How Microsoft's New AI Preferences Could Affect Your Brand's AI Presence

What changed in Microsoft's AI preferences?

Microsoft announced new publisher controls on October 30, 2026 that treat AI training and AI grounding as separate choices. AI training covers whether content can be used to modify the learned parameters of future generative models. AI grounding covers whether a page can be used as input to help inform a generated response.

That distinction matters because a publisher can now restrict training while still allowing its work to inform answers in Microsoft experiences. Microsoft says the new preferences apply to web content collected through its crawlers and used for training or web-grounded responses in Bing, Copilot, and Microsoft 365. Traditional Bing search eligibility does not change, subject to normal crawl and indexing requirements.

Microsoft is also adopting a shared AI training preference for robots.txt. The shared signal is based on an active individual Internet-Draft, Machine-Readable Training Preference Signal, rather than a completed IETF standard. The IETF Datatracker states that the draft is not endorsed by the IETF and has no formal standing in its standards process.

For publishers, the operational change is clear. Training can be managed at site, path, or page level. Grounding must be managed page by page. A brand that treats both controls as one setting can accidentally limit where its expertise appears in AI-generated answers.

1. AI Training controls use of content to train future generative models.
2. AI Grounding controls use of content to help inform generated responses.
3. Neither setting changes eligibility for traditional Bing search results, according to Microsoft.
4. Neither setting guarantees inclusion, traffic, or payment.

Why does AI grounding affect brand presence in AI answers?

AI grounding affects brand presence because it determines whether Microsoft can use a page to help inform an AI-generated response. Microsoft states that allowing grounding makes content eligible to inform responses, with links and citations providing a path back to the original publisher.

That eligibility is not the same as a citation guarantee. A page can be available for grounding without appearing in a response, and a page can rank in Bing search without being eligible to inform a generated answer. The practical question is whether a publisher wants its content available when a user asks a question that the page can answer well.

Microsoft says a grounding setting of Disallow or DisallowExtended will reduce brand visibility in AI experiences. Disallow prevents grounding use. DisallowExtended permits grounding in Microsoft-operated experiences such as AI Mode, AI Overview, Copilot, and Microsoft 365, while excluding grounding services Microsoft provides to third parties, including Web IQ and grounding through Foundry API access.

That makes AI grounding a content-distribution policy as much as a crawl-control policy. A company can preserve ordinary Bing visibility while narrowing its presence in answers. Teams that want to be cited should treat the setting alongside their editorial coverage, source quality, and measurement of Citation Share.

1. Allow means content may help inform generated responses and is the default when no applicable grounding preference is published.
2. DisallowExtended limits grounding to Microsoft-operated experiences.
3. Disallow prevents content from helping inform generated responses.
4. Training and grounding choices are independent under Microsoft's new system.

Which publishers need to act before March 1, 2027?

Publishers that currently use NoArchive or NoCache need to act before March 1, 2027. Microsoft says it will begin honoring the new preferences as it crawls on February 1, 2027, then stop reading NoArchive and NoCache as training or grounding preferences after the March deadline.

The largest immediate risk sits with sites that used those older tags to keep content out of grounding. Microsoft says pages without the required transition setting by March 1 can default to allowing grounding until Microsoft recrawls them. That creates a gap between a team's intended policy and the policy Microsoft can apply.

The change also matters for sites that do not use NoArchive or NoCache. Their default is important: where no applicable grounding preference is published, content may be used to help inform generated responses. A publisher that wishes to restrict grounding needs an explicit page-level setting rather than assuming a robots.txt rule will cover it.

Content teams should inventory templates, headers, content delivery rules, and legacy directives before changing preferences. A global tag can affect thousands of pages. A single policy decision can therefore change the pool of pages available to appear as sources in generated answers.

1. Audit all NoArchive and NoCache use before February 1, 2027.
2. Identify pages where the business wants to allow, limit, or prevent grounding.
3. Set the new preferences before March 1, 2027 if legacy tags are in use.
4. Record the policy owner and preserve a test record for each implementation.

