How to Opt Out of AI Training Without Losing AI Search Traffic

On October 30, 2026, Google and Microsoft both changed how bloggers tell AI companies what they can do with their content. Google said it plans to honor a shared AI training opt-out in robots.txt. Bing split AI training from AI grounding and set a March 1, 2027 deadline for anyone still relying on NoArchive.

That turns one vague question (“should I block AI?”) into three separate ones: can your posts train AI models, can AI answers quote and link them, and can Google’s AI Overviews and AI Mode show them. You can opt out of AI training and keep most of your AI search visibility, as long as you use the right control for each.

Short version: add a __usage_ai-training group and a Google-Extended group to robots.txt, and leave Bing’s grounding preference and Google’s Search generative AI control at their defaults. Your pages stay eligible for Google’s AI features and Bing’s AI answers. The one thing you give up is grounding in the Gemini app. For OpenAI, add a GPTBot group as well.

What Changed on October 30, 2026

Google’s Search Central post, written by John Mueller, says Google plans to support a shared AI training control in robots.txt, based on work from the IETF’s AIPREF working group. If your robots.txt opts out of training with the shared policy token and doesn’t mention Google-Extended, Google plans to treat that as a Google-Extended disallow. The post targets 2027, and Google’s crawler documentation now says early 2027.

Microsoft’s Bing announcement goes further. Bing adopts the same shared training preference, extends it to individual pages, and adds a separate page-level grounding preference. It starts honoring both from February 1, 2027, and after March 1, 2027 it stops reading NoArchive and NoCache as AI preferences. Bing says none of this changes eligibility for regular Bing search results.

The shared signal comes from an IETF Internet-Draft posted the same day. It’s an individual draft, which means it has no formal standing in the IETF standards process yet. Bing has committed to it and Google plans to support it, which is what makes it worth setting up now.

Training, Grounding, and AI Overviews Are Separate

Most of the confusion comes from treating “AI” as one use. The new controls split it into three.

AI training means using your content to change a model’s learned parameters. A training opt-out applies to future training. It doesn’t decide whether an AI answer can cite your page today.

AI grounding means using your page as input when an AI answer is generated, usually with a link or citation back to you. Bing is explicit that indexing and ranking pages in regular search results doesn’t count as grounding.

Google’s AI features in Search (AI Overviews, AI Mode, and generative AI in Discover) get their own Search Console setting, separate from both Google-Extended and the shared training token.

This table maps each control as of October 30, 2026:

The odd one out is Google-Extended, because it bundles two uses: training for Gemini models and grounding in the Gemini app and Vertex AI. Google has no setting today that blocks training and keeps the grounding. Google’s post says sites that explicitly allow Google-Extended stay eligible to appear in Gemini app responses, so a disallow costs you that visibility.

Pick Your Setup by Goal

Keep Everything On (the Default)

Do nothing. Every one of these preferences defaults to allow, so your content stays eligible for every use covered in this post. For a blog that lives on search traffic, this is still a reasonable starting point.

Block Training but Stay in AI Answers

This is the setup most bloggers who object to training want. Add two groups to robots.txt:

# Shared training opt-out
User-Agent: __usage_ai-training
Disallow: /

# Google (Gemini)
User-agent: Google-Extended
Disallow: /

The first group is the shared signal. Bing reads it from February 1, 2027, and Google plans to in 2027. The second group works with Google today. Keep both: the shared token does nothing at Google until Google turns it on, and an explicit Google-Extended rule means Google reads your choice directly instead of inferring it. Yoast SEO and Rank Math both include a robots.txt editor, or you can edit the file at your site root.

The shared token only covers companies that adopt it, and neither announcement names any beyond Google and Microsoft. For everyone else, use each company’s own crawler name. OpenAI’s crawler docs, for example, say a GPTBot disallow tells OpenAI not to use your content to train its foundation models, while OAI-SearchBot can still crawl for ChatGPT search:

# OpenAI (training only)
User-agent: GPTBot
Disallow: /

If your blog runs behind Cloudflare, its Disallow AI Training setting (added in September 2026) publishes a no-training preference in robots.txt for you and keeps Googlebot and Bingbot crawling for search. Check your live robots.txt after turning it on so you know which lines are already there.

Leave Bing’s grounding preference unset and keep Google’s Search generative AI control on “Include.” Your pages stay eligible for Bing and Copilot answers and for Google’s AI Overviews and AI Mode.

Leave Google’s AI Overviews and AI Mode

Google’s switch is site-wide and lives in Search Console:

1. Open Search Console and pick your property. Use the top-level domain property if you have one.
2. Go to Settings, then Search generative AI.
3. Choose “Exclude my site’s links and content from Search generative AI features” and save.

According to Google’s help page, exclusion takes effect within a few days. Your pages won’t appear or be linked in AI Overviews, AI Mode, or generative AI features in Discover, and you’ll get no impressions or traffic from them. Google says the setting isn’t used as a ranking signal for the rest of Search, and it doesn’t touch AI training, which stays with Google-Extended.

Child properties inherit the setting from the closest parent that changed it. A URL-prefix property for /blog/ follows your domain property unless the /blog/ property, or a property between the two, sets its own value.

Check what you’d lose before you flip it. The Generative AI performance report in Search Console shows AI Overview and AI Mode impressions by page (how to read it next to your query data). If your best posts get a big share of their impressions there, excluding the site gives up those impressions and the clicks that come with them.

Keep Bing’s AI but Skip Third-Party APIs

Bing’s grounding preference has a middle setting called DisallowExtended. Your pages can still inform Microsoft’s own experiences (Bing’s AI Mode and AI Overview, Copilot, and Microsoft 365), but they’re kept out of grounding services Microsoft provides to other companies, such as Web IQ and the grounding API in Foundry. Grounding is page-level only, so the tag goes in each page’s head:

<meta name="Bingbot" content="ai-grounding:disallowextended">

You can send the same preference as an HTTP header instead: X-Robots-Tag: Bingbot: ai-grounding:disallowextended.

