Bing announced AI training and grounding preferences on October 30, 2026, replacing the older NoArchive and NoCache signals, according to a timeline tracking Microsoft announcements. The new controls include a shared robots.txt training token and page-level grounding options of Allow, DisallowExtended, and Disallow, honoured from February 1, 2027, with the old tags stopping counting on March 1, 2027.
The distinction is the point of the new controls. Training is when a company uses your pages to build its AI model; grounding is when an AI assistant fetches your pages to answer a user question, and can send that user to your site. Bing controls let site owners make choices about training, grounding, or both, while remaining eligible for Bing search results.
That separation matters because the two choices have opposite commercial effects. Blocking training protects content from being absorbed into a model, while blocking grounding removes the site from AI answers entirely, including the citations that can still send traffic. Publishers should decide each choice on its own merits instead of treating AI as a single on-or-off switch.
The shared training token lets publishers set AI training preferences once in robots.txt and have them apply across several AI companies, rather than configuring restrictions separately for each one. Page-level grounding options then refine what happens inside answers: Allow, DisallowExtended, or Disallow, giving page-by-page control instead of a site-wide verdict.
The timetable gives publishers months to prepare: honoured from February 1, 2027, with old tags stopping counting on March 1, 2027. That window is the time to audit which pages should stay in AI answers and which should not, rather than rushing a site-wide block that also removes valuable citations.
First, inventory your content by value in AI answers: pages that earn citations and traffic stay grounded, while pages whose value is purely in the text itself may warrant a training block. Second, coordinate the Bing settings with the shared token work happening across Google, OpenAI, Microsoft, and Anthropic, so the rules agree instead of contradicting each other.
Third, record the decision and the date, because these controls will be audited later, whether by a licensing negotiation or a traffic investigation. A documented choice made in November 2026 is far more defensible than a forgotten robots.txt edit discovered during a dispute.
The February 1, 2027, honouring date and the March 1, 2027, retirement of old tags give publishers a concrete deadline to work toward. The sensible project plan starts with an audit of current robots files and meta tags, mapping every existing NoArchive and NoCache usage to its replacement under the new scheme, and noting where the old tags were doing work the new controls handle differently.
The second step is a page-level decision log: which sections allow grounding, which disallow it, and which allow it only in extended form, with a named owner for each decision. The third step is testing, using Bing Webmaster Tools to confirm the new signals are read correctly before the old tags stop counting. A publisher that finishes that sequence in November or December will meet the switchover calmly; one that starts in February will discover its errors at the same moment everyone else does.
Publishers should also watch how the new controls interact with licensing deals. A site that licenses content to an AI company may want different robots settings from a site that does not, and a blanket training block could conflict with a paid agreement. The controls are tools, not policies, and the policy decision, what each partner may do with each content type, should be written down before the tools are configured.
Sources: Keywords Everywhere Copilot timeline, read November 1, 2026; Search Engine Roundtable, October 30, 2026.
Linkbaits helps you create content that earns citations and backlinks. Start with the free tools and build your first linkable asset this week.
Start free trial