How do the new controls differ from the old Microsoft tags?

The old Microsoft approach used NoArchive and NoCache as controls for training and grounding. The new approach replaces them with separate preferences that state the intended use more directly. Microsoft says that when it detects either new AI Training or AI Grounding preferences on a site, it will disregard any remaining NoArchive or NoCache preferences on that site.

The new model gives publishers a more specific choice. AI Training can be set through robots.txt for a site or path, or at page level through a meta tag or HTTP response header. A page-level AI Training preference overrides the robots.txt setting for that page. AI Grounding is page-level only.

This matters for sites with mixed content. A publisher may decide to make general educational pages available for grounding while excluding paid research, customer documentation, account pages, or other restricted material. The implementation should reflect the actual content policy, not a vague preference for more or less AI exposure.

A broad disallow may be appropriate for a publisher that has a clear rights or distribution constraint. It is a poor substitute for an editorial decision about which content should be eligible to answer customer questions. The policy should be reviewed by the people who own content, legal constraints, and acquisition goals.

1. Before: NoArchive and NoCache were Microsoft's training and grounding controls.
2. After: AI Training and AI Grounding are independent preferences.
3. Before: legacy tags could is an existing restriction.
4. After: legacy tags stop being read as training or grounding preferences after March 1, 2027.

What should brands do if they want to be cited in Microsoft AI answers?

Brands that want citation opportunities should preserve AI Grounding: Allow on pages that provide clear, current, source-backed answers. Microsoft says allowing grounding makes a page eligible to help inform generated responses, but eligibility is only the beginning. The page still needs to answer a real question with enough specificity for an answer engine to use it.

Start with pages that match the questions customers ask before choosing a vendor, service provider, or local business. Give each page a direct answer near the top, explain the mechanism or evidence, and maintain dated sources where the subject requires external proof. Pages written only to repeat category terms have little reason to be cited.

Next, segment content by purpose. Product pages, comparison pages, technical documentation, help content, and editorial research may deserve different grounding policies. A page-level preference gives a company the ability to apply those distinctions without removing the entire site from Microsoft's grounded answers.

Finally, measure before and after the deployment. Monitor whether the brand appears on the question set that matters, which pages appear as cited sources, and whether named competitors gain or lose answer presence. Omnicite calls the percentage of relevant AI answers in a category that cite a brand Citation Share.

1. Keep grounding available on pages intended to answer customer questions.
2. Use current evidence and direct answers that can stand alone in a generated response.
3. Apply page-level restrictions only where a documented policy requires them.
4. Track Citation Share, Answer Presence, and competitor Share of Voice after release.

Should a brand allow grounding but block AI training?

A brand can allow grounding while blocking AI training under Microsoft's new preferences. Microsoft gives the example of setting AI Training to Disallow and AI Grounding to Allow, which keeps content eligible to inform generated responses without allowing it to train generative models.

That configuration fits a publisher that wants visibility and citations in current AI experiences while retaining a separate policy on future model training. It does not guarantee a citation, traffic, or payment. Microsoft states those limits directly, and a sound content strategy should preserve them rather than treating the preference as a performance switch.

The right setting depends on content rights, customer commitments, product strategy, and the role AI answer visibility plays in acquisition. A public FAQ may have a different policy from an original dataset, a gated resource, or a customer-specific document. Teams should document the rule before an engineer implements it.

The same split does not map neatly across every provider. Google's October 30, 2026 announcement says that when a site uses the shared opt-out signal and does not explicitly specify Google-Extended, Google plans to interpret that as a Google-Extended disallow. Google says sites that specifically allow Google-Extended are eligible to appear through grounding in Gemini Apps. Provider-specific controls still matter.

1. Microsoft permits independent AI Training and AI Grounding choices.
2. Allow grounding plus disallow training can preserve eligibility to inform generated answers.
3. Google's planned interpretation of the shared training control differs from Microsoft's separate grounding preference.
4. Use provider-specific documentation before applying one rule across all engines.

