Crawlerpeek
Check a site

noai Meta Tag: Does Any AI Operator Honor It? (2026)

The noai meta tag has no documented reader at the AI operators we checked on 1 Nov 2026. See 13 AI opt-out signals compared by scope and status.

Published
9 min read
  • noai
  • robots-txt
  • ai-training
  • google-extended
  • content-signals
  • tdmrep
  • opt-out

The noai meta tag has no documented reader among the operator pages we read on 1 Nov 2026 (OpenAI, Google, Anthropic, Apple, Common Crawl), and Meta's page names it only to prefer robots.txt. Of 13 AI opt-out signals compared here, four appear in no AI operator statement we found: noai, Content-Signal, Content-Usage and tdm-reservation.

As of 1 Nov 2026.

You want to keep your articles out of AI training, and you find five different answers: a robots.txt line, a meta tag, an HTTP header, a Cloudflare content signal and an IETF draft. Each one is real. Only some are read by anyone who matters to you, and the answer changes with what you want to stop: training, live answers, or Search features. This table is built only from the operators' and authors' own pages.

Which AI opt-out signals exist, and who says they honor them?

Signal Where it lives Scope Who documents honoring it (pages checked 1 Nov 2026) Status
GPTBot token robots.txt Training (OpenAI) OpenAI: "Disallowing GPTBot indicates a site’s content should not be used in training generative AI foundation models." Vendor-specific
Google-Extended token robots.txt Training and grounding in Gemini Apps and Vertex AI; not Search Google crawler docs Vendor-specific
Applebot-Extended token robots.txt Training of Apple foundation models Apple Vendor-specific
ClaudeBot token robots.txt Training (Anthropic) Anthropic support article, dated 7 Apr 2026 Vendor-specific
Meta-ExternalAgent token robots.txt Training and indexing crawl Meta Vendor-specific
__usage_ai-training robots.txt Training, any participating provider Microsoft from 1 Feb 2027; Google "early 2027"; Anthropic implementing Individual Internet-Draft (not a working group document)
Content-Signal: ai-train=no robots.txt Search, grounding (ai-input), training (ai-train) Cloudflare publishes it; no AI operator statement found Cloudflare policy
Content-Usage: train-ai=n robots.txt or HTTP header Training No operator statement found IETF working group draft
ai-training and ai-grounding <meta name="Bingbot"> or X-Robots-Tag Page-level training and grounding (Microsoft) Microsoft, from 1 Feb 2027 Vendor-specific
noai, noimageai <meta name="robots"> or X-Robots-Tag Use as input to AI models DeviantArt for its platform; no AI operator statement found Convention, not a standard
tdm-reservation HTTP header, HTML meta, /.well-known/tdmrep.json Text and data mining reservation No operator statement found W3C Community Group report
Search Console AI setting Search Console Google AI Overviews, AI Mode, AI Overviews in Discover Search Console Help ("Search generative AI control"); Google blog post of 3 Jun 2026, updated 31 Aug 2026 Vendor-specific
nosnippet, max-snippet Meta tag or header What Google Search shows Google AI features guide Vendor-specific

The first five rows are one mechanism with five vendor tokens; how to block AI crawlers lists them with copy-paste rules. "No operator statement found" means the pages named in Sources do not mention the signal; it does not mean no operator reads it. Which of the newer entries has the least behind it?

How firm are the newer signals?

The shared token __usage_ai-training is the newest, an individual IETF Internet-Draft dated 30 Oct 2026. Microsoft says it will honor the new preferences from 1 Feb 2027, and Google's crawler page says it "intends to support" the token "in early 2027". Our token article covers the draft, the start dates and a parser test. OpenAI's crawler page does not name the token as of 1 Nov 2026; PPC Land reports that OpenAI aims to implement it "in the coming months", which OpenAI's page does not confirm.

Cloudflare's Content Signals are older: the policy post is dated 24 Sep 2025. It defines search, ai-input and ai-train, and it says plainly that "content signals express preferences; they are not technical countermeasures against scraping. Some companies might simply ignore them."

The IETF working group uses other names. Its attachment draft, draft-ietf-aipref-attach-05 (published 19 Aug 2026, expires 20 Feb 2027), defines Content-Usage: train-ai=n for an HTTP header and for robots.txt. Its vocabulary draft, draft-ietf-aipref-vocab-08 (published 14 Sep 2026, expires 18 Mar 2027), labels the categories train-ai, ai-use and search. So Cloudflare's ai-train and the IETF's train-ai are different strings, each defined in its own document. Kelley wrote on 30 Oct 2026 that his token draft "differs from the attachment draft that we put aside in order to make progress on the vocabulary", and both IETF documents are still drafts.

tdm-reservation has the longest history of the web-facing formats. It comes from a W3C Community Group Final Report dated 10 May 2024, which says it "is not a W3C Standard". It defines a header, an HTML meta element and a tdmrep.json file under /.well-known/. Where does noai sit next to these?

Does the noai meta tag work, and what does it do?

The noai meta tag is a convention DeviantArt introduced on 11 Nov 2022 to say that a page's content may not be used as input to AI models. As of 1 Nov 2026, none of the AI operator pages we read documents honoring it, so it works as a statement of intent, not as a block.

