---
title: "Bing AI Grounding vs AI Training: How to Set Each"
description: "Microsoft split Bing AI use into AI Training and AI Grounding. See the syntax, defaults and dates, and what to do if you use noarchive or nocache."
url: https://seenlens.com/guides/bing-ai-grounding-and-training
last_updated: 2026-11-06
---
# Bing AI grounding and AI training: how to set each preference
On 2026-10-30 Microsoft announced two separate Bing preferences: AI Training and AI Grounding. Most sites need to do nothing, because Allow is the default for both. Sites that use `noarchive` or `nocache` to keep content out of Bing and Copilot AI answers must publish the new preferences before 2027-03-01, when Bing stops honoring those two directives for AI use.
By SeenLens. Last updated: 2026-11-06
## What are the two Bing AI preferences?
AI Training covers the use of your content to train generative models. AI Grounding covers the use of your content as input that helps inform AI generated responses in Bing, Copilot and partner integrations. The two preferences are independent. Limiting either one does not change whether your pages are eligible for Bing search results.
| | AI Training | AI Grounding |
| --- | --- | --- |
| What it controls | Use of content to train generative models | Use of content as input to help inform AI generated responses |
| Where to set it | robots.txt for a site or path, or page level | Page level only: meta tag or HTTP header. Not in robots.txt |
| Options | Allow, Disallow | Allow, DisallowExtended, Disallow |
| Default | Allow | Allow |
| Page level syntax | `ai-training:disallow` | `ai-grounding:disallowextended` or `ai-grounding:disallow` |
`DisallowExtended` limits grounding to Microsoft operated experiences. It excludes partner use, such as grounding API access in Microsoft Foundry and Web IQ. `Disallow` means the content cannot be used to help inform generated responses. For the term itself, see [grounding queries](https://seenlens.com/glossary/grounding-queries).
## When do the new preferences apply?
| Date | What happens |
| --- | --- |
| 2026-10-30 | Microsoft announces the preferences. Nothing changes that day. |
| 2026-12-15 | Temporary grounding transition preferences open in Bing Webmaster Tools. They are only for sites that use NoArchive or NoCache. |
| 2027-02-01 | Microsoft starts honoring AI Training and AI Grounding preferences as pages are recrawled. |
| 2027-03-01 | NoArchive and NoCache are no longer honored as training or grounding preferences. |
| 2027-09-30 | Temporary preferences expire, or at recrawl, whichever is earlier. |
Before the new rules apply, Bing has used `noarchive` and `nocache` to limit use of content in its AI answers and training. Bing Webmaster Tools has no temporary training setting. A site that wants to keep content out of AI training must publish the new training preference itself.
## What should I set for my goal?
| If your goal is | Set |
| --- | --- |
| Stay available for training and AI answers | Nothing. Allow is the default. You can state it in robots.txt. |
| Opt out of training for the whole site | `User-agent: __usage_ai-training` with `Disallow: /` in robots.txt |
| Opt out of training for one folder | `User-agent: __usage_ai-training` with `Disallow: /folder/` in robots.txt |
| Opt out of training for one page | `` or the matching header |
| Keep content in Microsoft products only | `ai-grounding:disallowextended` on the page |
| Keep content out of Bing and Copilot answers | `ai-grounding:disallow` on the page |
| You use noarchive or nocache for AI limits | Publish `ai-grounding` or `ai-training` before 2027-03-01 |
| You use noarchive or nocache for another reason | Nothing for AI. After 2027-03-01 Bing no longer reads them as AI preferences. |
A page level AI Training preference overrides the robots.txt preference for that page. An Allow does not override a robots.txt rule that blocks Bingbot from crawling. If you block Bingbot, Bing cannot read the page or its tags. See [Bingbot](https://seenlens.com/bots/bingbot) and the [robots.txt guide for AI crawlers](https://seenlens.com/guides/robots-txt-for-ai-crawlers).
## What is the exact syntax?
AI Training uses the shared industry token from IETF Internet-Draft draft-kelley-aipref-training-preference-signal. Microsoft extends it to page level with the Bingbot name.
```text
User-Agent: __usage_ai-training
Disallow: /
# or only one path
User-Agent: __usage_ai-training
Disallow: /example-path/
```
```text
```
```text
X-Robots-Tag: Bingbot: ai-training:disallow
X-Robots-Tag: Bingbot: ai-grounding:disallowextended
X-Robots-Tag: Bingbot: ai-grounding:disallow
```
Use the header for files that have no HTML head, such as PDFs. The robots.txt token is the only place where a site wide training preference lives. AI Grounding cannot be set in robots.txt.
