A publisher that wants Bing to link to its articles in AI answers but does not want those articles used to train future models will be able to express those choices separately. Microsoft’s October 30 announcement says the new settings begin taking effect as Bing crawls pages from February 1, 2027. On March 1, the older NoArchive and NoCache tags stop expressing training and grounding preferences. If your site relies on either tag, find out where it is set before adding the replacement anywhere.
Jump to a section
What each choice does
Training is the use of content to help build future models. Grounding is the use of a page as a source when an AI system generates an answer. Microsoft says the new preferences leave eligibility for ordinary Bing search results intact. The settings cover different uses of the same page, so deciding one does not decide the other.
The Bing Webmaster Tools implementation guide describes three practical choices. A publisher can disallow training while leaving grounding available. It can restrict grounding to Microsoft-operated products using DisallowExtended, excluding grounding services Microsoft provides to third parties. Or it can disallow grounding altogether. Decide training separately in the last two cases. Eligibility for a citation gives you a chance to appear, and the actual appearance is something you can investigate with our Bing citation reporting guide.
| Publisher goal | Training preference | Grounding preference |
|---|---|---|
| Allow AI citations and opt out of training | Disallow training | Leave grounding available |
| Allow Microsoft’s AI products to cite you, but restrict third-party services | Set separately | DisallowExtended |
| Keep pages out of AI answers grounded by Microsoft | Set separately | Disallow grounding |
Why a partial rollout can change more than one page
Microsoft says that once it detects either new preference on a site, it will disregard the old NoArchive and NoCache preferences on that site. Consider a publisher that adds a training restriction to its public news pages while leaving its paid archive on a template that still uses NoArchive. The public change can remove the old grounding signal the archive was relying on. Audit both templates before introducing the first new signal, including rules added by your hosting or caching service.
The decisions belong in different places. Microsoft accepts a site or path training preference in robots.txt and allows a page-level training preference to override it. Grounding is a page-level preference. A newsroom can therefore choose a default for training and document an exception for a specific page, while making sure each page that needs a grounding restriction actually serves one. Keep access control independent. The paywall access guide explains why a response header cannot substitute for restricting who receives subscriber-only text.
A rollout example for a site with a paid archive
Suppose a publication wants its free reporting to appear in AI answers but wants to keep its paid research out. It also wants to disallow training on both. The editor can describe those as three outcomes: public reporting remains eligible for grounding, paid research is excluded from grounding, and neither is available for future training under the new Microsoft preferences. Give that statement to the person who controls the templates and to the person who controls robots.txt. It gives both people a way to catch a rollout that meets only part of the policy.
Make a small launch sheet with one URL per template and a column for training, grounding and signed-out access. For example, the public article should load for a signed-out reader and remain eligible for grounding. The research article should expose only what the publisher intends a signed-out reader to see and serve the intended grounding restriction. Check the homepage too if it uses a different template or republishes summaries. If the public story passes but the research URL still serves only a legacy tag, the archive check has failed even though the new training setting is visible.
Check syndication separately. If a partner publishes the same article under a different domain, a preference on your original URL does not configure the partner’s copy. Ask the partner what its version serves. Treat that as a separate item in the rollout sheet, particularly for full-text copies that a reader can find without visiting your site.
How to verify the transition on your site
Start with a public article, a subscriber article and one older archive article. Save the current robots.txt file and inspect each page as a signed-out reader. Capture response headers and page source. Ask which system owns each signal. A setting injected by the content management system can coexist with a conflicting header added at the edge, and an archive template may omit a change made to the current article layout.
If you are carrying over an existing grounding restriction, Microsoft says to publish the page preference and record the temporary transition preference in Bing Webmaster Tools before March 1. The temporary setting applies per page until Bing recrawls it or September 30, 2027, whichever comes first. Microsoft does not offer that temporary setting for training. Keep the date, sample URLs and observed settings together so your team can distinguish a deployed preference from a page still waiting to be crawled.
After February 1, repeat the checks on the same URLs. If one page serves the intended setting and another does not, fix the template before drawing conclusions from changes in citations. If both serve the right setting, a movement in citations still tells you little about whether a specific rule caused it. Compare the same pages and dates, and keep recrawl status and changes to the article itself in view. The test here verifies what your site sends. Microsoft’s application of the preference is a separate observation.
What to hand off after the settings are live
Record the deployment date, the URLs tested and which templates they represent. Include a signed-out response for each. Assign one person to check the sample again after the next template or caching change. That gives the team something specific to compare if a restriction disappears or a previously open page starts serving the wrong preference.
If you track Bing search traffic and AI citations, save separate baselines before changing the policy. A drop in AI citations can be consistent with restricting grounding. A change in ordinary search traffic deserves its own investigation, including whether a template change also altered indexing or access. Keeping the two records separate makes it easier to diagnose the change without assuming the new preference affected both.
