Publisher takeaway: On October 30, 2026, Microsoft Bing (Krishna Madhavan) announced independent preferences for AI Training vs AI Grounding while keeping traditional Bing search eligibility unchanged. Training uses the shared robots.txt User-Agent __usage_ai-training (site/path) plus optional page-level Bingbot overrides; grounding is page-level only, with Allow / Disallow / a new DisallowExtended that limits grounding to Microsoft-operated experiences. These prefs replace NoArchive/NoCache for training and grounding on a 2027 timeline — plan by March 1, 2027 if you relied on the old tags.
Primary source: the official Bing Search blog (Oct 30, 2026). Industry context: Search Engine Land and Search Engine Roundtable. This complements our morning write-up on Google/OpenAI’s shared robots.txt AI training policy token — Bing adopts that shared training signal and goes further with separate grounding controls.
Training vs Grounding: two independent levers
Microsoft is explicit: training and grounding are independent. Your training setting does not change your grounding setting, and vice versa. Both default to Allow when no applicable preference is published. Traditional Bing rankings and indexing are not flipped by these usage prefs (subject to normal crawl/index requirements). Neither Allow nor Disallow guarantees inclusion, traffic, or payment.
| Preference | Where you set it | Values | What it governs |
|---|---|---|---|
| AI Training | robots.txt (site/path) and/or page meta / X-Robots-Tag | Allow (default) / Disallow | Using content to modify learned parameters of a generative AI model (IETF-aligned training definition) |
| AI Grounding | Page-level only (no site-wide robots.txt grounding) | Allow (default) / Disallow / DisallowExtended | Using content as input to help inform AI-generated responses (not traditional search ranking) |
A practical combo Microsoft highlights: set Training Disallow + Grounding Allow so pages stay eligible to inform Copilot / Bing AI answers (with citations) without feeding generative model training.
AI Training: shared robots.txt + Bing page overrides
For site or path training prefs, Bing uses the same shared policy token industry partners are adopting:
User-Agent: __usage_ai-training
Disallow: /
Path-level example:
User-Agent: __usage_ai-training
Disallow: /example-path/
Allow is the default — you do not need a robots.txt group to allow training. An optional explicit Allow is also documented. Microsoft extends the same preference to the page level so a single URL can override robots.txt:
<meta name="Bingbot" content="ai-training:disallow">X-Robots-Tag: Bingbot: ai-training:disallow
When Microsoft detects the new AI Training or AI Grounding prefs on a site, it will disregard remaining NoArchive/NoCache prefs on that site for those uses. Page-level training overrides robots.txt for that page.
AI Grounding: page-level Allow, Disallow, DisallowExtended
There is no site-wide robots.txt grounding preference. Publish grounding per page via meta or HTTP header:
<meta name="Bingbot" content="ai-grounding:disallow">(orai-grounding:disallowextended/ allow)X-Robots-Tag: Bingbot: ai-grounding:disallow(same value variants)
| Grounding value | Effect |
|---|---|
| Allow (default) | Content may help inform generated responses |
| DisallowExtended | Grounding limited to Microsoft-operated experiences (AI Mode, AI Overview, Copilot, Microsoft 365). Excludes third-party grounding Microsoft provides (e.g. Web IQ, Foundry grounding API) |
| Disallow | Content is not used to help inform generated responses |
Microsoft warns that Disallow or DisallowExtended will significantly reduce AI visibility without affecting Bing search eligibility. Indexing/ranking in classic results is not grounding. Bing’s traditional search syndication API (links for classic search) is also described as out of scope for the grounding preference.
Timeline: Dec 2026 → Sep 2027
| Date | What happens |
|---|---|
| Dec 15, 2026 | Temporary grounding transition prefs open in Bing Webmaster Tools — only for sites using NoArchive/NoCache as of that date |
| Feb 1, 2027 | Microsoft begins honoring new AI Training / AI Grounding prefs as it crawls |
| March 1, 2027 | Deadline: after this, NoArchive/NoCache are no longer read as training/grounding prefs. NoArchive/NoCache users who want grounding limits must (1) publish page grounding prefs and (2) record a temporary preference in BWT |
| Sep 30, 2027 | Temporary BWT grounding prefs expire (or earlier when Microsoft recrawls and discovers the new on-page pref) |
Training has no temporary BWT setting — publish robots.txt or page training prefs by March 1, 2027. If you skip the temporary grounding record by March 1, pages that only had NoArchive/NoCache default to Allow for grounding until recrawl discovers a new preference.
