The verdict
Blocking Google-Extended stops Gemini training and some grounding outside Search. Google says it does not affect inclusion or ranking in Search, so it will not take you out of AI Overviews. Those follow the ordinary snippet and index controls, which also change your normal results. For most local and ecommerce businesses, I would block nothing.
“Can you opt us out of Google’s AI?” is a reasonable thing for a client to ask, and it does not have one answer. There are three different controls, they do three different things, and the one most people reach for does not do what they think.
Google-Extended controls training, not Search
Google-Extended is a token you can name in robots.txt. Google’s crawler documentation says it manages whether content may be used for training future Gemini models, and then says the part that matters: “Google-Extended does not impact a site’s inclusion in Google Search nor is it used as a ranking signal in Google Search.”
So blocking it keeps your pages out of Gemini training and out of grounding in some of Google’s other systems. It does not keep them out of AI Overviews or AI Mode, because those are part of Search.
What does control AI Overviews
The same controls that govern every other result. Google’s AI features documentation lists them: nosnippet, data-nosnippet, max-snippet and noindex.
The catch is that none of them is AI-only. A nosnippet on a page removes its text snippet from the ordinary results too. noindex removes the page from Search altogether. There is no setting that keeps you in the blue links and out of the AI answer above them. Anyone who promises one is describing something Google does not offer.
And llms.txt does nothing for Google
The other file people ask about is llms.txt, a proposed way of telling AI systems what is on a site. Google’s guide to generative AI in Search is clear that “You don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search.” It neither helps nor hurts. It is simply not read.
What I would actually do
For most of my clients, nothing. A clinic, a trades business, a store. These are businesses that want to be found and quoted, and blocking AI systems removes them from the answer without removing the question. Somebody else gets named instead.
Most sites agree, or have not thought about it. HTTP Archive’s 2025 Web Almanac found gptbot named in 4.2% of robots.txt files and claudebot in 3.4%. Roughly one site in twenty-five has a view either way.
The businesses where it deserves a real conversation are the ones whose product is the content: publishers, course providers, anyone whose archive is the thing being sold. That is a commercial decision about licensing, not an SEO one, and I say so rather than pretend I can optimise it.
Blocking is one more AI decision people are sold as though it were optimisation. The wider picture is in what Google says about AI Overviews and the Search index.
The short version
- Google-Extended is about Gemini training. Google says it has no effect on Search inclusion or ranking.
- AI Overviews follow normal Search controls. nosnippet, data-nosnippet, max-snippet, noindex.
- None of those is AI-only. They change your ordinary results too.
- llms.txt is not read by Google Search.
- Most businesses should block nothing. Being quoted is the point.
- Content businesses face a licensing decision. Not an SEO one.
If a site should be ranking and it isn’t, that’s the work I do. If you are deciding what to allow or block and want it thought through for your business, that’s part of SEO consulting.
