Does the noindex meta tag stop AI training?

No. None of the major AI crawlers document the meta robots tag as an opt-out for training, and a real question asking exactly this sat unanswered on Stack Overflow for nearly three years. Robots.txt is what they actually check.

What the meta tag was built for

The meta robots tag is a search-indexing directive:

<meta name="robots" content="noindex">

Google and Bing use it to decide whether a page can appear in search results. It says nothing about crawling for other purposes, and no current documentation from OpenAI, Google or Anthropic extends it to cover AI training.

What the AI crawlers actually check

I checked each vendor's own current documentation rather than assume. OpenAI's page for GPTBot, OAI-SearchBot and ChatGPT-User says these crawlers are managed through robots.txt; it does not mention a meta tag anywhere. Google's crawler documentation is the same shape: Google-Extended is a robots.txt token with no crawler of its own, and Google has never documented a meta tag equivalent for it. Wanting a per-page or per-account switch instead of a whole-domain robots.txt rule is a reasonable ask, since robots.txt only groups rules by crawler, not by page owner or by path unless you write it that way. But none of the crawlers that matter for training or citations read a meta tag for it today.

The noai tag going around SEO blogs

A number of SEO articles this year describe a tag like this, sometimes with a made-up per-vendor variant named after Google-Extended, as if setting it opts a page out of AI training:

<meta name="noai" content="noai, noimageai">

I could not find this documented as respected by GPTBot, Google-Extended, ClaudeBot or PerplexityBot on any of their own current pages. It traces back to an opt-out DeviantArt proposed in 2022 for image-generation training, aimed at a different set of tools entirely. Adding it to a page costs nothing, but there is no evidence it does anything for the crawlers this site is about, and I am not going to tell you it works without that evidence.

What is actually coming

Two real mechanisms exist for finer-grained control than a domain-wide robots.txt block, and neither is a meta tag. Cloudflare's Content Signals policy, launched in September 2025, adds a line inside a robots.txt user-agent group:

User-agent: *
Content-Signal: search=yes, ai-train=no

That is still grouped by crawler, not by page or account, and Cloudflare's own announcement says plainly that some crawlers may simply ignore it. Separately, the IETF's AI Preferences working group has an active draft, dated August 2026, defining a Content-Usage HTTP response header:

Content-Usage: train-ai=n

alongside a path-scoped robots.txt rule for the same vocabulary. A header set per response is the first mechanism here that could actually vary by page or by account, since a CMS could emit a different value for each user's content. It is still a draft on the way to becoming a standard, not something any crawler has committed to reading.

The honest answer for a multi-user site today

If you run a CMS where different accounts want different AI-training preferences on pages sharing one domain, there is no mechanism any major crawler will honor yet. Robots.txt groups apply by crawler, not by author, and the meta tag was never extended to cover this. The nearest working option today is a domain-wide decision, or moving the content you want excluded to a separate path or subdomain with its own robots.txt group. Watch the Content-Usage header once the IETF draft ships, since it is the one built for exactly this case.