Yes, you can. Bing still sends you customers while Microsoft's AI is told to leave your pages alone, as long as ordinary Bingbot crawling and indexing stay switched on. You may want to block Bing AI from using your website without losing a single search visitor, and Microsoft now documents settings built for exactly that.
The trouble is that the settings sound alike and do very different jobs. One rule can keep Microsoft from training its models on your pages. Another decides whether Copilot and Bing's AI answers can draw on a page. A third kind of rule can take you out of Bing search altogether, which is rarely what an owner wants.
Pick the outcome first, then the control that matches it. Once you have, the exact lines to use are short, and a ten-minute check on your own site shows where you stand today.
Key Takeaways
Microsoft treats AI Training and AI Grounding as separate preferences, and both are separate from search indexing, so you can limit either one and stay in Bing search.
AI Training can go in robots.txt under a `__usage_ai-training` group, but AI Grounding can only be set page by page with a Bingbot meta tag or an HTTP header.
It removes a page from Bing search, and a Bingbot block cuts off crawling, so neither one works as an AI-only opt-out.
Which Bing AI setting matches what you want?
Start by writing down which of four outcomes you are after. Each one maps to a different control, and only the last one takes you out of Bing search. Microsoft's AI preferences help page documents the training and grounding controls, and this table pairs each one with a common outcome.
| What you want | Control to use | Where it goes |
|---|---|---|
| Stay in Bing search and allow AI answers | No AI preference needed, or `allow` on grounding | Nothing to add |
| Stay in Bing search, block AI answers from using the page | `ai-grounding:disallow` | Bingbot meta tag or HTTP header, per page |
| Stay in Bing search, block training, allow grounding | `__usage_ai-training` group with `Disallow: /` | robots.txt, or the page-level training tag |
| Come out of Bing search entirely | Bing's removal tools or NOINDEX | Separate job from AI preferences |
If you want both training and AI answers blocked, you combine the second and third rows. Nothing about that combination touches search indexing, so your pages stay eligible for the regular blue-link results.
Removing a page from search is a different job from setting an AI preference. It uses Bing's content removal tools or a NOINDEX tag, and it takes the page out of search for everyone, not only for AI.
What do AI Training and AI Grounding each change?
AI Training controls whether Microsoft may use your page content to train generative models. AI Grounding controls whether your page content may help generate answers in Microsoft experiences such as Copilot, Bing AI Mode, Bing AI Overviews and Microsoft 365.
Those two uses are independent of each other. You can allow grounding and block training, block grounding and allow training, or block both. Microsoft documents them as separate preferences, so a setting for one does nothing to the other.
Neither one is a search setting. Microsoft says blocking training or grounding does not automatically remove a page from Bing search. It also says an `Allow` setting does not override a robots.txt rule that blocks Bingbot, so a page Bing cannot crawl stays unreachable no matter what the AI preference says.
Grounding is the one that matters if your worry is Copilot quoting you. Microsoft's Copilot web search documentation says Copilot sends generated queries to Bing, receives web results and composes a grounded answer with citations that point back to source pages. Public website grounding uses content Bing has indexed.
How do you set each control without leaving Bing search?
For training across the whole site, add a special group to the robots.txt file at the root of your domain. Microsoft names the group `__usage_ai-training`, and the rule under it works like any other robots.txt rule.
``` User-Agent: __usage_ai-training Disallow: / ```
That group is separate from your normal Bingbot rules. Leave your regular Bingbot access alone, and Bing search keeps working. You can also point the rule at a single folder instead of the whole site if only part of it should be off limits.
For one page, use a Bingbot meta tag inside the page head. The training version looks like this, and an `X-Robots-Tag` HTTP header can do the same job.
``` <meta name="Bingbot" content="ai-training:disallow"> ```
Grounding has no robots.txt version. It must be set on each page with a meta tag or a matching header, and its values are `allow`, `disallowextended` and `disallow`.
