How to Manage Your Content's Use in AI with Microsoft's New Controls

What changed in Microsoft's AI content controls?

Microsoft announced new preferences on October 30, 2026 that separate two uses of publisher content: AI Training and AI Grounding. AI Training concerns whether content can modify the learned parameters of a generative AI model. AI Grounding concerns whether content can is input that informs an AI-generated response at the time that response is produced.

That distinction changes the decision publishers make. A site can disallow training while allowing grounding, which keeps its pages eligible to inform and be linked from AI responses. It can also permit grounding only in Microsoft-operated experiences through DisallowExtended, while excluding Microsoft grounding services offered to third parties.

The new controls replace Microsoft's legacy NoArchive and NoCache controls for training and grounding. Microsoft says NoArchive and NoCache were designed for cached search pages and treated training and grounding as one decision. The new names describe the use being controlled, and the preferences can be set independently.

The calendar matters. Microsoft calls October 30, 2026 an announcement date, not an enforcement date. It says it will begin honoring the new preferences as it crawls on February 1, 2027, and that recrawling all content will take time. Publishers using NoArchive or NoCache have a transition path in Bing Webmaster Tools. Publishers without those directives need no transition action before the new controls take effect.

For teams working on AI search visibility, grounding is the decision that most directly affects whether content may help inform a generated response. Microsoft explicitly says permission does not guarantee inclusion, traffic, or payment. Eligibility is not a citation promise. It is still the condition that keeps a page available for consideration when an answer engine selects sources.

1. AI Training controls use of content to train generative models.
2. AI Grounding controls use of content to inform generated responses.
3. Traditional Bing search indexing remains governed by normal crawl and indexing guidance.
4. Microsoft says Allow is the default for both training and grounding.

How do the old and new controls differ?

The old approach gave publishers broad legacy directives that did not distinguish between a model learning from content and an answer drawing on content at response time. The new approach asks publishers to make that distinction directly. AI Training can be configured at a site or path through robots.txt, then overridden for an individual page with a meta tag or HTTP response header.

AI Grounding is narrower in location and broader in choice. Microsoft says it is available at the page level through a meta tag or HTTP response header, rather than through robots.txt. A publisher can allow grounding, disallow it entirely, or choose DisallowExtended. That middle choice allows grounding in Microsoft-operated products while excluding third-party grounding services supplied by Microsoft.

This is a practical before-and-after change. Before the transition, a publisher relying on NoArchive or NoCache had one legacy control carrying consequences for cached pages, training, and grounding. After the transition, the publisher needs a recorded policy for two separate uses and must apply page-level grounding choices where the policy differs by page type.

The change also creates a stronger reason to classify content. A company may reach a different conclusion for public research, pricing pages, customer documentation, gated resources, and pages containing licensed material. A single sitewide reflex will miss those differences. The settings should follow a content policy that has an owner and a review date.

1. Before February 1, 2027: legacy NoArchive and NoCache may be part of a site's existing setup.
2. From February 1, 2027: Microsoft begins honoring separate AI Training and AI Grounding preferences as it crawls.
3. Training: Allow or Disallow, set in robots.txt at site or path level and available at page level.
4. Grounding: Allow, DisallowExtended, or Disallow, set at page level through a meta tag or HTTP response header.

Who does Microsoft's new AI content control affect?

The change affects publishers whose pages Microsoft crawls and any team responsible for robots.txt, response headers, metadata, content governance, or AI search strategy. It matters most when a site has a reason to distinguish model training from answer grounding, rather than treating every AI use as the same category.

Editorial publishers may want their reporting to inform Microsoft-operated AI responses while keeping it out of third-party grounding services. A B2B software company may decide that product documentation should remain eligible for grounding because customers ask answer engines implementation questions. A business with licensed data or partner material may require a more restrictive page-level rule for that material.

Technical ownership is rarely held by the same person who owns the policy decision. SEO teams may oversee indexing, legal teams may set content-use requirements, engineering teams may deploy headers, and editorial teams may know which pages contain content with special restrictions. The safest implementation process gives each decision a named reviewer before any directive changes production behavior.

Publishers should also avoid overstating what these controls do. Microsoft says the preferences govern training and grounding and do not change eligibility for traditional Bing search results. Blocking grounding does not automatically remove a page from Bing search. Allowing grounding does not guarantee a link, a citation, traffic, or commercial compensation.

For Omnicite's audience, this reinforces a basic point about Citation Share: being eligible to support an answer is different from being chosen as a cited source. Eligibility keeps the door open. Quality, coverage, freshness, and clear evidence determine whether a page is useful enough to be selected when an AI system assembles an answer.