Leave Bing’s AI Answers Entirely

Use ai-grounding:disallow in the same tag or header. Bing warns that Disallow and DisallowExtended both reduce your visibility in AI answers. Neither one affects your eligibility for regular Bing results.

If You Use NoArchive, Switch Before March 1

Bing has treated NoArchive and NoCache as its AI training and grounding controls since 2023, and that is ending. If those tags are on your site, these are the dates from Bing’s announcement and its AI preferences FAQ:

• December 15, 2026: temporary grounding preferences open in Bing Webmaster Tools, for sites using NoArchive or NoCache on that date.
• February 1, 2027: Bing starts honoring the new training and grounding preferences as it crawls.
• March 1, 2027: the deadline. After this date, NoArchive and NoCache no longer count as training or grounding preferences.
• September 30, 2027: any remaining temporary grounding preference expires.

To keep a restriction, publish the new training preference (robots.txt or page tags) and the new grounding tags on your pages. Then record a temporary grounding preference in Bing Webmaster Tools before March 1. Bing asks for that step because it may not recrawl every page between February 1 and March 1. Skip it, and pages with NoArchive default to allowing grounding until Bing recrawls them.

If you added NoArchive years ago for some other reason, decide now whether you still want it. Once Bing finds one of the new preferences on your site, it ignores any NoArchive and NoCache tags left on that site.

To check whether you use them, view the source of a post and search for noarchive and nocache, then look at the response headers for an X-Robots-Tag:

curl -sI https://yourblog.com/sample-post/ | grep -i x-robots-tag

On WordPress, also look at Rank Math’s robots meta settings, which include a No Archive option both site-wide and per post.

Avoid the robots.txt Grouping Trap

The shared token borrows the User-Agent line, so normal robots.txt grouping applies. Under RFC 9309, a group is one or more user-agent lines followed by rules, and every user-agent line in the group shares those rules. That makes this edit dangerous:

User-Agent: __usage_ai-training
User-agent: *
Disallow: /

The intent was “no training,” but the asterisk line sits in the same group, so every crawler, Googlebot included, is now told to stay off the whole site. Put the policy token’s own rules directly under it, before the next User-agent line, as in the earlier example.

Two more things to know. The policy token isn’t a crawling directive, so it won’t stop any bot from fetching your pages. And Bing applies a preference only after it recrawls the page, so a change you publish today isn’t active until then. Bing says IndexNow can raise the chance of a recrawl, but it can’t promise a date.

Keeping a bot off your pages entirely is a crawl decision, covered in our guide to AI crawler access. The new preferences describe what a company may do with content it has already fetched.

Add Bing’s Page Tags in WordPress

Bing’s grounding preference has no robots.txt version, so a site-wide choice means a tag on every page. On WordPress, a few lines in a child theme’s functions.php file or a code snippets plugin will print it:

add_action( 'wp_head', function () {
	echo '<meta name="Bingbot" content="ai-grounding:disallowextended">' . PHP_EOL;
} );

Change the value to ai-grounding:disallow if you want out of Bing’s AI answers entirely, or add a second line for ai-training:disallow. To tag some posts and skip others, check the post ID or category with is_single() or in_category() before printing. Afterward, view the source of a live page to confirm the tag is there, since a caching plugin can keep serving the old HTML for a while.

A Sensible Default for Most Blogs

If your blog earns its traffic from search, AI answers are one more place readers find you, and many of them link to their sources. Opting out of grounding or AI Overviews removes you from those links, so do it only with a clear reason and numbers behind it.

Training is the easier call. Opting out costs nothing in Bing’s search or AI answers, and at Google it costs only Gemini app grounding. If you don’t want future models trained on your posts, add the two robots.txt groups (plus GPTBot for OpenAI) and move on.

Whatever you change, measure before and after. Search Console’s Generative AI performance report covers Google’s AI features, and Bing Webmaster Tools has its own AI Performance report. RightBlogger’s AI Brand Monitor checks whether ChatGPT, Gemini, Perplexity, Claude, and AI Overviews mention your site, which shows you what a setting cost.

FAQs About AI Opt-Outs

Does Blocking Google-Extended Remove You From AI Overviews?

No. Google says Google-Extended doesn’t affect inclusion in Google Search, and AI Overviews and AI Mode are Search features. They’re controlled by the Search generative AI control in Search Console.

Does NoArchive Still Block Bing’s AI?

For now. Bing stops reading NoArchive and NoCache as training or grounding preferences after March 1, 2027. It ignores them sooner on any site where it finds one of the new preferences.

Will Opting Out of AI Overviews Hurt Rankings?

Google says the control isn’t used as a ranking or inclusion signal for other parts of Search. You do lose the impressions and clicks those AI features would have sent you.

Does the AI Training Token Block Crawling?

No. The IETF draft says the policy token isn’t a crawling directive. Each bot keeps following the crawler rules you set for its own user agent.

Is the AI Training Signal an Official Standard?

Not yet. It’s an individual Internet-Draft (draft-kelley-aipref-training-preference-signal-00) with no formal standing in the IETF process. Microsoft will honor it from February 1, 2027, and Google plans to in 2027. Neither announcement names other providers that have committed.

Final Thoughts on AI Opt-Outs

Training and AI answers are now separate settings at both Google and Bing, with Gemini app grounding as the one exception. Add the training groups if you want them, leave the answer-side controls on unless your data says otherwise, and put March 1, 2027 on your calendar if NoArchive is anywhere on your site.