``` <meta name="Bingbot" content="ai-grounding:disallow"> ```
Microsoft says a page-level preference overrides the robots.txt training preference for that page. So, if your robots.txt blocks training but one page carries a tag that allows it, the page tag wins. A control takes effect when Microsoft recrawls the page, and the help page lists the date the new preferences begin taking effect, so read the current date there before you plan around a timeline.
On a site with a content management system, the tag usually goes into the template that writes the page head. Set it once in the template, and every page built from it carries the tag.
What did NOCACHE and NOARCHIVE do, and do they still apply?
Microsoft announced NOCACHE and NOARCHIVE for Bing Chat in a September 22, 2023 post on the Bing Webmaster Blog. They were the first controls over how Bing Chat used a page.
NOCACHE let Bing Chat use only the URL, the title and the snippet of a page. NOARCHIVE excluded the page content from Bing Chat answers and from training. Microsoft stated that pages carrying either tag could still appear in Bing search results.
If an old NOARCHIVE or NOCACHE tag is sitting in your templates, it was written for the 2023 system. The newer training and grounding preferences let you say what you want more precisely, so map each old tag to the outcome you actually intend. NOARCHIVE also has no single meaning across every company, because other platforms read it their own way.
| Tag or preference | What Microsoft documents it as | Effect on Bing search |
|---|---|---|
| NOCACHE (2023) | Bing Chat uses only URL, title and snippet | Page can stay visible |
| NOARCHIVE (2023) | Content excluded from Bing Chat answers and training | Page can stay visible |
| `ai-training:disallow` | No use of the page for training generative models | Not removed automatically |
| `ai-grounding:disallow` | No use of the page to inform AI answers | Not removed automatically |
If you have both old and new tags on the same page, decide which result you want and remove the one that contradicts it. Mixed signals make the outcome harder to predict and harder to explain later.
Does a Bingbot block or NOINDEX work as an AI opt-out?
Neither is built for that job, and both can cost you Bing visitors. A robots.txt rule that blocks Bingbot stops crawling, so the page can fall out of search. NOINDEX tells Bing not to index the page, so it leaves search results too.
A line of `Disallow: /` is also easy to misread. Under the `__usage_ai-training` group it limits training use. Placed under a normal Bingbot group, the same line shuts Bingbot out of your whole site. The wording is identical and the result is not, so check which group the line sits under.
If you want to stay in Bing, keep ordinary Bingbot access open and use the training and grounding preferences for the AI side. Microsoft's webmaster guidelines cover the crawl and indexing basics that your AI choices sit on top of.
Compare this with the Google side of the same question. Google runs its own controls, and our post on how Google AI training is controlled through robots.txt walks through those separately. A Bing setting does not change what Google does, and the reverse holds too.
What can you check in ten minutes?
You do not need a tool to see what your site publishes. A browser and a few minutes will do, and the checks are the same whether you plan to change anything or only want to know where you stand.
Ten-minute check
Visit yourdomain.com/robots.txt and look for a `__usage_ai-training` group, then note whether it holds `Disallow: /`, a narrower path rule or an explicit `Allow`.
Open a typical page's source and search for `ai-training`, `ai-grounding`, `noarchive`, `nocache` and `noindex`.
Check the response headers for an `X-Robots-Tag` line, since a header can say something different from the page.
Make sure no rule shuts out ordinary Bingbot access by accident.
Review the crawl information and the AI Performance report where your account shows them.
A clean check tells you what is published on your site right now. It does not mean every Microsoft system has already recrawled every page, so give a change time to be picked up before you judge it.
Test a few different page types, not only the home page. A site with separate templates for services, blog posts and product pages can carry a tag on one and miss it on another.
Will allowing Bing's AI answers bring citations, or cost you visits?
Allowing grounding makes a page eligible to inform an answer. It does not guarantee that any answer will cite you. Microsoft's Copilot Studio guidance on public websites describes retrieval, relevance checks and citations as steps an answer passes through, and indexing only gets a page into the pool.
Content that is built on the fly or changes often may appear incomplete or out of date in a grounded answer. If your prices, hours or availability change often, keep them in plain text on the page where Bing can read them, and keep your important facts consistent across your site.
