On 30 October 2026, Anthropic, Google, Microsoft and OpenAI backed a shared robots.txt signal that lets a site refuse AI model training on its content, according to the IETF and ITdaily. It applies to any website, including your Belgian company's. The technical step is simple. The decision affects your visibility in AI answers.
1. One token in robots.txt
The signal is called __usage_ai-training. According to the document filed with the IETF on 30 October 2026, it is written as an ordinary robots.txt group: a User-Agent line with this token, then Disallow: / for the whole site, or a path for a single section. Until now, each AI provider required its own separate settings.
According to ITdaily, Google, Anthropic, OpenAI and Microsoft will respect the signal. Meta says it supports the idea without saying how it will implement it.
2. A voluntary commitment with a staggered timetable
The text is an individual Internet-Draft that the IETF has not endorsed, and ITdaily notes that the standard is not final. According to AI Weekly, Anthropic has already started, Microsoft is aiming for 1 February 2027 and Google for 2027. OpenAI has not given a precise date.
3. What the signal does not do
According to AI Weekly, OpenAI describes it as a coordination mechanism, not a security control: a bot that ignores the rules can ignore it. It also does not remove what has already been collected in existing datasets or models.
4. The real choice: be cited or be protected
Microsoft warns that blocking training or answer grounding reduces a brand's visibility in AI applications, but that its Bing ranking does not depend on these settings, according to ITdaily.
For an SME that wants to be cited in AI answers, blocking everything has a cost. For a business whose content is the product, blocking nothing has another. The right answer differs by section of the site.
The AI lens, humans first
AI does the spotting: it lists the bots visiting your site, reads your robots.txt, suggests rules by section and checks that they are in place. It cannot know what you want to protect. Deciding which content should feed AI answers and which should stay yours is a management choice, not a technical setting. AI executes. Expertise decides, and keeps watch.
This week
Open your website's robots.txt file (your-site.be/robots.txt) and note what it already contains. Sort your pages into two groups: those you want AI assistants to cite, and those you would rather keep out of training. Change nothing before you have done this sorting: the first effects are only expected from 2027.
Vistalaro Build audits your site's robots.txt and headers, then sets up the rules section by section on European servers. Vistalaro Pilot helps you weigh content protection against visibility in AI answers, with a senior marketer who knows your business.
Should your content feed AI models?
Let's talk before 2027: a simple sort of your pages avoids a block that is too broad or a door left too open.
Let's talk- ITdaily, Websites can more easily restrict AI training on articles (3 November 2026)
- IETF Datatracker, Machine-Readable Training Preference Signal (draft-kelley-aipref-training-preference-signal-00) (30 October 2026)
- AI Weekly, OpenAI, Google, Microsoft, Anthropic Back One AI Opt-Out Token (30 October 2026)