1. Publishers using NoArchive or NoCache need to review Microsoft's transition plan.
2. Teams with public knowledge bases should decide whether those pages remain eligible for grounding.
3. Sites that license data or publish partner content should identify pages needing separate treatment.
4. SEO, engineering, legal, and editorial owners should agree on the policy before deployment.

How should publishers respond before February 2027?

Publishers should first inventory their current directives and classify the content they control. Search for NoArchive and NoCache in page metadata, HTTP headers, templates, edge rules, and robots.txt. Record the affected URLs or templates, the directive in place, the business reason, and the person responsible for deciding the replacement setting.

Next, decide separately whether each relevant content group may be used for AI Training and AI Grounding. Do not turn this into a blanket technical change before the policy is clear. If a page should remain eligible to inform AI-generated responses, Microsoft says no setting is needed because Allow is the default. If a page needs restrictions, select the narrowest rule that matches the policy.

DisallowExtended deserves careful treatment because it is a specific trade-off. Microsoft describes it as allowing grounding in Microsoft-operated experiences, including AI Mode, AI Overviews, Copilot, and Microsoft 365, while excluding grounding services Microsoft provides to third parties. Use it only when that distinction is intentional and documented.

Then map the policy to the right implementation layer. For training, Microsoft supports the shared AI Training preference in robots.txt for a site or path and also supports individual-page settings through a meta tag or HTTP header. For grounding, use a page-level meta tag or HTTP header. A header can be useful when an application cannot safely alter page templates, while a meta tag may be easier to inspect in rendered HTML.

Finally, test the production output. Confirm that the intended meta tag appears on the correct page types or that the HTTP response contains the expected header. Check that restrictive settings did not land on pages meant to remain eligible for grounding. Keep a dated change record, including the directive, affected template or route, reviewer, deployment date, and rollback method.

This work is content governance, not a trick for manipulating models. Microsoft has created controls that let publishers state preferences for defined uses. The commercial question remains editorial: which pages should remain available to inform answers where prospective customers and readers are looking for sources?

1. Inventory NoArchive and NoCache across templates, headers, and robots.txt.
2. Classify pages by permitted training use and permitted grounding use.
3. Set a page-level AI Grounding choice where the default Allow is not appropriate.
4. Use robots.txt, page metadata, or headers only where Microsoft supports that control.
5. Verify deployed directives on representative live URLs before recording the change.

What should a citation-focused content team choose?

A citation-focused team should begin from the content's purpose. If a page exists to answer a buyer's question with clear, current evidence, allowing grounding keeps it eligible to inform Microsoft AI responses. That is consistent with building content that an answer engine can assess, use, and link to. It does not create a guaranteed citation.

Training is a different policy question. Microsoft makes clear that AI Training and AI Grounding are independent. A publisher can disallow training and still allow grounding, or choose a restricted grounding path. Teams should not assume that a training decision automatically answers the grounding decision.

The best operational response is a simple policy matrix linked to page classes. For example, a team can label public editorial content, documentation, commercial pages, archived material, and licensed material, then assign an approved training and grounding choice for each class. The matrix should state where the setting is deployed and who can approve an exception.

Do not rely on the old labels after the transition. The new controls use names that correspond to their function, and Microsoft says they replace NoArchive and NoCache for training and grounding. A legacy directive may be evidence that a policy decision was once made, but it is not a complete decision record for the new controls.

The next action is concrete: complete the inventory before February 1, 2027, approve a written training and grounding policy for every page class that needs one, and test the live implementation after deployment. The source documents below should stay attached to the change record because Microsoft's published rules, dates, and product scope define what the controls mean.

1. Keep answer-oriented public pages eligible for grounding when that matches the content strategy.
2. Treat training and grounding as separate approvals.
3. Use DisallowExtended only when Microsoft-operated grounding is acceptable and third-party grounding is not.
4. Document policy exceptions by URL or template, with an owner and review date.

Sources

Source: Microsoft Bing Webmaster Tools

Microsoft announced independent preferences for AI Training and AI Grounding, with page-level controls and a February 1, 2027 enforcement start as it crawls. Microsoft Bing Webmaster Tools, 2026-10-30

Source: Microsoft Bing Blog

Microsoft's announcement explains the new publisher choices for training and grounding. Microsoft Bing Blog, 2026-10-30

Source: Microsoft Bing Webmaster Tools

Bing documents its crawlers and identifies robots.txt and Crawl Control as methods for controlling crawler interaction. Microsoft Bing Webmaster Tools, 2026-10-30