What does Google-Extended block? Gemini training and grounding, not Search or AI Overviews

Google-Extended is a robots.txt token. A Disallow for it tells Google not to use your pages to train future Gemini models or to ground answers in Gemini Apps. It does not affect Google Search and it does not switch off AI Overviews. For those, Google points to rules for Googlebot and to a separate Search generative AI control in Search Console.

This page rests on Google's own documentation, read on 1 November 2026, not on third-party summaries.

What does Google say Google-Extended controls?

Google calls it a standalone product token that publishers use to manage whether crawled content “may be used for training future generations of Gemini models”. Those models power Gemini Apps and the Vertex AI API for Gemini. The token also covers grounding in Gemini Apps and in Grounding with Google Search on Vertex AI. Google defines grounding as giving the model content from the Search index at prompt time.

Three facts follow from Google's text.

• It has no crawler of its own. Google says Google-Extended has no separate user agent string. Crawling uses existing Google user agent strings, so no request in your access log carries the token. A Disallow for it changes nothing you can see there. Find AI crawlers in your server logs covers what a log can show.
• It does not touch Search. Google says the token does not affect a site's inclusion in Google Search and is not a ranking signal.
• It looks forward. The wording is training of “future generations”. The page says nothing about content a model has already seen.

Google added the token on 28 September 2023. Its Search Central changelog lists the launch and later rewordings, one of which states that the token does not affect Google Search.

The group takes the same path rules as any other:

User-agent: Google-Extended
Disallow: /

Does Google-Extended affect AI Overviews or AI Mode?

No. Three Google pages put AI Overviews and AI Mode under Search, not under Google-Extended.

1. The common crawlers page. It names Gemini Apps and Vertex AI as the products Google-Extended affects. It does not mention AI Overviews or AI Mode. Its Googlebot entry says crawling preferences for Googlebot affect Google Search, “including Discover and all Google Search features”.
2. The AI features page. Google writes that AI is built into Search, so robots.txt rules for Googlebot are the control for crawling. A page needs to be indexed and eligible for a snippet to appear as a supporting link, with no further technical requirements. To limit what Search shows, it lists nosnippet, data-nosnippet, max-snippet and noindex. It mentions Google-Extended only for “AI training and grounding in some of Google's other systems”.
3. The Search Console control page. It lists AI Overviews, AI Mode and generative features in Google Discover as what its own control manages. It adds that the control does not affect AI training, and points to Google-Extended for limiting the training of the models behind those features.

The last point is the one to keep. Google names Google-Extended as the way to limit training of the models behind AI Overviews. The token does not decide whether your pages appear in them.

How do you opt out of AI Overviews?

Use the Search generative AI control. Google's help page says it rolled the control out to all websites worldwide on 31 August 2026. You find it in Search Console under Settings, then Search generative AI.

It has 3 values: include (the default), exclude, and inherit from a parent property. Excluding keeps your links and content out of those features, including as grounding. Google says you then get no traffic or impressions from them, and that content from other sites still appears.

Google adds 3 caveats:

• The control is not a ranking or inclusion signal for the rest of Search.
• It does not override your choices for services such as Merchant Center and Google Ads.
• A change generally takes a few days to show.

The help page names the Generative AI performance report for measuring the effect before and after. Google's page on controlling what you share, which the changelog says it updated on 21 October 2026, lists the control as a site-wide option. The help page names noindex as the way to block a page from Google Search completely.

The AI features page, read the same day, does not mention the control. The help page dates the rollout to 31 August 2026.

What does a Disallow for Google-Extended cost?

It costs grounding in Gemini Apps. On 30 October 2026 Google wrote that sites which specifically allow Google-Extended can appear in responses and receive clicks via grounding in Gemini Apps. A Disallow gives that up along with training. The AI crawler blocking guide shows the file that blocks training and keeps search.

It costs nothing in Search, by Google's account: the token does not affect inclusion in Google Search.

A group that names a token replaces the * group for that token. Say your file has a * group full of Disallow lines and a bare Allow: / group for Google-Extended. The * rules then do not apply to the token. AI crawler user agents explains how RFC 9309 picks the group.

Is a Content-Signal line enough?

Google's pages do not mention Content-Signal. They name a Disallow for the token as the control. A Content-Signal: ai-train=no line is a statement of intent, and Content-Signal in robots.txt covers who reads it.

Google's 30 October post announces a shared training control for robots.txt, based on the IETF AI Preferences work. When a site uses it to opt out of training and does not name Google-Extended, Google plans to read that as a Disallow for Google-Extended. It aims to support the control in 2027. It is a plan, and no robots.txt line works that way today.

What do real sites do?

In the scan records of 20 October 2026, 1 of the 5 exemplars names Google-Extended. Cloudflare gives it its own group with an allow rule. Vercel, Resend, MDN and GitHub leave it on the * group.

How do you check what your robots.txt says to Google-Extended?

Read the group that applies to the token:

curl -sS https://example.com/robots.txt | grep -i -A3 '^user-agent: google-extended'

No output means robots.txt has no group for the token, so under RFC 9309 it follows *. The AI crawler checker resolves the governing group for Google-Extended and 17 other tokens. The AI robots.txt generator writes a group for it from a tick box.

None of the Google pages read for this article documents a report or log line that shows the token at work. Bing, Copilot and noarchive covers Bing's controls, and the AI grounding directives check reads them on your homepage.

To see how your file addresses all 18 AI crawler tokens, scan your site.