Published November 8, 2026 · LLM Results. Microsoft's October 30 announcement separates training preferences from permission to inform AI answers. Enforcement is planned to begin February 1, 2027, as content is crawled. This is preparation guidance, not a report of a completed rollout.
Choose Training and Grounding Separately
Microsoft's publisher FAQ describes training as changing a generative model's learned parameters, and grounding as supplying content for generated answers. Training preferences can cover a site or path in robots.txt, with page-level overrides. Grounding preferences are page-level only.
Grounding has three choices: Allow, Disallow, or DisallowExtended. The last permits Microsoft-operated experiences but excludes Microsoft's grounding services for third parties. Both training and grounding default to Allow without an applicable preference. These controls do not change eligibility for traditional Bing results or override blocked crawl access.
The Migration Dates That Matter
The implementation guide sets this timetable for legacy restrictions:
- December 15, 2026: temporary grounding preferences open in Bing Webmaster Tools for sites using NoArchive or NoCache on that date.
- February 1, 2027: Microsoft starts honoring new preferences on crawl.
- Before March 1, 2027: publish replacement training preferences; maintaining grounding restrictions also requires page-level preferences and the temporary Webmaster Tools choice. The old tags cease expressing these AI preferences after that date.
- September 30, 2027: temporary grounding preferences expire, or earlier when recrawling discovers a page's new setting.
The FAQ warns that publishing settings does not activate them immediately: recrawling is required. There is no temporary training setting.
A Shared Signal Is Still Work in Progress
The linked October 30 Internet-Draft proposes __usage_ai-training groups in robots.txt. The Datatracker labels it an individual draft, without IETF endorsement or formal standards standing. It expresses a use preference separately from crawler access; it is not a universal block on AI companies.
What Publishers Should Do Now
Our recommendation: inventory existing restrictions, record training and answer-visibility choices separately, and assign the relevant migration dates. Check provider documentation before implementation. Eligibility is not evidence of citations or referral gains; measure those outcomes independently.