Skip to content
siterank.info
Sign in Start free

SEO

Google Search and AI Experiences: Do You Need Special GEO Markup?

Google says there are no additional requirements or special optimisations for its AI features. Here is what that means, and which controls actually exist.

By ankitkumarvig@gmail.com Published September 9, 2026 Last reviewed September 9, 2026 Sources verified September 9, 2026 9 min read

A large amount of GEO advice sold in 2026 amounts to inventing markup that Google has never asked for. Google's own documentation on AI features says the opposite of what that advice implies, in one sentence: there are no additional requirements to appear in AI Overviews or AI Mode, and no other special optimisations are necessary. This article works through what Google actually documents — the eligibility rule, the controls you genuinely have, and what Google-Extended does and does not do — and separates it from the practices that are unconfirmed guesswork.

What Google actually says

Google Search Central's page on AI features states that AI Overviews and AI Mode surface links to supporting web pages, and that to be eligible a page must be indexed and eligible to be shown in Google Search with a snippet. It then says plainly that there are no additional requirements to appear, nor other special optimisations necessary. The page is explicit that no new machine-readable file, no AI-specific text block, and no special Schema.org markup is required.

That is OFFICIAL PROVIDER GUIDANCE, and it is the single most load-bearing fact in this article. Any product, consultant or article that tells you to add a proprietary "AI markup" block, an "AI-readable" meta tag, or a schema type invented for answer engines is contradicting a first-party statement without a source. Ask for the source.

The same page describes the mechanism Google calls query fan-out: for one prompt, the system issues multiple related searches across subtopics, which lets it surface a wider and more diverse set of supporting links than a single ranked list would. That is a genuine, documented difference in how pages get selected, and it has a real implication — but the implication is about content coverage, not markup. If your site answers only the headline question and none of the adjacent ones, fewer of the fanned-out searches can land on you.

The eligibility chain

Requirement Where it comes from Tier
Page is crawlable by Googlebot robots.txt, RFC 9309 ESTABLISHED STANDARD
Page is indexed Google Search indexing OFFICIAL PROVIDER GUIDANCE
Page is eligible to be shown with a snippet Google's stated AI-features eligibility rule OFFICIAL PROVIDER GUIDANCE
Anything beyond that Not documented —

Every link in that chain is ordinary technical SEO. A noindex page cannot be a supporting link. A page carrying nosnippet cannot contribute text. A page blocked in robots.txt is not indexed on its own content at all. The SEO audit checks exactly this chain, which is why the honest sequence is: fix indexability first, then think about retrievability.

The controls that actually exist

Google documents four mechanisms that a site owner can use to constrain how its content appears in Search previews, and the AI features page states these are the same controls that apply to its AI experiences. All four are OFFICIAL PROVIDER GUIDANCE, and three of them are long-settled Search directives rather than anything new.

noindex

Removes the page from Search results entirely, and therefore from AI-feature eligibility, since eligibility requires indexing.

<meta name="robots" content="noindex">

Or as a header, which is the right choice for PDFs and other non-HTML resources:

X-Robots-Tag: noindex

This is the blunt instrument. It costs you conventional Search visibility too. Do not reach for it to control AI features specifically.

nosnippet

Suppresses the text snippet and video preview for the page. Google's robots meta tag documentation states directly that this also prevents the content from being used for AI Overviews and AI Mode.

<meta name="robots" content="nosnippet">

This is the closest thing to a page-level "do not use my text in a generated answer" control that Google documents, and it comes at a cost: you lose your ordinary Search snippet as well. There is no documented way to keep one and drop the other.

data-nosnippet

A per-element control, valid only on span, div and section elements. It excludes a fragment from snippets without removing the whole page.

<p>Our support hours are 9am to 5pm.
  <span data-nosnippet>Internal note: escalate to tier 2 after 3pm.</span>
</p>

This is the surgical option, and the one most sites should reach for when there is a specific passage — a boilerplate disclaimer, a paywall teaser, a legal notice — that keeps getting pulled into previews.

max-snippet

Caps the number of characters Google may use as a text snippet. 0 suppresses the snippet; -1 lets Google choose.

<meta name="robots" content="max-snippet:120">

A word of caution: capping snippet length is a preview-presentation control that predates AI features by years. Treating a low max-snippet value as a GEO tactic is HYPOTHESIS — nobody has published a mechanism by which restricting how much of your page Google may quote makes it more likely to be cited.

How the rules combine

Google's robots meta tag documentation states that when rules conflict, the more restrictive rule applies. Comma-separated rules combine, and X-Robots-Tag supports the same vocabulary as the meta tag, with optional user-agent targeting:

X-Robots-Tag: googlebot: noindex, nofollow

Two directives worth knowing are effectively dead: noarchive is no longer used because the cached-link feature no longer exists, and nocache is not used by Google Search. Removing them from a legacy template is harmless housekeeping.

Google-Extended: what it is and what it is not

Google-Extended is the most misunderstood item in this area, largely because people assume it is an AI-Overviews opt-out. It is not.

