How Google-Extended works
Google-Extended is not a crawler. It is a token you can name in robots.txt to tell Google whether content already fetched by Googlebot may be used to help improve Gemini apps and the grounded answers in Google’s AI developer products. No separate bot arrives at your server under that name, and no line appears in your access log for it.
That is the entire mechanism, and it explains why the control behaves oddly compared with the crawler tokens people are used to. You are not turning off a visit. You are declining one particular use of something the ordinary search crawler has already collected.
Why Google-Extended matters
It exists so a publisher can stay in Google Search while opting out of a single downstream use. Before it, the only lever available was blocking Googlebot, which meant leaving the index altogether — a price almost nobody would pay. Separating the two was a genuine improvement for anyone with a licensing position to protect.
For most small businesses the honest answer is that it changes very little. If your aim is to be found, quoted and recommended, you have no reason to reach for it. Its value belongs to publishers whose content is itself the product.
Common mistakes with Google-Extended
The big one is believing it controls AI Overviews. It does not. AI Overviews are a feature of Google Search, so they follow the ordinary search controls — whether Googlebot can crawl the page, whether it is indexed, and the snippet directives you set with a meta robots tag. Disallowing Google-Extended in the hope of disappearing from AI Overviews is a wasted change.
The second is fearing a penalty. Using it has no effect on crawling, indexing or ranking in Google Search, because it has no relationship with Googlebot at all.
The third is confusing it with other companies’ crawlers. A rule for Google-Extended says nothing about GPTBot or any other operator’s agent. Every vendor’s control is separate, and there is no single switch for all of them.
What to do about it
Ask one question: is my content itself the thing I sell? A news archive, a paid research library or a licensed database has a real reason to decline. A services business that wants to be recommended does not, and using the control gains it nothing at all.
If you do want it, add a user-agent group named Google-Extended to your robots.txt and disallow the paths concerned, remembering that a group takes only the rules written inside it. Then keep the snippet controls separate in your mind. Those are what govern how much of your page can be displayed in Search and its AI features, and those are the ones to review if display, rather than model training, is your actual concern.