On 30 October 2026, OpenAI, Google, Microsoft, and Anthropic each said they will honour one shared robots.txt token that lets a website refuse to have its content used for AI training. It is called `__usage_ai-training`. If you run a shop or cafe website, this concerns you. Just not as dramatically as the headlines suggest.

Our view is short: this is good news that is not yet protection. The sign is up, but only one guard has started reading it.

Three numbers from the 30 October news: four AI companies on board, two lines in robots.txt, and 1 February 2027 when Microsoft starts honouring the token

What actually changed

Until now, if you did not want your shop content used to train AI models, you had to write a separate rule for each company, using each one's crawler name. Google, for instance, uses Google-Extended. Now there is one shared name. In robots.txt you write `User-Agent: __usage_ai-training`, and beneath it `Disallow: /`. If some part of the site may be used, such as `/press-kit/`, add an `Allow:` line for it.

The token comes from an IETF draft titled "Machine-Readable Training Preference Signal", written by Kevin Kelley of Anthropic. It is an individual draft: not endorsed by the IETF and not an official standard. What is real is the commitment on the company side. All four say they will read the token.

A sign on the shop door

Think of robots.txt as a sign on your shop door. A polite visitor reads it before walking in. Until now your sign had to carry one sentence per visitor: "Team A, no photos", "Team B, no photos". Now one sentence will do: "No photos for training AI."

Two things about that sign. First, it is about use, not entry. The IETF draft says the token is not a directive for crawling behaviour, so bots may still open your pages. The token only states that what they read should not be used to train models. Second, a sign is still a sign. OpenAI itself describes it as a coordination mechanism rather than a security control. A visitor who chooses not to read it will not be stopped by a piece of wood.

Comparison of opting out of AI training before and after 30 October: per-crawler rules versus one shared token

What did not change

Start with timing. Anthropic says it has already started implementing support. OpenAI says it aims to do so in the coming months, with no date. Microsoft says 1 February 2027. Google says 2027, and its crawler documentation says early 2027. Meta, Amazon, and Apple have made no commitment. So if you add the token today, only some of the readers are reading it today.

Then there is strength. The IETF draft says private agreements may override a preference expressed this way. Old rules do not disappear either: names such as Google-Extended keep their current meaning. And blocking everything with `User-Agent: *` does not count as a stance on AI training. A shop that blocks all bots has not said anything about that.

As for older data, the sources we read do not explain what happens to content collected before the token was added. What they do say is that preferences are judged by the robots.txt in effect when the content was retrieved. Do not expect the token to pull anything back.

The shared token sits above per-crawler rules, private agreements that can override it, and voluntary compliance

One trap with Gemini

This is the part most likely to bite an online shop. Google says a shared opt-out will be treated as if you had disallowed Google-Extended. Google-Extended does not affect your position in Google Search. But it does cover Gemini, including grounding, which is when Gemini pulls in website content to answer a user's question.

That means an opt-out alone could take your shop out of Gemini's answers. Google's blog says sites that specifically allow Google-Extended remain eligible to appear in responses and receive clicks via grounding in Gemini Apps. If you want your content kept out of training but still findable, write an explicit `Allow` for Google-Extended. Google's own support is only scheduled for 2027, so this is about planning, not firefighting.

The decision you can finish this week

Nothing here needs doing at midnight. But three things can be finished within a week, and all of them are more about thinking than typing.

  1. Look at your robots.txt. Open yourdomain.com/robots.txt in a browser and note what is there. Many shop owners have never seen it.
  2. Decide your stance. Product photos, descriptions, and customer reviews: may they be used to train models or not? Two lines in robots.txt are easy. Deciding the answer is what takes time.
  3. Settle the Gemini question. Do you want to stay out of training but remain visible in answers, or step out of both? That choice decides whether you need the extra `Allow`.

After that, note one date: 1 February 2027, when Microsoft starts honouring the token. If your online shop platform locks robots.txt, ask the provider or your developer whether the file can be edited.

Three decisions that can be finished in a week: look at robots.txt, decide your stance, settle the Gemini question

The consequence is simple. A business that decides now and writes it down will not be scrambling when Microsoft starts reading the token in February. One that waits will decide under pressure, or not decide at all, and that is also a decision.

If you want to talk through the right stance for your business website, message us on WhatsApp at +62 823 1499 5005.