What Do Microsoft's New AI Training and Grounding Preferences Mean for Your Brand?

What changed in Microsoft's AI grounding preferences?

Microsoft announced two independent publisher preferences: AI Training and AI Grounding. AI Training covers whether content may be used to modify the learned parameters of future generative AI models. AI Grounding covers whether content may be used as input to help inform an AI-generated response in Bing, Copilot and Microsoft 365.

Before this change, Microsoft used NoArchive and NoCache tags as controls connected to training and grounding. From February 1, 2027, Microsoft says it will honor the new preferences as it crawls. If Microsoft detects the new AI Training or AI Grounding preferences, it will disregard remaining NoArchive and NoCache preferences on that site.

The separation matters because training and grounding are different uses of a page. A publisher can disallow training while allowing grounding. That approach keeps a page eligible to inform generated responses and receive the links or citations that Microsoft says can connect users back to the original publisher.

Microsoft says these choices do not change eligibility for traditional Bing search results, subject to normal crawling and indexing requirements. The preferences also do not create licensing or payment arrangements, and Microsoft says allowing either setting does not guarantee inclusion, traffic or payment.

For a brand, that creates a more precise question than whether to allow AI. Decide whether each important page should help answer live questions, whether it should contribute to future model training, and whether a technical control currently makes that choice for you.

1. AI Training can be set in robots.txt for a site or path, or at page level.
2. AI Grounding is a page-level setting.
3. No applicable preference means Allow is the default for both uses, according to Microsoft.
4. A page-level AI Training preference overrides the robots.txt training setting.

What is the before-and-after for publishers using NoArchive or NoCache?

The change replaces an older, combined control pattern with separate preferences that identify the intended use of content. Microsoft tells sites that currently use NoArchive or NoCache to publish the new preferences and complete the transition by March 1, 2027.

The transition is time-bound. Microsoft says the new preferences become effective from February 1, 2027 as it crawls, while NoArchive and NoCache will no longer be read as a preference for training or grounding after March 1, 2027. A brand that relies on those older tags should not assume that its present implementation continues to express its intended policy.

The operational risk is not only legal or governance related. Microsoft explicitly says that choosing Disallow or DisallowExtended for grounding reduces brand visibility in AI experiences. For publishers that want to be cited in answer engines, a broad block could remove pages from the pool used to inform generated responses.

This is why technical controls need an editorial inventory. A marketing team may want product, category, comparison and help pages to remain eligible for grounding. Legal, private account, duplicate, thin or obsolete pages may have a different policy. The rule should come from page purpose, not from a one-line robots.txt change made without a page review.

1. Before February 1, 2027: identify every use of NoArchive and NoCache.
2. From February 1, 2027: Microsoft honors the new AI Training and AI Grounding preferences as it crawls.
3. By March 1, 2027: publish the replacement preferences if older tags were intended to control training or grounding.
4. After March 1, 2027: do not rely on NoArchive or NoCache to express those preferences.

How do AI training and AI grounding preferences differ?

AI training and AI grounding answer separate policy questions. Training concerns whether a generative model may use an asset to modify learned model parameters. Grounding concerns whether a page may be used as input when an AI system creates a response.

That distinction is useful for brands because citation visibility depends more directly on grounding. Microsoft describes grounding as connecting generated responses to web content, with links and citations providing a path to the original publisher. A page can therefore be available to inform a response even when the publisher does not want it used for future model training.

Microsoft describes three grounding choices. Allow is the default and permits content to help inform generated responses. DisallowExtended permits grounding only in Microsoft-operated experiences, including AI Mode, AI Overview, Copilot and Microsoft 365, while excluding grounding services Microsoft provides to third parties. Disallow prevents grounding use.

The right choice is not universal. A business that wants broad visibility in Microsoft AI experiences may use Allow on public editorial pages. A publisher that wants visibility in Microsoft-operated products but not through third-party grounding services may assess DisallowExtended. A page that should not inform generated answers needs Disallow at page level.

The important point for brand teams is that a training decision no longer has to be a grounding decision. Treat them as two documented policies, then verify the live implementation on the specific URLs that matter for citation share.

1. Training policy asks whether a page can contribute to future generative-model training.
2. Grounding policy asks whether a page can help inform a generated response now.
3. Traditional Bing search eligibility is separate from these preferences.
4. Citation opportunities depend on an answer system being able and willing to use a page for grounding.

Who does Microsoft's change affect?

The change affects any publisher whose content is crawled by Microsoft and could be used for training or web-grounded responses in Bing, Copilot or Microsoft 365. It is especially relevant to brands that publish public product pages, category pages, comparison pages, support content, location pages or editorial resources.

SEO teams are affected because robots.txt and page directives are often managed inside technical SEO workflows. Content teams are affected because page-level grounding requires a judgment about what a page is for. Legal and governance teams are affected because the settings express preferences about how public content may be used.

Brands that only monitor conventional rankings should still review the change. Microsoft says grounding is distinct from traditional indexing and ranking. A page can remain eligible for Bing search results while being disallowed from grounding, which means a search visibility check alone cannot confirm answer visibility.

Companies that use NoArchive or NoCache have the most immediate transition work. Microsoft provides a temporary grounding preference path in Bing Webmaster Tools for sites using those tags. For each page, Microsoft says the temporary preference applies until recrawl or September 30, 2027, whichever comes first.

The change also affects organizations that use a shared training control across providers. Google announced plans to support a shared AI training control in robots.txt, based on work from the IETF AIPREF working group. Google says it aims to support the new control in 2027, while existing provider-specific controls continue to function.