Google's common crawlers documentation describes Google-Extended as a robots.txt user-agent token — not a separate crawler with its own fetches. Crawling is done with the existing Google user agent strings; the token exists purely in a control capacity. What it controls is whether content Google crawls from your site may be used to train future generations of Gemini models, and for grounding in Gemini Apps and Vertex AI.

The documentation then states the part that matters most: Google-Extended does not impact a site's inclusion in Google Search, and is not used as a ranking signal in Google Search.

Question Answer Tier
Does it block Googlebot? No — it is a control token, not a crawler OFFICIAL PROVIDER GUIDANCE
Does it remove you from AI Overviews or AI Mode? Not documented as doing so OFFICIAL PROVIDER GUIDANCE
Does it affect Search ranking or inclusion? Explicitly no OFFICIAL PROVIDER GUIDANCE
Does it govern Gemini model training and Gemini/Vertex grounding? Yes OFFICIAL PROVIDER GUIDANCE

So the practical answer to "how do I opt out of AI Overviews without hurting Search?" is: you cannot, cleanly. nosnippet and data-nosnippet remove text from previews generally, including AI features. Google-Extended addresses a different product family. Anyone selling you a clean separation is describing a control Google has not published.

A Google-Extended block is written like any other robots.txt group:

User-agent: Google-Extended
Disallow: /

Whether you should write it is a business decision about Gemini training and grounding, not an SEO decision. Should you block AI training crawlers? works through that trade-off properly.

Note also GoogleOther and Google-CloudVertexBot in the same crawler list. They are distinct tokens with distinct purposes, and a robots.txt group written for one does not cover the others. Our AI crawler policy tool inspects your robots.txt against the current registry rather than assuming a single "Google AI" agent exists.

What not to assume

Do not assume structured data is an AI-features lever. Google's AI features page says no special Schema.org markup is necessary. Schema is a well-documented part of Search and an ESTABLISHED STANDARD way to describe entities on a page — it is simply not documented as an AI-citation mechanism. Structured data for AI and search covers where it genuinely earns its place.

Do not assume being cited is a rank. An AI Overview supporting link is not a position. There is no ordered list to be third in. Language like "we rank #3 in AI Overviews" describes a thing that does not exist, and it will lead you to optimise for a number nobody publishes.

Do not assume an appearance is stable. Generated surfaces vary between runs for reasons that have nothing to do with your page. If you want to say anything about your presence in AI answers, you need repeated observation with an interval around it, not a screenshot. That is what /methodology/ exists to explain.

Do not assume Google-Extended is a privacy or copyright control. It expresses a preference to Google about specific Google products. It has no effect on any other company's crawler, and it is not enforcement.

Do not assume "AI content" is penalised or rewarded. Google's position is about content quality and usefulness, not production method. Nothing in the AI features documentation changes the spam and quality guidance that already applies.

What to actually do

Everything below is either an established standard or first-party guidance. Nothing here is invented for AI.

  1. Confirm indexability of the pages you care about. Canonical correct, no accidental noindex, no robots.txt block on the content itself. ESTABLISHED STANDARD / OFFICIAL PROVIDER GUIDANCE.
  2. Confirm snippet eligibility. A site-wide nosnippet or an aggressive max-snippet from a legacy template silently disqualifies you from the AI-features eligibility rule. OFFICIAL PROVIDER GUIDANCE.
  3. Use data-nosnippet for specific passages you do not want quoted, rather than page-level nosnippet. OFFICIAL PROVIDER GUIDANCE.
  4. Decide your Google-Extended position deliberately, understanding it governs Gemini training and grounding, not Search. OFFICIAL PROVIDER GUIDANCE.
  5. Cover the adjacent questions, not just the headline one. Query fan-out is documented; broad, genuine subtopic coverage is the honest response to it. The practice itself is good content strategy — treating it as a guaranteed route into AI Overviews is EMERGING PRACTICE at best.
  6. Write passages that stand alone. A retrieval system selects fragments. Self-contained sections with clear headings are plausibly easier to select. EMERGING PRACTICE — plausible mechanism, no provider confirmation. See passage-level content structure.
  7. Keep entity identity consistent site-wide. Same organisation name, same URL, same About page. EMERGING PRACTICE for AI retrieval; ESTABLISHED STANDARD as ordinary Schema.org practice.

Notice what is absent: no AI-specific file, no new tag, no markup type. That absence is the finding.

Key takeaways

  • Google states there are no additional requirements or special optimisations for AI Overviews and AI Mode. Eligibility is: indexed, and eligible for a snippet.
  • The real controls are noindex, nosnippet, data-nosnippet and max-snippet. nosnippet is documented as also preventing use in AI Overviews and AI Mode, but it costs you your Search snippet too.
  • Google-Extended is a robots.txt control token for Gemini training and Gemini/Vertex grounding. Google states explicitly that it does not affect Search inclusion or ranking. It is not an AI Overviews opt-out.
  • Query fan-out is documented and argues for breadth of genuine subtopic coverage — not for markup.
  • Any recommendation to add "GEO markup" should be met with a request for the first-party source. There isn't one.

Official sources & further reading