It isn't live yet. OpenAI says it will implement it "in the coming months" and Google talks about 2027. But it's worth understanding now, because it touches a decision every company with a website will have to make: whether it wants AI to learn from its content, and whether it wants AI to cite it.

◢What exactly was announced

The idea is simple. Instead of naming each crawler, a website uses a special name in its robots.txt that means "AI training", and participating providers read it as a preference about that specific use. According to the draft, training means using content to modify the parameters of a generative AI model.

OpenAI says the signal doesn't replace its existing controls, which will keep working. Google explains that if a site uses the shared signal to opt out of training and says nothing specific about its own control, Google-Extended, it will treat it as if Google-Extended were blocked.

◢How the rule is written

The draft's own example is this:

User-Agent: __usage_ai-training
Disallow: /
Allow: /press-kit/

That says "don't use anything on my site to train AI, except the press kit folder". It follows the same path rules you already know from robots.txt. One detail that trips people up: the general User-Agent: * rule doesn't count as this signal. If you want to express the preference, you have to use the special name.

◢What this signal doesn't do

This is the part I care about most, because it's easy to get wrong.

It doesn't block anyone. robots.txt is a request, not a locked door. Providers that commit to honoring it will, but a crawler that doesn't take part can ignore it.

And watch out for this in Google's case: according to its announcement, sites that allow Google-Extended are the ones eligible to appear in Gemini responses with a link to the source. If you block it, you may lose that visibility.

◢Should I use it?

It depends on what your business lives on.

If your content is your product (a paid publication, a database, courses, photography), it makes sense not to want it used to train models without an agreement. For you, this signal is good news because it saves you maintaining a list of crawlers that keeps growing.

If your website exists so people can find you (a services company, a consultancy, a law firm, a shop), the decision is different. What you want is for ChatGPT, Gemini, Claude or Perplexity to cite you when someone asks about what you do. That's what I explained in what GEO is and how to rank your website in AI answers. In that case, what matters is not blocking the search and answer crawlers by mistake, which aren't the same as the training ones.

On this site, for example, I deliberately allow AI crawlers in robots.txt: I want them to read me and cite me.

◢What to do now

There's no need to change anything today, because the signal doesn't work with any provider yet. But it's a good moment to check your robots.txt and know what it says. Many sites blocked everything that sounded like AI in 2023 and are now missing from ChatGPT's answers without realizing it.

Look at which crawlers you're blocking, decide whether your priority is protecting your content or being found, and note this signal for when OpenAI and Google switch it on. If you use it, combine it with specific rules for the search crawlers you do want to let through.

I'll update this story when there are concrete dates or more providers join.