Nobody can tell you in advance whether letting AI answers use your pages will send more visitors or fewer. Make the choice on what you want Microsoft to do with your content, then watch your own numbers. Our post on whether Bing is worth it for a small business helps you judge how much of your traffic is at stake, and how ChatGPT and Google AI read your website covers the rest beyond Microsoft.
The IETF is also working on a shared vocabulary for AI usage preferences with categories such as AI Training, AI Use and Search. Its draft vocabulary is a working document without Working Group consensus, so treat it as a sign of where standards may go and not as a rule any crawler must follow.
When is it worth paying someone to set this up?
If your site has one simple robots.txt file and no conflicting page tags, you can make this change yourself in an afternoon. You do not need a paid package to add a `__usage_ai-training` group or a single meta tag.
Outside help earns its fee when the situation is messier. That includes several page templates, a CMS that writes tags automatically, custom HTTP headers, old NOARCHIVE or NOCACHE rules, content built on the fly, more than one domain, or a need for a written record of the decision.
What that help sells is review, implementation, testing and documentation. Someone maps your intended outcome to the right control, checks that Bing search access stays open, resolves clashes between robots.txt, page tags and headers, tests a sample of pages, and watches the result after Bing recrawls. Anyone who promises a citation, a ranking or a traffic result is promising something Microsoft does not offer.
Which Bing AI control fits?
Pick an answer to begin.
1. You want to stay in Bing search but stop Copilot from using a page to inform answers. What do you add?
2. Where can AI Grounding be set?
3. A page carries a NOINDEX tag meant to block AI. What happens to its Bing search eligibility?
Frequently Asked Questions About Keeping Your Site Out of Bing AI Answers
Can I block Bing AI and stay in Bing search?
Yes, as long as ordinary Bingbot crawling and indexing stay allowed. AI Training and AI Grounding preferences are separate from search indexing.
Does robots.txt control AI grounding?
No. Microsoft documents AI Training in robots.txt, but AI Grounding is set only per page with a meta tag or an HTTP header.
What does ai-training:disallow do?
It tells Microsoft not to use that page to train generative models, under the preference system Microsoft documents.
What does ai-grounding:disallow do?
It tells Microsoft not to use the page to inform Copilot and other Microsoft AI answers. The page can still be indexed for regular search.
Does NOARCHIVE stop Bing AI training?
Under Microsoft's 2023 Bing Chat announcement, NOARCHIVE excluded content from Bing Chat answers and training. The newer training and grounding preferences let you state each choice separately.
Does allowing grounding guarantee a citation?
No. It makes a page eligible, and a citation is never guaranteed.
What This Means for You
Pick the result you want before touching a file. If you want Bing search and no AI use, set the training preference in robots.txt or on the page, set the grounding preference on each page, and leave ordinary Bingbot access open. If you want to leave Bing search, that is a removal job with its own tools.
Run the ten-minute check first, because the answer to your question may already be sitting in your files. A clean setup keeps your Bing customers coming, keeps your content under your own rules, and gives you a record of what you chose and why.
If your site is more complicated than one robots.txt file, Web Leveling can review it, set the right controls and test them, through our generative engine optimization work. We cannot promise a citation or a traffic result, and we will not pretend otherwise. We work with small and medium businesses across the country and overseas. Tell us which pages and templates need a closer look, and we will map each one to the right Microsoft control.
Terms
Bing AI words in this post
Tap a term to see what it means.
AI Training. Microsoft's use of page content to train generative models, which you can limit in robots.txt or per page.
AI Grounding. The use of page content to help produce answers in Copilot and Bing AI experiences, set per page only.
Bingbot. Microsoft's crawler that visits pages so they can appear in Bing search.
robots.txt. A text file at the root of a domain that tells crawlers which paths they may visit.
X-Robots-Tag. An HTTP response header that carries crawl and indexing instructions outside the page itself.
NOINDEX. A directive that tells search engines not to list a page in results.
