Microsoft says NoArchive and NoCache stop counting as AI preferences after March 1, 2027. How do I keep my pages out of Copilot and Bing's AI answers without dropping out of Bing search, and what do the new ai-grounding and ai-training settings actually do?

What should I check, step by step?

1. Decide training and grounding separately. Microsoft treats them as independent: blocking training doesn't keep a page out of Copilot answers, and blocking grounding doesn't stop training. Neither one is a search setting.

2. Leave Bingbot allowed in robots.txt. Microsoft says an Allow preference doesn't override a robots.txt rule that blocks Bingbot, and blocking the crawler or adding noindex takes you out of Bing search, which is more than the AI settings ask for.
3. To keep a page out of AI answers, add <meta name="Bingbot" content="ai-grounding:disallow"> to its head, or send X-Robots-Tag: Bingbot: ai-grounding:disallow as a response header (the header form also covers PDFs and other non-HTML files). There is no site-wide robots.txt grounding setting, so it has to go on every page you want excluded, usually through a template.
4. If you want to stay in Copilot and Bing's own AI answers but not be fed to other companies' AI apps through Microsoft's grounding services, use ai-grounding:disallowextended instead.
5. To opt out of training for the whole site, add a group with "User-Agent: __usage_ai-training" and "Disallow: /" (or Disallow lines for specific paths). For one page, use <meta name="Bingbot" content="ai-training:disallow">; Microsoft says the page setting overrides robots.txt.
6. If your templates carry noarchive or nocache today, map each to what you meant. Under the 2023 rules, NOARCHIVE kept a page out of Chat and Copilot answers and out of training; NOCACHE allowed it in answers showing only the URL, title and snippet; with both, Bing treated the page as NOCACHE. Write the matching ai-grounding and ai-training values; don't assume the old tags carry over.
7. Sites using NoArchive or NoCache on December 15, 2026 can record a temporary grounding preference in Bing Webmaster Tools. Do it before March 1, 2027: without it, those pages default to Allow for grounding until Bing recrawls them and finds the new tag. A temporary preference ends at recrawl or on September 30, 2027, whichever comes first.
8. Publishing a tag doesn't activate it; Bing has to recrawl the page. Microsoft says submitting changed URLs with IndexNow raises the chance of a recrawl but doesn't guarantee a date.
9. Expect a cost if you disallow grounding. Microsoft says it significantly reduces your visibility in AI-generated experiences, and that allowing it makes your pages eligible for answers with links and citations. It also says neither choice guarantees inclusion, traffic or payment.
10. Keep a dated record of what you published and when: the robots.txt version, the template change and the Bing Webmaster Tools entry. If a page shows up in an answer after you opted out, that record is what you can point to.

What is the difference between Bing's ai-training and ai-grounding preferences?

AI Training covers using a page to change the learned parameters of a generative AI model. Microsoft adopts the shared robots.txt syntax from the IETF AI Preferences work ("User-Agent: __usage_ai-training" with Allow or Disallow lines) for a site or path, and adds a page-level form, ai-training:disallow in a Bingbot meta tag or X-Robots-Tag header, that overrides robots.txt for that page.

• AI Grounding covers using a page as input to help inform an AI-generated response. It's Microsoft's own preference, set only per page, with three values: Allow (the default), DisallowExtended (Microsoft-operated experiences such as AI Mode, AI Overview, Copilot and Microsoft 365 only, excluding grounding Microsoft provides to third parties such as Web IQ and grounding through API access in Foundry) and Disallow (not used to inform generated responses).
• Microsoft says indexing and ranking pages in traditional results isn't grounding, so these settings don't change whether a page can appear as an ordinary Bing result. It also says the current Bing search syndication API returns links for traditional search, isn't meant for AI summaries and isn't affected by the grounding preference. The preferences don't create any license or payment arrangement.

Will Bing still honor noarchive and nocache for Copilot and AI training after 2027?

Microsoft introduced NOARCHIVE and NOCACHE as Bing Chat controls in September 2023, and its robots meta tag page still describes them that way: NOARCHIVE means don't link in Chat and Copilot and don't use the content for training; NOCACHE means show only the URL, snippet and title in Chat or Copilot and train only on those; both together are treated as NOCACHE.

• The new preferences replace them. From February 1, 2027 Microsoft honors the new settings as it crawls, and once it detects a new AI Training or AI Grounding preference on a site, it disregards any NoArchive or NoCache left there. After March 1, 2027 the old tags aren't read as a training or grounding preference at all.
• The transition window covers grounding only. From December 15, 2026, sites that use NoArchive or NoCache on that date can record a temporary grounding preference in Bing Webmaster Tools, because Microsoft may not recrawl every page between February 1 and March 1. There is no temporary training setting: publish the training preference in robots.txt or on the pages before March 1, 2027.

What do you lose by blocking Bingbot or disallowing AI grounding, and does it affect Bing search?

Blocking Bingbot or adding noindex: you leave Bing search, and Copilot's web answers draw on what Bing has indexed, so you also lose any chance of being cited. ai-grounding:disallow: you stay eligible for Bing results, but Microsoft warns your visibility in AI experiences drops significantly. disallowextended: you keep Microsoft's own answers and drop only the grounding it supplies to other companies' AI products. ai-training:disallow: no effect on search or grounding.

• name="Bingbot" scopes a tag to Bing. Microsoft's examples use the Bingbot name for the new values. A generic name="robots" nosnippet or noarchive also affects other search engines, and noarchive doesn't mean the same thing everywhere, so keep the Bing-only form when Bing is what you mean.
• The __usage_ai-training group isn't specific to Microsoft: Google, OpenAI and Anthropic announced support on the same day, so a whole-site Disallow there is a training opt-out aimed at every company that reads it. There's no shared equivalent for grounding; each company has its own control (Google's is in Search Console, Apple's is nosnippet).