Gemma: Google Open Models, Web Access and Training Data
Gemma is a family of open models from Google. It is a model, not a chat service with web access. Gemma does not crawl, search or fetch web pages by itself, so there is no Gemma user agent to allow or block.
What Gemma is
Google describes Gemma as lightweight open models built from the same research and technology as Gemini. According to the Gemma documentation, Google DeepMind and other teams at Google develop it. The newest version on that page is Gemma 4. Developers can get Gemma variants from Kaggle and Hugging Face. They run the models in their own apps, on their own hardware, on mobile devices or on hosted services. Google marks that documentation as deprecated and points to gemma.ai for new pages.
Web access in apps built on Gemma
A Gemma model reads only the text and files that an app gives it. The Gemma 4 model card says that Gemma 4 has native support for function calling (structured tool use). So a developer can connect it to a web search or fetch tool. In that case the app sends the requests, with the user agent that its developer chooses. These requests do not come from a Google crawler, so the Google crawler pages do not list them.
If a bot that uses Gemma visits your site, identify it by its own user agent and IP address. A rule for a Google token does not apply to it.
Web text in Gemma training data
The same model card says that the training data includes web text in more than 140 languages. The cutoff date of the data is January 2025. The card also says that Google filtered the data for safety and quality, and removed certain personal information and other sensitive data with automated methods.
The model card does not name the crawler that collected the web text. It does not mention robots.txt or Google-Extended. Google’s Google-Extended documentation names Gemini models and the Vertex AI API for Gemini. It does not name Gemma. So no Google page confirms that a Google-Extended rule removes your pages from future Gemma training data.
For Google’s documented AI controls, read the Google-Extended guide and the Google Gemini guide.