IETF AIPREF · 2026-10-30 · major
AI training opt-out token — one robots.txt rule for OpenAI and Google
The __usage_ai-training token lets a website opt out of AI training with one robots.txt group. OpenAI and Google both said on October 30, 2026 that they will honor it; the IETF draft is written by Anthropic's Kevin Kelley.
A single robots.txt group can now tell several AI providers not to train on a site, instead of one rule per crawler.
Quick facts
| Token | __usage_ai-training |
|---|---|
| Where | robots.txt User-Agent line |
| Draft | draft-kelley-aipref-training-preference-signal-00 |
| Draft author | Kevin Kelley, Anthropic |
| OpenAI rollout | In the coming months |
| Google rollout | Aims for 2027 |
| Status | Individual Internet-Draft, not IETF-endorsed |
What is it?
The Machine-Readable Training Preference Signal is a new IETF Internet-Draft that adds a shared policy token, __usage_ai-training, to robots.txt. On October 30, 2026 OpenAI said it will honor the signal, and Google said it plans to support it along with other AI providers. Kevin Kelley of Anthropic wrote the draft, and discussion runs on the IETF AI Preferences (AIPREF) working group list.
How does it work?
The draft adds no new directives or grammar. A site writes an ordinary robots.txt group whose User-Agent line is the policy token, then uses normal Disallow and Allow paths, so 'Disallow: /' with 'Allow: /press-kit/' opts out everywhere except the press kit. The token is matched exactly and without case; a 'User-Agent: *' group does not count, and if no group names the token, no preference has been expressed.
Why does it matter?
Publishers today must list each AI crawler by name, such as Google-Extended, and keep that list current. With the shared token, Google says an opt-out will count as disallowing Google-Extended when Google-Extended is not named separately. OpenAI says its existing crawler controls stay in place, so the token adds a signal rather than replacing the old rules. Neither company has shipped it yet.
Who is it for?
website owners, publishers, SEO and web platform teams
Frequently asked questions
- When will OpenAI and Google start honoring the __usage_ai-training token?
- OpenAI says it aims to implement the __usage_ai-training signal in the coming months and will publish implementation details later. Google's Search Central blog says Google aims to support the shared AI training control along with other AI providers in 2027 and will share progress updates. As of October 30, 2026, neither company has a live implementation.
- Does the AI training opt-out token replace GPTBot and Google-Extended rules?
- No. OpenAI says the new training preference signal will not replace the controls it already provides through robots.txt, and Google says company-specific controls such as Google-Extended keep working alongside it. Google adds that if a site opts out with the shared token and does not name Google-Extended, Google will treat that as a Google-Extended disallow.
- Is the __usage_ai-training token an official IETF standard?
- Not yet. draft-kelley-aipref-training-preference-signal-00 is an individual Internet-Draft by Kevin Kelley of Anthropic, dated 30 October 2026 and expiring 3 May 2027. It states that it is not endorsed by the IETF and has no formal standing. Discussion happens on the AI Preferences working group mailing list, and the wider AIPREF vocabulary is still being debated.
- Can a site opt out of AI training for only some pages?
- Yes. The training preference draft reuses robots.txt path rules and longest-match precedence, so a site can mix Disallow and Allow lines inside the __usage_ai-training group. The draft's own example disallows the whole site but allows /press-kit/, which signals that press materials may be used for AI training while the rest of the site may not.
Try it
Add to robots.txt: User-Agent: __usage_ai-training / Disallow: /