Glossary
Google-Extended
In short: Google-Extended is a robots.txt control token that lets site owners opt out of their content being used to train Gemini models without affecting Google Search or AI Overviews.
Google-Extended is not a crawler. It is a token you can name in robots.txt to tell Google whether pages fetched by Googlebot may be used to train and ground Gemini's models. There is no separate bot with that name and no separate request in your logs; Googlebot fetches the page as usual and Google applies your Google-Extended rule afterwards.
The important consequence is what it does not control. Disallowing Google-Extended has no effect on Search rankings, on AI Overviews or on AI Mode. Those are all powered by the normal Search index and you cannot opt out of them without blocking Googlebot and leaving Search entirely. The token only affects training of standalone Gemini models and some grounding of the Gemini app.
A typical rule to opt out of training while keeping everything else looks like this:
User-agent: Google-Extended
Disallow: /
For most businesses the better choice is to leave it allowed, since being represented accurately in model training reduces the chance of being confused with a similarly named company.
Seoptist's audit reports your Google-Extended rule alongside those for GPTBot, ClaudeBot, PerplexityBot and OAI-SearchBot, and explains which engines each rule does and does not affect.