Bing Lets Marketers Separate AI Citations From AI Training"

AI Search Visibility

A marketing team can now keep its content out of model training and still compete for citations in AI answers. The defaults and the March 2027 deadline decide how much work that takes.

Marketing leaders can now make two artificial intelligence (AI) decisions separately. On October 30, 2026, Microsoft announced one setting for whether its systems may train generative models on your web content and another for whether AI answers may draw on it. Neither setting changes a page's eligibility for traditional Bing search results (Madhavan).

Until now, Microsoft read two older tags, NoArchive and NoCache, as the publisher's preference for both uses. A site that set them limited training and grounding together. The new settings let a team answer each question on its own terms.

Opting out of training no longer costs visibility in Microsoft's products

Training means using content to change a generative model's learned parameters. Grounding means using content as input when an AI system writes an answer, often with a link and citation back to the source (Madhavan).

Microsoft treats the two as independent. A site can disallow training and keep grounding allowed, which leaves its pages eligible for citations in AI answers and for standard search results. That removes the main visibility argument against opting out of training, at least across Microsoft's products.

That removes the main visibility argument against opting out of training, at least across Microsoft's products.

Training still deserves a real discussion. Microsoft says allowing it can help improve future models, and some companies will want their category expertise represented in those models. Neither choice comes with payment or a licensing arrangement (Madhavan).

Restricting grounding carries a visibility cost Microsoft states openly

Microsoft says that choosing Disallow or DisallowExtended for grounding will reduce a site's visibility in AI experiences (Madhavan). Grounding feeds Copilot and Microsoft 365, along with Bing's AI Mode and AI Overview.

DisallowExtended sits in the middle. It keeps a page available to those Microsoft-operated experiences and excludes the grounding services Microsoft provides to third parties, such as Web IQ and grounding through application programming interface (API) access in Foundry (Madhavan). It suits openly published content a company treats as proprietary, such as full research articles. Buyers still find that content in the Microsoft tools they use at work, and Microsoft stops passing it to other companies' AI products through its own services.

The setting covers Microsoft's services only. Other companies' crawlers follow their own rules.

Grounding decisions happen page by page

Microsoft offers no site-wide robots.txt setting for grounding. The preference goes in a meta tag or HTTP header on each page (Madhavan). That makes grounding a content operations decision. The team that knows which pages generate demand and which pages hold premium value should set the policy, and the content management system (CMS) templates should apply it by content type.

Two limits matter for planning. The settings govern how Microsoft uses content, so an Allow setting does nothing for a page that robots.txt blocks from Bingbot. A new setting also takes effect only after Microsoft recrawls the page.

The training syntax in robots.txt comes from an Internet Engineering Task Force (IETF) draft Microsoft is developing with other industry participants, while the grounding tag is Microsoft's own and addresses Bingbot (Madhavan).

Sites using the old tags default to Allow after March 1, 2027

Microsoft begins honoring the new settings on February 1, 2027. After March 1, 2027, it no longer reads NoArchive or NoCache as a training or grounding preference, and a page without a new setting defaults to Allow for both (Madhavan).

A company that used those tags to limit AI use needs to publish the new settings before that date to keep its restrictions. For grounding, Microsoft also asks those sites to record a temporary preference in Bing Webmaster Tools, available from December 15, 2026, because a recrawl may not happen before the deadline. The temporary preference lasts until Microsoft recrawls the page or until September 30, 2027, whichever comes first (Madhavan).

Those tags often date back to an earlier web team or a CMS default, so many marketing leaders have never reviewed them. The deadline is a good reason to find out what your site says today.