How should technical teams implement the change safely?

Technical teams should implement the change through a controlled inventory, policy decision, deployment, and verification cycle. Microsoft's documentation states that the AI Training preference can appear in robots.txt or at page level, while AI Grounding requires a page-level setting. That technical split should appear in the rollout plan.

First, crawl or query the site for NoArchive and NoCache directives, including meta tags, HTTP response headers, CMS templates, middleware, and CDN rules. Record the URL pattern, current directive, intended training preference, intended grounding preference, and owner. Do not remove a legacy directive until the replacement policy is ready to ship.

Second, choose whether each page group should allow grounding, use DisallowExtended, or use Disallow. Test on representative URLs in each template class. Verify the rendered HTML and HTTP headers because a CMS setting that appears in an admin interface may not reach the public response.

Third, publish the new controls before the deadline and retain evidence of the implementation. Microsoft says its transition grounding preference in Bing Webmaster Tools applies until recrawl or September 30, 2027, whichever is earlier. Treat that as a temporary safeguard, not a replacement for page-level implementation.

Fourth, monitor recrawl and answer visibility after deployment. Check grounded answers for high-intent prompts, record cited URLs, and investigate unexpected changes. A drop in citations could reflect a new preference, weaker content coverage, a competitor's stronger source, or a change in the answer surface. The deployment record makes that investigation possible.

1. Inventory legacy directives across tags, headers, templates, middleware, and CDN rules.
2. Document an intended training and grounding policy for each URL group.
3. Test the public HTML and HTTP response on representative pages.
4. Deploy, monitor recrawl, and compare answer presence against a pre-change baseline.

What is the practical before-and-after timeline?

The practical timeline has two dates. Microsoft announced the new system on October 30, 2026 and says it will honor the new preferences as it crawls beginning February 1, 2027. It will stop recognizing NoArchive and NoCache as training or grounding preferences after March 1, 2027.

The narrow gap between those dates means a transition should not wait for the deadline. A team with legacy controls needs enough time to find all implementations, choose a page policy, publish the replacement settings, verify production output, and monitor how Microsoft crawls the updated pages.

For brands seeking AI presence, the work is not limited to technical compliance. The pages left eligible for grounding should earn that eligibility through clear answers, maintained facts, useful comparisons, and visible sources. That is the content side of Citation Engineering: build authoritative coverage that an AI system can use and a reader can verify.

Microsoft's new controls make the policy more explicit. They also make the trade-off visible. A brand can limit grounding and expect less visibility in Microsoft AI experiences, or it can allow grounding and compete to become a cited source. The setting should reflect a deliberate business decision.

1. October 30, 2026: Microsoft announced separate training and grounding preferences.
2. February 1, 2027: Microsoft begins honoring the new preferences as it crawls.
3. March 1, 2027: Microsoft stops reading NoArchive and NoCache as training or grounding preferences.
4. After deployment: monitor answer presence and cited URLs on the prompts that matter to the business.

Sources

Source: Microsoft Bing Blog

Microsoft announced independent AI Training and AI Grounding preferences, the February 1, 2027 implementation date, and the March 1, 2027 retirement of NoArchive and NoCache for these uses. Microsoft Bing Blog, 2026-10-30

Source: Google Search Central Blog

Google plans to support a shared AI training control in 2027 and says a shared opt-out without an explicit Google-Extended setting will be interpreted as a Google-Extended disallow. Google Search Central Blog, 2026-10-30

Source: IETF Datatracker

The shared machine-readable training preference is an active individual Internet-Draft that is not endorsed by the IETF and has no formal standing in the IETF standards process. IETF Datatracker, 2026-10-30

Source: Search Engine Journal

Search Engine Journal reported the announced shared preference, the Microsoft transition dates, and the differences between Google and Microsoft handling of the signal. Search Engine Journal, 2026-11-01