On 30 October 2026, Google said it plans to honour a single, shared robots.txt rule that tells AI companies not to train on your content. Microsoft said the same for Bing, and OpenAI said it would follow. For bloggers, that sounds like the long-awaited "one switch" for AI training. It is closer than anything we have had, but it isn't live yet, it isn't a standard yet, and opting out has a trade-off most posts about it skip. Here is what the rule is, what it does and doesn't do, and how I'd decide whether to use it.
The rule in one paragraph
The shared signal comes from an Internet-Draft titled "Machine-Readable Training Preference Signal", published on 30 October 2026. It uses a special policy token, __usage_ai-training, written where you would normally name a crawler in robots.txt. This example from the draft says "don't train on my site, except the press kit":
User-Agent: __usage_ai-training
Disallow: /
Allow: /press-kit/
No new syntax, no new file. It is ordinary robots.txt grammar, keyed to a token instead of a bot name.
What Google actually committed to
Google's announcement on the Search Central blog, posted by John Mueller, says that when a site uses this token to opt out of training, and does not separately mention Google-Extended, Google plans to treat it as if the site had disallowed Google-Extended. Google-Extended is the token Google introduced in 2023 for controlling use of your content in its AI models.
Three details matter:
Timing: Google says it aims to support the control in 2027, with progress updates to come. It is not active today.
Company-specific controls stay: Google-Extended keeps working alongside the shared signal. If you set Google-Extended explicitly, that is what Google reads.
The trade-off: Google's post notes that sites allowing Google-Extended are eligible to appear in Gemini Apps responses and receive clicks through grounding. Opting out of Google-Extended, directly or through the new token, gives that up.
Other providers: a detailed write-up on fireup.pro reports that Microsoft will honour the token for Bing training, with preferences taking effect from 1 February 2027, and that OpenAI plans to support it "in the coming months". I couldn't find a firm date from every major provider, so treat coverage as partial for now.
What the token does not do
This is where people will get it wrong. Per the draft itself:
It is not a crawl block. The draft says the token "is not a directive for crawling behavior". Bots can still fetch your pages. To limit crawling, you still need rules for specific crawler names.
User-Agent: *doesn't count. A wildcard group must not be read as a training preference. Your existing catch-all rules say nothing about AI training under this scheme.Silence is not a "no". If your robots.txt has no
__usage_ai-traininggroup, you have expressed no preference either way.It is a preference, not enforcement. Only providers that choose to honour it will. It also can't pull your posts out of models that were already trained.
It is a draft. The Internet-Draft is an individual submission, not endorsed by the IETF, and expires on 3 May 2027 unless updated. The syntax could still change.
Should a blogger opt out?
Here is the honest trade-off, without the outrage or the hype.
Reasons to opt out
You sell your writing, courses or research, and AI summaries of it compete directly with what you charge for.
You simply don't want your work used to train commercial models, and you are fine losing some AI visibility for that principle.
Reasons to stay opted in
Your blog is a top-of-funnel channel. Being cited in AI answers is one of the few new sources of discovery, and at Google, opting out of Google-Extended also removes you from Gemini grounding.
You are trying to be quoted. If AI citations are part of your strategy, as we discussed in how ChatGPT, Gemini and Perplexity choose what to cite, blocking everything works against you.
The middle path most blogs should consider
The draft's Allow lines make a mixed setup easy. Keep public, top-of-funnel posts open, and opt out only the parts you sell:
User-Agent: __usage_ai-training
Allow: /
Disallow: /members/
Disallow: /courses/
That protects paid material and leaves your discovery content where AI systems can learn from and cite it.
A five-minute checklist for this month
Open your current robots.txt and note any rules for Google-Extended, GPTBot or other AI crawlers you added in the past. Decide whether they still reflect what you want.
Write down your decision (opt out, opt in or mixed) before 2027. Support is coming next year; your choice doesn't need to wait for it.
If you add the token now, keep it consistent with any Google-Extended rule. Google reads an explicit Google-Extended setting over the shared token, so contradictory lines just confuse future you.
Don't confuse this with Search Console controls. They are separate. We covered the Search Console AI performance report and its opt-out in October.
Keep working on what earns visibility anyway: original reporting, clear authorship and loyal readers. Google's new Preferred Sources feature rewards exactly that.
My take: this is a genuinely useful step, because one line beats maintaining a list of 30 bot names. But it is a 2027 feature built on a draft, and the bigger decision (whether AI visibility is worth more to you than training control) is yours to make now.