1. Publishers using NoArchive or NoCache need a transition plan.
2. Brands seeking answer-engine citations need a page-level grounding policy.
3. Technical SEO owners need to audit robots.txt, meta tags and HTTP response headers.
4. Legal and content owners need an agreed rule for public pages with different business purposes.

How should a brand respond to the new AI grounding preferences?

Start with a control inventory, then make a page-purpose decision before editing directives. The first job is to find existing NoArchive, NoCache, Bingbot directives, robots.txt rules and response headers. The second is to identify which URLs the brand wants available to inform generated answers.

Prioritize pages that explain a product, compare options, answer buyer questions, establish expertise or support a location. These are the pages most likely to matter when a prospective customer asks an answer engine for help. Omnicite calls the resulting measure Citation Share: the percentage of relevant AI answers in a category that cite a brand.

Do not treat a single crawl directive as a growth tactic. Microsoft says that allowing grounding does not guarantee inclusion, traffic or payment. Answer inclusion still depends on whether a system can use the page for a relevant response. Brands need clear, current content that covers the questions their buyers ask and can be checked against authoritative sources.

Set ownership before implementation. Technical SEO can publish a robots.txt change, but page-level grounding needs editorial and legal review. An approved policy should state which page groups receive Allow, DisallowExtended or Disallow, who can change those settings, and how the change is verified after deployment.

Measure the result repeatedly. Citation Share is a useful headline metric, while Citation Count per day records volume and Answer Presence shows breadth across the question set. Track a fixed set of relevant prompts across the engines that matter to the business, then investigate a change before attributing it to one directive.

1. Audit existing directives and affected URL templates.
2. Classify public pages by their intended answer-engine role.
3. Choose training and grounding preferences separately.
4. Publish and verify the live robots.txt, meta tags or HTTP headers.
5. Monitor citation visibility with a consistent prompt set after implementation.

What should brands avoid when changing these controls?

Avoid assuming that a training opt-out preserves grounding visibility everywhere. Microsoft separates the choices, but each provider has its own implementation. Google says that when a website uses the shared policy token to opt out of training and Google-Extended is not explicitly specified, Google plans to interpret that as disallowing Google-Extended. Google also says sites that specifically allow Google-Extended are eligible for grounding in Gemini Apps.

Avoid applying a page-level grounding block without recording why the page was blocked. Microsoft says Disallow and DisallowExtended reduce brand visibility in AI experiences. A page that must be excluded may deserve that outcome, but a public comparison page or buyer guide could lose a route into relevant answers.

Avoid treating NoArchive and NoCache as a permanent answer after March 1, 2027. Microsoft says it will no longer read those directives as a training or grounding preference after that date. A site with no replacement may end up expressing no applicable preference, where Allow is the default.

Avoid confusing search eligibility with answer eligibility. Microsoft says its preferences do not change traditional Bing search-result eligibility, but grounding determines whether content can help inform generated answers. Measure both surfaces if search discovery and AI citations are business goals.

Avoid promising a citation outcome from a technical setting. The mechanism is eligibility and publisher choice. Citation performance still requires coverage, quality and freshness that make a page useful for a relevant answer.

1. Do not copy a control from one provider into another provider's policy without checking the documented behavior.
2. Do not use a site-wide decision when page purpose differs across the site.
3. Do not make the change without a rollback record and a URL-level verification method.
4. Do not equate an Allow setting with guaranteed citations.

What is the practical deadline for AI grounding preference changes?

The practical deadline is March 1, 2027 for sites that currently use NoArchive or NoCache as their training or grounding preference. Microsoft says the new preferences begin taking effect as it crawls from February 1, 2027, then stops reading NoArchive and NoCache for training and grounding after March 1, 2027.

That gives brands a clear sequence. Audit current directives before the effective date, approve separate training and grounding policies, implement the correct setting for each affected page group, and verify live output after deployment. Microsoft's temporary Bing Webmaster Tools option can help sites that use the older tags manage grounding preferences during the transition.

For brands that want to be cited, the default working position should be deliberate rather than accidental. Keep public pages that are intended to answer real buyer questions eligible for grounding where that fits the organization's policy. Then measure whether those pages appear across relevant answer-engine prompts.

Microsoft's change does not settle every cross-provider control question. Google's shared training-control plan is for 2027, and the IETF work is still moving toward a final standard. has a provider-by-provider record of what each directive does, when it takes effect and which page types it covers.

1. February 1, 2027: Microsoft begins honoring the new preferences as it crawls.
2. March 1, 2027: Microsoft stops reading NoArchive and NoCache as training or grounding preferences.
3. September 30, 2027: latest stated end date for a temporary Bing Webmaster Tools grounding preference, unless recrawl happens earlier.
4. Ongoing: review provider documentation before changing shared or provider-specific controls.

Sources

Source: Microsoft Bing Blog

Microsoft introduced independent AI Training and AI Grounding preferences, described the transition away from NoArchive and NoCache, and set February 1, 2027 and March 1, 2027 transition dates. Microsoft Bing Blog, 2026-10-30

Source: Google Search Central Blog

Google announced plans to support a shared AI training control in robots.txt in 2027 and described its relationship to Google-Extended. Google Search Central Blog, 2026-10-30

Source: Jayant Solanki

The weekly source summarized Microsoft's split between training and grounding preferences and the replacement of NoArchive and NoCache. Jayant Solanki, 2026-11-01