What is Google-Extended?

A robots.txt control token that tells Google not to use your content for training Gemini models and Vertex AI generative APIs. It's not a separate crawler — regular Googlebot does the fetching and checks the token.

Does blocking Google-Extended affect my Google rankings?

No. Google states explicitly that Google-Extended has no impact on Search indexing or ranking. Googlebot governs Search; Google-Extended only governs generative model training.

Does blocking Google-Extended remove me from AI Overviews?

No. AI Overviews are a Search feature built from the Search index, not controlled by Google-Extended. The only levers over AI Overviews are snippet controls like nosnippet and max-snippet, which also affect regular search snippets.

Should I block Google-Extended?

Block it if you object to your content training Gemini — it costs nothing in Search visibility. Keep in mind models trained without your content are less likely to know or mention your brand.

Google-Extended explained: what it controls (and what it doesn't)

Google-Extended is not a crawler — it's a robots.txt token that Google's existing crawlers check to decide whether your content may train Gemini models. Blocking it has zero effect on Google Search rankings and zero effect on AI Overviews. Here's exactly what it does and doesn't control, because almost every summary online gets part of this wrong.

What Google-Extended actually is

When Google introduced Google-Extended in September 2023, it did something unusual: instead of launching another crawler, it created a product token — a name you can address in robots.txt that changes how already-crawled content may be used.

User-agent: Google-Extended
Disallow: /

You will never see "Google-Extended" in your server logs. Regular Googlebot fetches your pages exactly as before; the token is consulted afterward, as a usage permission. That's why our bot directory classifies it differently from true crawlers like GPTBot.

What blocking it does

Disallowing Google-Extended opts your content out of:

Training Gemini models — your pages won't be used to improve future Gemini versions.
Grounding in Vertex AI — Google Cloud's generative APIs won't use your content as source material for their answers.

That's the whole list. It is Google's equivalent of blocking GPTBot or ClaudeBot: a training opt-out, cleanly separated from search.

What blocking it does NOT do

This is where the confusion lives, so one claim at a time:

It does not affect Google Search rankings. Google has stated this directly and repeatedly. Indexing and ranking are governed by Googlebot, which you're not touching.
It does not remove you from AI Overviews. AI Overviews are generated from the Search index as a Search feature. Google treats them like featured snippets, not like Gemini training. Blocking Google-Extended while ranking well means you can still be summarized and cited in AI Overviews.
It does not stop Gemini from browsing to your page live. If a Gemini user asks about your URL and the product fetches it in real time via Search infrastructure, that's a different pathway.
It does not retroactively remove your content from models Google has already trained.

So how DO you control AI Overviews?

There is no dedicated AI Overviews opt-out. Your levers are the classic snippet controls, which apply to AI Overviews the same way they apply to featured snippets:

<meta name="robots" content="nosnippet"> — nothing from the page may be shown or used in AI Overviews. Also kills your regular search snippet.
max-snippet:[n] — caps how many characters can be used. Less nuclear, still affects normal snippets.
Blocking Googlebot entirely — removes you from Search altogether, which is almost never what anyone wants.

In practice, most sites should aim the other way: being cited in AI Overviews is distribution, and our AI Overviews structure checklist covers how to win those citations rather than hide from them.

Should you block Google-Extended?

The trade-off is cleaner than for most AI crawlers precisely because search is unaffected:

Block it if you object to your content training Google's models — original research, creative work, paid content. You give up nothing in Search. This is the rare free opt-out.

Allow it if you want future Gemini models to "know" your content and brand. Models trained on your material are more likely to mention you unprompted — a fuzzy but real benefit, especially for brands whose category answers you want to shape.

Because the cost side is ~zero, Google-Extended blocking is common even among sites that allow every search-type AI agent. If that's your policy, our robots.txt Generator for AI Bots can produce exactly that split — block training tokens (Google-Extended, Applebot-Extended, GPTBot, ClaudeBot), allow search agents — and the AI Crawler Access Checker verifies what your live file actually says.

The pattern to remember

Google-Extended and Applebot-Extended represent a model the industry may standardize on: fetching and usage as separate permissions. One crawler gathers pages; named tokens govern what each product may do with them. Whenever you read "block X to stay out of AI," ask which layer X controls — access, or usage. Google-Extended is usage-only, training-only. Everything else you've heard about it is probably one of the four myths above.