OpenAI to Honor a Shared robots.txt AI Training Signal

On October 30, 2026, OpenAI said it will honor a shared robots.txt signal that lets websites state whether their content may be used to train AI models. The signal comes from an IETF draft and is meant to work across several AI providers at once. OpenAI plans to implement it in the coming months.

What we know

The announcement: OpenAI describes it as a voluntary initiative so that site owners can state one preference for several participating providers, instead of keeping a separate rule for each crawler.
Timing: OpenAI says it aims to implement the signal in the coming months and will publish details when it does. No date is given.
What stays: OpenAI says the signal will not replace its existing robots.txt controls, which it will keep honoring.
Where it comes from: the work of the IETF AI Preferences working group. The Internet-Draft draft-kelley-aipref-training-preference-signal-00, dated October 30, 2026 and authored by Kevin Kelley of Anthropic, defines the token `__usage_ai-training`.
How it is written: per the draft, it is an ordinary robots.txt group with that token on the `User-Agent` line. Its example states a preference against training on everything except the `/press-kit/` folder.
Two key rules in the draft: a `User-Agent: *` group does not count as the signal, and a file with no `__usage_ai-training` group expresses no preference at all.
Not a crawl directive: the draft says the token states a usage preference, not a crawling instruction. Crawler-specific rules still decide what gets fetched.
Its status: the IETF page notes it is an individual draft with no formal IETF endorsement. It expires on May 3, 2027 unless renewed.
Google: its Search Central blog published an October post titled "Supporting a shared AI training control for websites." Its body could not be read for this article, so its timeline is not reported here.

What changes and what doesn't

What changes: how you say "don't train on my site." One group could cover several providers once each of them supports it.
Nothing changes today: OpenAI does not apply it yet. Until then, per-crawler rules are what count.
Search and citations are separate: OpenAI's crawler documentation separates GPTBot (training) from OAI-SearchBot (appearing in ChatGPT search answers). Blocking one does not block the other.
Not a guarantee: it is a voluntary preference. The draft also says private agreements may take precedence.

How to tell if it affects you

Open your robots.txt. Go to yourdomain.com/robots.txt. On WordPress an SEO plugin usually generates it; on Drupal it is a file in the web root or managed by a module.
Split three decisions. Do you accept AI training on your content? Do you want to appear in AI search tools such as ChatGPT search? Do you want to limit crawling? Each is controlled separately.
Don't rely on the wildcard. If you add the signal, it goes in its own group with `User-Agent: __usage_ai-training`. A block under `User-Agent: *` does not express it, and it can also remove your site from Google.
If your site brings in customers, don't block OAI-SearchBot or Googlebot by mistake. OpenAI says ChatGPT search picks up robots.txt changes in about 24 hours.
Wait before changing things. OpenAI will publish how it implements the signal. Until then, if you already block GPTBot, leave that rule as it is.

Sources

Updates: when OpenAI publishes how it applies the signal, or Google's timeline can be confirmed in a readable source, it will be added here with a link.