What to Prepare for the Shared AI Training Opt-Out Signal

Google, Microsoft, and OpenAI said they'll support a shared robots.txt signal: one line that tells every AI company at once that a site should be excluded from AI training. None of the three companies honor it yet, and each has a different rollout date, but knowing about it now means you won't be scrambling later. Here's whether you need to change robots.txt right away or can leave it as is.

Why this signal matters now

Until now, blocking AI training meant adding a separate line for each company: Google-Extended to block Google, GPTBot for OpenAI, ClaudeBot for Anthropic. Every new company meant one more line to track.

The new '__usage_ai-training' signal is an attempt to collapse all of those into a single line. It's still a draft, but since three major companies announced support at once, other AI companies may follow, so it's worth understanding the shift now rather than scrambling to change settings later.

Who created this signal

The signal was first proposed by Kevin Kelly, an engineer at Anthropic, who submitted a draft to an international working group that handles internet standards.

This announcement doesn't actually include Anthropic among the three companies listed. Kelly said his own employer has also started working on it internally, but bringing it to the whole product lineup is still a ways off. The fact that the person behind the draft and the companies that announced it aren't the same shows this signal hasn't been locked in as an official standard yet.

Why you don't need to act immediately

All three companies are still ignoring this signal for now. Adding the '__usage_ai-training' line to robots.txt today won't do anything until each company starts applying it.

Google said early 2027, OpenAI said within a few months, and Microsoft will phase in a temporary setting from December 15 before a new setting takes effect February 1. There's no need to rush, but you should double-check your settings as each rollout date approaches.

Should you keep your existing rules

There's no reason to remove rules like Google-Extended, GPTBot, or ClaudeBot that you've set up for individual companies. Until the shared signal actually takes effect, these are the only rules doing any work. Even after the shared signal rolls out, keeping company-specific rules in place lets you keep handling crawlers from any AI company that hasn't adopted it yet.

Why adding the new signal early can't hurt

Even though it has no effect yet, adding the '__usage_ai-training' line now costs nothing. It reuses existing robots.txt syntax, so there's no extra setup involved, and it simply starts working once each company begins applying it.

One thing to note: if a site uses only this new signal without separately listing Google-Extended, Google will treat that as opting out of Google-Extended too. That's fine if you intend both settings the same way, but if you've been running them with different intentions, it's worth knowing this interpretation rule in advance.

Each company's rollout timeline

The three companies' schedules differ. Google will support the signal starting early 2027, and the same rule covers Gemini app grounding too. Visibility in AI Overviews and AI Mode is handled elsewhere, through Search Console's own generative AI visibility setting.

OpenAI plans to start honoring it in a few months; before that, its current GPTBot policy continues unchanged. Microsoft is managing training and grounding settings separately: a temporary setting arrives in Bing Webmaster Tools on December 15, and the old tags stop being recognized starting March 1 next year. It's worth marking all three dates on your calendar so none slip by.

If you use Bing Webmaster Tools

Sites that have relied on noarchive or nocache tags to manage grounding should pay particular attention to this change. Starting December 15, Bing Webmaster Tools will offer a temporary grounding setting; it actually kicks in on February 1 next year, and the old tags stop being recognized once March 1 arrives. Pages left unconfigured during this window stay open to grounding until they're next crawled, so if you've had a clear reason for using noarchive or nocache tags, check and migrate your setting in Bing Webmaster Tools by early February at the latest.

Blocking AI training is a separate issue from AI search visibility

Blocking AI training through robots.txt and being cited in AI Overviews or ChatGPT answers are two separate settings. For Google, AI Overviews and AI Mode visibility is managed through a separate Search Console setting, and the training opt-out signal doesn't affect it.

If you want to stay out of AI training while still being cited in AI search answers, keep these two settings straight rather than confusing them. It's a common mistake to change one setting and assume the other changed with it, so after making any change, it's worth double-checking which setting you actually touched.

For smaller teams

There's no need to feel pressure to change robots.txt right now. Keeping your existing Google-Extended, GPTBot, and ClaudeBot rules in place and jotting down each company's rollout date (Google: early 2027, OpenAI: a few months out, Microsoft: February 1 next year) is enough for now.

If you manage grounding through Bing Webmaster Tools, just keep an eye on the mid-December update; otherwise, it's fine to revisit robots.txt whenever each company makes its next official announcement.