On 30 October 2026, Google announced a plan on the Google Search Central Blog for a single joint AI training rule in robots.txt. The proposal is based on the work of the Internet Engineering Task Force’s (IETF) AIPREF working group. With this rule, you can communicate your preferences for training AI models to multiple AI providers at once in a single step.
Google is building on Google-Extended, the setting the company has used since 2023. If you use the new rule to opt out of training and do not mention Google-Extended separately, Google will treat this as an opt-out for Google-Extended. Specific rules per AI developer will continue to work. Organisations that explicitly allow Google-Extended will remain eligible for citations and links via grounding in Gemini apps.
What this means for non-profits and government
For civil society organisations, this choice cuts both ways. You want to maintain control over how AI models use your copyright-protected reports or policy documents for training. At the same time, you want citizens and donors to find your official information when they ask questions to AI assistants.
Take a registered charity that publishes public research reports. With the new setting, you can specify in robots.txt that AI models cannot use your texts as training data. If you explicitly allow Google-Extended alongside this, Gemini apps can still cite your reports and link to your website when someone searches for that topic.
What you can do now
Google aims to support the new rule from 2027. You do not need to adjust your settings right away, but you can prepare:
- Check your current robots.txt file for specific instructions for AI crawlers such as Google-Extended.
- Determine per content type whether you make material available for AI training or solely for citations in responses.
- Consult your web team about how to record your preferences once the IETF standard is finalised.
Source: Google Search Central
