Training and answer-time grounding are different choices
On October 30, 2026, Microsoft Bing and Google announced plans to support a shared training preference in robots.txt, using the __usage_ai-training policy token. The linked IETF document is an individual Internet-Draft, not an approved standard.
Microsoft separates training from grounding: supplying content to inform generated answers. Publishers could reject training while allowing grounding. Its page-level DisallowExtended option would permit Microsoft-operated experiences but exclude third-party grounding services. These usage preferences do not themselves remove eligibility for ordinary Bing results; normal crawling and indexing requirements remain.
Announcement is not activation
Microsoft plans to honor discovered settings as it crawls from February 1, 2027. Existing NoArchive and NoCache users need to review the transition before those tags stop serving as training and grounding preferences after March 1. Google targets support in 2027, interpreting a shared training opt-out as a Google-Extended disallow when that control is not explicitly set.
For publishers, the practical task is to separate crawler access, model training and answer citations before changing configuration. Recheck implementation and recrawl status. Permission signals do not guarantee inclusion, traffic or payment, and do not establish licensing agreements.