Google says Workspace prompts are not used to train Gemini

What the documentation states about training

Google's Generative AI in Google Workspace Privacy Hub, as retrieved on 16 September 2026, states that user prompts are treated as customer data under the Cloud Data Processing Addendum and that Workspace does not use customer data to train its models without the customer's prior permission or instruction. Google's separate data-governance documentation for Gemini in Google Cloud makes the equivalent statement for that product line, adding that Gemini does not use prompts or responses to train its models, with a named exception for an opt-in Trusted Tester Programme used for product improvement rather than model training.

What is left to admin settings

The privacy hub is specific about retention, and retention is where the real configuration choices sit. Gemini features inside Workspace apps such as Gmail, Docs and Meet can retain conversation data from 90 days up to an indefinite period, and administrators control that window. The separate Gemini app has its own retention setting, running up to 36 months by default, also administrator-controlled, while Gemini in notebook-style products is described as not retained once a session ends. Administrators can additionally choose whether users may delete individual conversations themselves, a capability enabled by default but one that can be restricted centrally.

Why the retention detail matters more than the training promise

A no-training promise addresses one specific risk: that a competitor's prompt or a confidential document becomes part of a shared model. It does not address a second, more mundane risk: that a prompt containing sensitive information sits in Google's systems for months or years because an administrator left a long retention window in place, or never checked it. The documentation makes the retention period a matter of admin configuration rather than a fixed vendor commitment, which means the actual privacy posture of a given Workspace tenant depends on decisions made locally, not only on the vendor's stated policy.

• What retention period is actually configured for Gemini in your Workspace tenant, and who set it?
• Does your admin console allow users to delete their own conversations, and is that setting known to staff?
• Which products in your suite fall under the Cloud data-governance page versus the Workspace privacy hub, and do their terms match?

Google's no-training language is a clear and checkable commitment; the more consequential setting for most teams is the retention window sitting one menu below it, which nobody enforces by default.