Microsoft Bing introduced three publisher settings on Friday, October 30, 2026, that treat AI model training and AI answer grounding as separate decisions. Grounding means a page’s content helps produce an AI-generated answer. According to the announcement, classic Bing rankings are untouched by whichever choice a site makes.
Krishna Madhavan, Principal Product Manager at Microsoft Bing, announced the change, and Barry Schwartz of Search Engine Roundtable reported on it. Madhavan described the preferences as simpler and clearer than before. A site can choose training, grounding, or both while still qualifying to appear in Bing’s results.
Three settings define the new model:
- AI Training lets a site permit or refuse the use of its content in model training.
- AI Grounding controls whether its content can feed the answers AI experiences generate.
- Limited Grounding lets Microsoft’s own AI products draw on the content but withholds it from third-party grounding services.
The settings can be mixed. Search Engine Roundtable offers an example: a publisher keeps Microsoft from training models on its articles, yet Copilot may still quote and link them. The reverse choice carries a cost: blocking grounding can reduce a site’s visibility in AI experiences. Neither Microsoft’s announcement nor the coverage of it quantifies that loss.
Microsoft published a staged timeline for sites that currently depend on the legacy NoArchive or NoCache tags:
- December 15, 2026: Bing Webmaster Tools offers temporary grounding controls to eligible sites.
- February 1, 2027: Microsoft starts honoring the new preferences when it crawls pages.
- March 1, 2027: Publishers on the old directives must configure the new preferences to keep their restrictions.
- September 30, 2027: The temporary grounding preferences expire.
Schwartz wrote that the changes appear to build on controls Bing already had over how site content surfaces in machine-written answers. Separately, Bing Webmaster Tools includes an AI Performance report. It lists which URLs get cited and the queries used for grounding, across the AI experiences Bing supports. That report is the closest thing publishers have to a baseline before they change anything.
The announcement also fits a wider move. Microsoft said it is adopting the common robots.txt signal for training that its peers announced, and Google, OpenAI, and Anthropic have backed the same approach. Microsoft goes further: training choices can also be set page by page, and grounding gets its own switch.
The operational risk sits in the migration. A site that has leaned on NoArchive or NoCache to hold back AI use of its pages has until March 1, 2027 to restate that intent in the new settings. Missing the date means those restrictions are not preserved, per the timeline Microsoft published. Teams with large archives or many templates will want to know which pages carry the old directives well before then.
The announcement as reported contains no independent measurement of what any combination does to citations or traffic. Microsoft is describing its own controls and its own products, so the effect of each setting remains a claim until sites test it. That makes a pre-change record of current citations more valuable than the settings themselves.
Before December 15, pull the AI Performance report for your top pages and note which are cited today. Then decide per section of the site whether training, grounding, or both should stay open, and schedule the configuration work ahead of the March 1 deadline.
Based on Microsoft Bing’s announcement by Krishna Madhavan on the Bing blog; Search Engine Roundtable (Barry Schwartz) covered it on October 30, 2026.
