On October 30, OpenAI, Google and Microsoft announced plans for a shared AI training preference.
The big change
One shared rule can reduce separate provider instructions while leaving Bing search eligibility unchanged. It expresses a use preference, not crawl restrictions; participants must first implement it.
The proposed token
The draft expresses a whole-site training opt-out in robots.txt:
User-Agent: __usage_ai-training
Disallow: /The draft also allows path-specific preferences.
This individual Internet-Draft is not endorsed by the IETF and has no formal standards-process standing.
OpenAI says it aims to implement the signal in the coming months and retain its existing crawler controls. Google plans support in 2027. Google plans to treat the shared opt-out as a Google-Extended disallow when Google-Extended is unspecified. Explicitly allowing Google-Extended makes pages eligible for grounding in Gemini Apps. Provider-specific controls remain available.
BIG CHANGE has not tested provider implementations.
Training and grounding
Training changes model parameters; grounding informs answers. Crawling retrieves pages.
Microsoft will honor shared training and separate page-level grounding preferences as it crawls from February 1, 2027.
Microsoft says it will disregard legacy training and grounding preferences on detecting new settings. After March 1, 2027, NoArchive and NoCache no longer express those preferences. By that date, sites using those tags on December 15, 2026 must publish page-level grounding preferences and record temporary Bing Webmaster Tools protection to retain restrictions; it lasts until recrawl or September 30, 2027, whichever comes first.
Limits
A nonparticipant may ignore the signal. A published preference does not prove compliance.