## How does this relate to Content-Signal?
`Content-Signal: ai-train=no` is a robots.txt line that Cloudflare introduced. Bing documents the `__usage_ai-training` token and does not document Content-Signal. A site can publish both. This is what a robots.txt looks like when a site allows everything and states the training preference explicitly:
```text
User-agent: *
Content-Signal: search=yes, ai-input=yes, ai-train=yes
Allow: /
User-agent: __usage_ai-training
Allow: /
```
Google announced on 2026-10-30 that it intends to support the `__usage_ai-training` token in early 2027. When a site opts out of training with it and has no Google-Extended group, Google plans to treat that as a Google-Extended disallow. Until then, Google-Extended is the control Google reads today. For the Content-Signal line, see the [Content Signals guide](https://seenlens.com/guides/content-signals-robots-txt) and the [fix for adding Content Signals](https://seenlens.com/fixes/add-content-signals).
## What does SeenLens check?
The scan reads ``, ``, the `X-Robots-Tag` header and the `__usage_ai-training` robots.txt group. It fails a page with `ai-grounding:disallow`, warns on `noarchive` or `nocache`, and notes `ai-grounding:disallowextended` as information. A training opt-out alone passes. Details and fixes are on the [Bing AI grounding blocked fix page](https://seenlens.com/fixes/bing-ai-grounding-blocked). The scan is free at [the SeenLens home page](https://seenlens.com/). To see how Bing reports Copilot citations, read the [Bing AI performance report guide](https://seenlens.com/guides/bing-ai-performance-report). A `noindex` tag is a separate problem; see [noindex blocking AI answers](https://seenlens.com/fixes/noindex-blocks-ai-answers).
[Scan your page with SeenLens](https://seenlens.com/)
## Frequently asked questions
### What does ai-grounding:disallow do?
It tells Bing that the page content cannot be used to help inform AI generated responses. Copilot and Bing AI answers then do not use the page as input, so they do not cite it from that use.
### Does ai-grounding:disallow remove my page from Bing search?
No. Microsoft says limiting training or grounding does not change eligibility for Bing search results.
### Can I set AI Grounding in robots.txt?
No. AI Grounding is page level only: a meta tag or an HTTP header. Only AI Training has a robots.txt form.
### What does the __usage_ai-training token do?
It is a robots.txt group name from an IETF Internet-Draft. A `Disallow` under it states a training opt-out for the site or path. Bing documents it and starts honoring it on 2027-02-01. Google says it intends to support it in early 2027 and treat an opt-out as a Google-Extended disallow; until then, use Google-Extended for Google.
### Is noarchive still honored by Bing for Copilot?
Until 2027-03-01, yes. Bing treats `noarchive` and `nocache` as AI usage controls before the new rules apply. After that date they are no longer honored as training or grounding preferences.
### Do I need to change anything today?
Only if you use `noarchive` or `nocache` to keep content out of AI answers or training. Publish the new preferences before 2027-03-01. Every other site can keep the default, Allow.
## Sources
- [Bing Webmaster Tools help: New AI Preferences](https://www.bing.com/webmasters/help/new-ai-preferences-2t1cc16d) (options, defaults, syntax, precedence, timeline; checked 2026-11-06)
- [Bing Blog: Your content, your choice, clearer preferences for training and grounding](https://blogs.bing.com/search/2026/10/your-content-your-choice-clearer-preferences-for-training-and-grounding/) (announcement of 2026-10-30; checked 2026-11-06)
- [IETF Internet-Draft draft-kelley-aipref-training-preference-signal](https://datatracker.ietf.org/doc/draft-kelley-aipref-training-preference-signal/) (the `__usage_ai-training` token; checked 2026-11-06)
- [Google Search Central Blog: Supporting a shared AI training control for websites](https://developers.google.com/search/blog/2026/10/ai-training-control) (planned support, Google-Extended mapping; checked 2026-11-06)
- [Google: common crawlers](https://developers.google.com/crawling/docs/crawlers-fetchers/google-common-crawlers) (Google-Extended, AIPREF support planned for early 2027; checked 2026-11-06)