The convention comes from DeviantArt's post of 11 Nov 2022, which says: "When an artist declares that their content may not be used as input to AI models, an HTML tag with the noimageai directive is placed on the page and an HTTP header with the noai directive is sent when the image is directly downloaded from DeviantArt’s servers." The post publishes the syntax and invites others to adopt it:

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

X-Robots-Tag: noai
X-Robots-Tag: noimageai

DeviantArt wrote: "We encourage other platforms who enable creators to publish their work to implement similar protections." Of the operator pages we read, only Meta mentions the tag, and only to steer people elsewhere: "We make it easy for site managers and content owners to indicate their preferences by using industry-standard practices like robots.txt rather than non-standard formats like NoAI tags." Setting noai costs one line of HTML, but we found no AI operator that documents honoring it. Treat it as a statement of intent, not a block. Is there anything that works page by page?

What changes per page, and what happens to NoArchive and NoCache?

Microsoft documents AI-specific page-level controls for training and grounding, and says it starts honoring them on 1 Feb 2027. Its blog (30 Oct 2026) gives these forms:

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

X-Robots-Tag: Bingbot: ai-training:disallow

Training can be set in robots.txt or per page, and a page-level training setting overrides robots.txt. Grounding is page-level only: "There is no site-wide robots.txt grounding preference." Microsoft adds that "These settings govern use of content, not crawl access", so crawl access is controlled separately. If you used NoArchive or NoCache, the dates matter:

Date What Microsoft says happens
15 Dec 2026 Temporary grounding preferences open in Bing Webmaster Tools, only for sites using NoArchive or NoCache as of that date
1 Feb 2027 Microsoft starts honoring the new training and grounding preferences as it crawls
1 Mar 2027 Deadline: after it, NoArchive and NoCache are no longer read as a training or grounding preference
30 Sep 2027 Temporary grounding preferences expire

Google runs a different split. Its AI features guide says robots.txt for Googlebot controls crawling for Search, and nosnippet, data-nosnippet, max-snippet or noindex limit what Search shows. Search Console Help documents a "Search generative AI control" under Settings, which Google's blog calls a toggle in Search Console; the blog post's 31 Aug 2026 update says Google has "rolled out these features to all websites worldwide". The same post warns: "Sites that opt out will not receive traffic or impressions from our generative AI features." The help page adds that the control "doesn't affect AI training" and points to Google-Extended for that. That is a trade, not a free switch. Where does llms.txt fit among these?

Is llms.txt an opt-out signal?

No. The llms.txt site describes it as "A proposal to standardise on using an /llms.txt file to provide information to help agents use a website." It points agents at content, so it expresses no permission or refusal. Our llms.txt guide covers what it can and cannot do. It belongs in neither column of the table: the proposal describes help for agents, not a permission or a refusal.

Which signals should you set?

Start from the goal, because the cost differs. Microsoft states that choosing Disallow or DisallowExtended for grounding "will significantly reduce your content’s visibility in AI-generated experiences", and Anthropic notes that disabling Claude-SearchBot "may reduce your site's visibility and accuracy in user search results".

Your goal What to set today Documented reader
Keep content out of model training A robots.txt group per operator token you care about: GPTBot, Google-Extended, Applebot-Extended, ClaudeBot, Meta-ExternalAgent Each operator, per its own page
Express the same preference once Add __usage_ai-training next to the vendor tokens Microsoft from 1 Feb 2027; Google early 2027
Keep a page out of Bing grounding ai-grounding:disallow on the page Microsoft from 1 Feb 2027
Stay out of Google's AI features in Search The Search Console setting Google
Signal intent to unnamed tools Content-Signal, Content-Usage or noai, understanding no AI operator documents reading them None found

Takeaway: the vendor tokens are the controls with an operator statement today, plus Microsoft's page tags from 1 Feb 2027, so keep them and treat the rest as additions. Paste your file into the robots.txt validator to see how a parser reads it, then use the robots.txt generator to write the groups. Our study of 151 robots.txt files covers which AI crawler tokens those files name.

Common follow-up questions

Does the noai meta tag work to stop OpenAI or Google from training on my pages?

We found no statement from either that it reads noai. OpenAI documents GPTBot and Google documents Google-Extended for training. Use those robots.txt tokens, as in blocking AI crawlers.

Do Content Signals replace robots.txt tokens?

Not according to Cloudflare, which says signals are preferences and "Some companies might simply ignore them". They sit beside Allow and Disallow rules.

Is Google-Extended the same as the Search Console AI setting?

No. Google's crawler page says Google-Extended covers Gemini training and grounding and "does not impact a site's inclusion in Google Search". The Search Console setting governs AI Overviews, AI Mode and AI Overviews in Discover.

Do I still need NoArchive or NoCache for Bing?

Microsoft says that after 1 Mar 2027 they stop counting as a training or grounding preference, and that once it detects the new preferences on a site it disregards leftover NoArchive and NoCache. Publish the new training and grounding preferences before then.

Sources

Checked 1 November 2026.