traffic in GA4

This list covers technical eligibility first, then editorial value, in the order we run it on a real page. The first six points decide whether the page can be shown, the last six decide whether it deserves to be quoted over another one.

• The page is indexed and URL Inspection confirms the version served to Googlebot.
• No nosnippet, max-snippet:0 or noindex directive blocks the snippet.
• No data-nosnippet attribute wraps the passage you want quoted.
• The Search generative AI control in Search Console is set to Include, inheritance included.
• robots.txt, CDN and firewall let Googlebot through, and the page returns a 200 status code.
• Important content is present in the served HTML, without depending on client rendering.
• The answer to the main question arrives in the first lines of the page.
• The content brings experience, a test or data nobody else publishes.
• Headings restate real questions and split the subject in a readable way.
• Internal links from your strong pages make the page easy to reach.
• Images and videos serve the point and follow Google best practices.
• Structured data matches the visible text, and your Merchant Center and Business Profile details are current.

Useful tools. Check AI crawler access | Free AI visibility audit | Glossary, AI Overviews

Eligibility for AI Overviews comes down to four technical checks and one editorial demand. The checks take an hour. The editorial demand, publishing what nobody else can publish, takes the rest of the year.

The next step is knowing where you actually stand. Our free audit checks crawler access, page structure and citability signals, and Analyzer then tracks page by page what feeds the answers of AI engines.