Microsoft is expanding the choices available to web publishers by distinguishing two uses of online content that are often conflated: training AI models and grounding AI-generated responses. In an October 2026 announcement titled Your content, your choice: clearer preferences for training and grounding, the company says publishers will be able to express independent preferences for these activities without changing their eligibility to appear in traditional Bing search results.
The distinction matters for publishers who want their pages discovered in search but have different policies about the use of those pages in model development and AI-generated answers. It also signals a broader move toward standardized, machine-readable choices across the web.
AI training and grounding are different uses of content
AI training refers to using content as part of the process of developing or updating a model's learned parameters. A training preference addresses whether a publisher permits that use of its material.
AI grounding is different. It involves retrieving current or relevant information to inform an AI response at the time the answer is generated. Grounding can help AI systems reference information that was not included in training data, including recent updates and specialist sources.
A website may be comfortable with one use but not the other. For example, a publisher might wish to remain a source for AI-generated answers while declining to contribute its material to model training. Another might prioritize control over both activities. Independent preferences make these distinctions easier to express.
What Microsoft is proposing
Microsoft's announcement describes expanded publisher choice built on a shared industry AI training preference, with grounding handled as a separate decision. The company states that these choices will not change eligibility for Bing's conventional search results. This is a meaningful assurance for site owners concerned that an AI-specific preference might unintentionally remove their pages from standard web discovery.
However, eligibility is not a guarantee of ranking, impressions or traffic. Nor should a statement about conventional Bing search be interpreted as a promise that restricting grounding will leave visibility in AI-generated experiences unchanged. These are distinct discovery surfaces with distinct uses of publisher content.
How this fits the emerging shared training standard
The IETF AI Preferences (AIPREF) working group is developing interoperable mechanisms for publishers to communicate preferences about AI-related uses of their material. Google and OpenAI have also announced plans to support a common AI training preference through robots.txt. Microsoft's announcement builds on that direction while emphasizing that training and grounding should be independently controllable.
Standardization could reduce the need to maintain separate provider-specific instructions, but the real effect depends on final specifications, documented implementation and provider adoption. An announced plan is not evidence that every AI service already honors a new signal. Publishers should rely on current official technical documentation before deploying new directives.
What about Bing's existing controls?
Microsoft has offered publisher controls for AI-related content use before. In September 2023, Bing explained how the nocache and noarchive meta tags affected use of content in Bing Chat and training, while preserving eligibility for standard Bing search. Those historical controls were tied to specific product behavior and should not be assumed to be identical to the newly announced independent preferences.
Microsoft has also expanded Bing Webmaster Tools' AI Performance reporting, giving site owners information about citations, grounding queries and pages referenced in supported AI experiences. These metrics are useful for evaluating AI visibility, but they do not themselves establish or change permission to train models on a site's content.
What publishers and technical SEO teams should do
Begin by separating three objectives in your content policy: eligibility for conventional search indexing, permission to use content for model training, and permission to use content to ground AI answers. A single 'allow AI' or 'block AI' policy is often too broad to describe what a publisher actually wants.
Next, audit current robots.txt directives and page-level meta tags. Document which controls are already active, which providers they address and whether the website has commercial licensing agreements that require different treatment for particular content. Do not remove existing controls merely because a new industry-wide mechanism has been announced.
Finally, watch Microsoft's implementation documentation for exact syntax, rollout timing, scope and interactions with existing directives. The announcement describes the direction of policy, but a reliable implementation guide must specify what site owners can deploy and how conflicts are resolved. Until those details are confirmed, publishing speculative configuration snippets risks misleading readers.
Why the separation matters for AI Search
The business implications extend beyond compliance. Grounding can influence whether content appears as a source in AI answers, while training is a different kind of downstream use that may not produce a direct referral opportunity. Keeping those permissions distinct gives publishers a more precise way to balance intellectual-property preferences with discoverability.
For SEO and GEO professionals, the lesson is to measure search rankings and AI citations separately, and to avoid equating an AI training opt-out with deindexing. Microsoft is framing the future around publisher choice rather than a single all-or-nothing switch. The effectiveness of that approach will ultimately depend on clear specifications, broad adoption and transparent enforcement.
Sources: Microsoft Bing announcement (URL supplied in the editorial brief; page not independently retrievable during preparation); Microsoft's 2023 publisher controls; Bing Webmaster Tools AI Performance; IETF AIPREF working group.