A Shared Signal to Opt Out of AI Training: What You Need to Know

Google, Microsoft, OpenAI and Anthropic support a shared signal that would allow websites to express their opposition to having their pages used to train AI models. Announced on October 30, 2026, the mechanism is based on a proposed directive for the robots.txt file. It could make it easier for publishers to express their preferences, without giving them guaranteed control over how their content is used.

A signal separate from crawling rules

Named __usage_ai-training, the signal is intended for the training of generative models. It is not a directive designed to stop bots from crawling a website: crawling and the subsequent use of content for training are two separate matters.

The mechanism has been proposed as an individual draft of the IETF, the Internet standards organization. It is not a finalized standard. According to PPC Land, the participating companies support its development on a voluntary basis. Its presence in a robots.txt file therefore does not, by itself, guarantee that content will be excluded from training data.

The announced timelines differ

The reported commitments vary from company to company. Microsoft plans to take the signal into account starting February 1, 2027. Google is aiming for early 2027. OpenAI has not announced a date, and Anthropic says it has begun developing support for the signal without announcing a timeline.

These announcements do not mean that the feature is already uniformly available. Website owners will need to check for updates and review each company’s rules before changing their robots.txt file.

What the setting does not address

The signal does not remove pages that have already been collected, or content that may already have been included in training datasets. Nor does it block a website from being crawled. Finally, because adoption is voluntary, it offers no general guarantee against pages being used.

For publishers, the proposal could provide a clearer way to express a preference, if the services concerned choose to respect it. However, it does not replace checking each provider’s policies or other access-management measures. Its practical effect will depend on how it is actually implemented, which is still to come for several companies.

AI-generated illustration · Ayo.