What was announced
Today, each AI provider needs its own rule in robots.txt. Google's is called Google-Extended. According to Google, a shared rule that applies to all participating providers will join it. There is a draft at the IETF, and Google uses the same name in its crawler documentation:
User-Agent: __usage_ai-training Disallow: /
Disallow states a preference against training and Allow a preference in favour, path by path as usual. A catch-all rule for every crawler (User-Agent: *) does not count as a statement about training, and without the new rule you have simply expressed no preference.
For now this is a proposal. The draft was submitted by Kevin Kelley of Anthropic, and the IETF has not endorsed it so far. Google's blog post says 2027, its documentation says "early 2027". Google does not name the other providers taking part. It does point to the EU Code of Practice for general-purpose AI models, which recognises robots.txt as a primary method for such machine-readable preferences.
What the rule also switches off at Google
If the new rule says Disallow and Google-Extended appears nowhere in the file, Google plans to treat that as if Google-Extended were disallowed. According to the documentation, that one switch covers two things: training future Gemini models and so-called grounding in Gemini Apps, where the model receives content from Google's index while it answers.
The announcement says: "Websites that specifically allow Google-Extended are eligible to appear in responses and receive clicks via grounding in Gemini Apps."
Our reading: if you use the shared rule only to rule out training, you may also drop out as a source in Gemini answers. Google does not say this in so many words, it follows from the two texts taken together.
Each provider's own rules keep working, so you can decide one by one. Ruling out all participating providers while admitting Google would look like this:
User-Agent: __usage_ai-training Disallow: / User-Agent: Google-Extended Allow: /
That Allow then covers both, so Google may train as well. "No training, but appear in Gemini" cannot be set with Google-Extended as it stands.
What stays the same
According to the documentation, Google-Extended has no effect on regular Google Search: neither on inclusion in Search nor on ranking. AI Overviews and AI Mode have their own switch, the Search generative AI control in Search Console, which applies to the whole site.
Crawling is a separate matter too. The draft says explicitly that the new rule is not a directive for crawling, so a bot can still fetch your pages. To limit that, you still need rules for the individual crawlers.
The Content-Signal line described in our knowledge article is a different proposal. Google's announcement does not mention it, and if you have set it, you can leave it in place.
What you should do
Nothing for now. Google does not support the rule yet, and as long as it is not in your robots.txt, you express no preference with it in 2027 either.
We would advise a café, a tradesperson or a medical practice against setting it later as well. What counts for them is being named in answers. It can look different for content someone earns a living from, such as specialist articles, photos or courses. There it is worth looking at the finished state in 2027 and then deciding provider by provider.
If Google-Extended is already in your file, the new rule changes nothing about that, because at Google the explicit statement takes precedence. And if a service helps maintain your robots.txt, Cloudflare for example, take a look at what it says in 2027.