Google plans to support a shared website control for AI training in 2027, potentially giving publishers one instruction that participating providers can understand. If you run a portfolio, shop or publication, the useful part is simpler maintenance. The immediate caution is that this isn't a replacement for your existing controls yet.
In its announcement of the shared training control, Google says a new preference in robots.txt will work alongside company-specific settings. That matters because a common instruction could become your baseline while individual exceptions remain possible. It doesn't mean every AI service will start respecting the same rule at the same time.
A shared preference, still being developed
The Machine-Readable Training Preference Signal proposal uses the name __usage_ai-training in an ordinary robots.txt rule group. A publisher could express a preference against training across a site while allowing a particular directory, such as a press kit.
Its status deserves attention: this is an individual Internet-Draft, not an approved IETF standard. The proposal separates preferences about training use from the rules governing whether a crawler may fetch it. Software that doesn't understand the new signal can ignore it, so adding the text is not a technical barrier around your content.
Anthropic engineer Kevin Kelley’s explanation of the draft says the company has begun implementing support, with more work needed across its services. He also explains that existing wildcard rules don't automatically become a training preference. A site hasn't expressed this new preference simply because it already has a robots.txt file.
Keep search and training decisions separate
Google's proposed handling has an important qualification. If a site opts out through the shared signal and hasn't explicitly specified Google-Extended, the company plans to treat that preference like a Google-Extended disallow. Explicit company-specific settings will continue to matter.
The current crawler documentation explains why you should review that consequence carefully. Google-Extended controls both training future Gemini models and grounding in specified Gemini products, where retrieved content helps inform an answer. It does not affect inclusion or ranking in Google Search. Those are different surfaces, even when they belong to the same company.
For a small site, the sensible preparation is an inventory rather than a rushed configuration change. Keep a copy of your current rules, record why each exception exists, and decide which material you want available for which uses. A public press kit and a collection of original essays needn't have identical preferences.
Leave working provider-specific controls in place while support develops. When implementation details arrive, compare them with your actual goals before changing the file. A shared vocabulary will be useful only if you understand what each participating service does with it.