What this means vs Google’s shared training token
Same day (Oct 30), Google and OpenAI announced plans around the shared __usage_ai-training token — covered in our policy token guide. Bing’s post confirms Microsoft is participating in that shared training preference and shipping a separate Microsoft grounding control with DisallowExtended. Do not collapse:
- Shared training token — cross-provider training preference in robots.txt (Bing + Google/OpenAI direction).
- Bing page training override — Bingbot meta / X-Robots-Tag for one URL.
- Bing grounding — Microsoft-only page prefs; not a Google Search Console generative AI control equivalent.
- Classic Bingbot crawl Disallow — still about crawl access; usage Allow does not override a robots.txt block on Bingbot.
Update mental models from our older AI crawlers allow/block cheat sheet: treating noarchive/nocache as the durable Copilot / Bing AI usage switch is being retired. Migrate to the new training + grounding prefs on the calendar above.
Activation caveats publishers miss
- Publishing ≠ activation. Microsoft must crawl the page to discover the setting.
- IndexNow can signal change and may improve recrawl chance — it does not guarantee a date.
- Defaults matter. Sites that never set NoArchive/NoCache and never publish new prefs stay Allow/Allow.
- NoArchive/NoCache users should not assume old restrictions carry over after March 1, 2027.
Publisher checklist (Oct 31, 2026)
- Inventory pages/templates that emit NoArchive or NoCache today — those are your migration set.
- Decide Training vs Grounding independently (e.g. Training Disallow + Grounding Allow for citation eligibility without training).
- Publish training via
__usage_ai-trainingin robots.txt and/or Bingbot page overrides where needed. - Publish grounding page-by-page (Allow / Disallow / DisallowExtended) — there is no robots.txt grounding group.
- If you use NoArchive/NoCache: calendar Dec 15, 2026 (BWT temporary prefs), Feb 1, 2027 (honor-as-crawled), March 1, 2027 (hard deadline), Sep 30, 2027 (temp expiry).
- Use IndexNow after publishing prefs to improve odds of timely recrawl — still verify in Bing Webmaster Tools.
- Measure AI visibility separately from classic SEO — see tracking AI referral traffic in GA4 (Bing AI Performance reporting is a useful companion surface) and keep Google halo/CTR context from the Shepard opt-out notes.
- Do not confuse with Google’s GSC generative AI Include/Exclude — different product; see our GSC control guide when you manage Google Search generative visibility.
FAQ: Bing AI Training vs Grounding controls
Does Disallow grounding hurt Bing search rankings?
No — Microsoft states these usage preferences do not change eligibility for traditional Bing search results (normal crawl and indexing rules still apply). Disallow / DisallowExtended mainly reduce visibility in AI-generated experiences.
What does DisallowExtended actually block?
It limits grounding to Microsoft-operated products such as AI Mode, AI Overview, Copilot, and Microsoft 365, and excludes grounding services Microsoft provides to third parties (examples cited: Web IQ and Foundry grounding API access).
Can I set grounding once in robots.txt for the whole site?
No. AI Grounding is page-level only. AI Training can be site/path in robots.txt via __usage_ai-training, with optional per-page Bingbot overrides.
We already use NoArchive / NoCache for Copilot — are we done?
Not for the long term. New prefs start being honored as crawled from Feb 1, 2027. After March 1, 2027, NoArchive/NoCache are no longer read as training/grounding prefs. Sites on those tags as of Dec 15, 2026 should also record temporary grounding prefs in Bing Webmaster Tools and publish new page prefs.
Is publishing a meta tag enough?
It is necessary but not sufficient until crawl. Microsoft must discover the setting; IndexNow may help recrawl odds but does not guarantee timing. Temporary BWT prefs bridge NoArchive/NoCache sites until recrawl or Sep 30, 2027.
How does this relate to Google’s shared __usage_ai-training token?
Bing is adopting the shared training preference in robots.txt (same industry direction as Google/OpenAI). Separately, Bing adds Microsoft-specific grounding controls and page-level training overrides. Read both this guide and our Google/OpenAI policy token article.
Bottom line: Bing’s Oct 30, 2026 controls split AI Training (shared __usage_ai-training + Bing page overrides) from AI Grounding (page-only Allow / Disallow / DisallowExtended). Classic Bing rankings stay separate; AI visibility shrinks if you Disallow grounding. Migrate off NoArchive/NoCache before March 1, 2027, and treat crawl discovery as part of activation. For adjacent publisher tooling, keep the AI crawler cheat sheet, shared training token notes, and getting cited in AI search in the same ops doc.
Leave a Reply
You must be logged in to post a comment.