Disable Gemini model training in Google Workspace

We write as an implementation team, not a law firm: a free guide on what works in practice, ready to verify against your documents. We start with a single setting in the Google Workspace Admin console: disable the use of employee prompts and files to train Gemini models. After reading, you will have an inactive toggle in your panel, confirmation of the change, and a rollback path if the setting proves broader than needed. Before opening the panel, check that you hold the Super Admin role or have privileges to manage Google Workspace service settings; without these, the fields may be invisible. You also need a Google Workspace license that includes Gemini; free personal accounts usually lack this control.

Krok 1: Otwórz ustawienie prywatności Gemini (5 minut)

Open admin.google.com and sign in to your administrator account. On the left side, click Apps, then Google Workspace, and locate Gemini or Gemini Settings. In versions of the console where this item is not visible, type "Gemini" in the search bar at the top. Inside, look for the section named Privacy or Privacy Settings in Google documentation: find the link in Google materials. Click Edit next to the option that includes "train" or "improve" in its description: depending on your edition, it is named "Data used to improve models" or similar. Set the toggle to Off. Save by clicking Save.

AIAI-generated image.

Krok 2: Sprawdź, czy zmiana weszła (10 minut)

Go back to Apps > Google Workspace > Gemini and open the same setting. Verify that the toggle remains in the Off position and that the status field shows an entry with the timestamp assigned to your administrator account. In the audit center or admin console logs, search for the setting change event: under reporting, select Security Center, then the section dedicated to Gemini, and filter events for today's date. If logs are not available, run a quick test: open a document in Google Docs, invoke Help me write in Gemini, and return to the setting to confirm that nothing was re-enabled. Keep in mind that propagating the change across user profiles can take up to 24 hours in larger organizations.

Krok 3: Zapisz zmianę w rejestrze (5 minut)

In your company's AI tools registry, log the date, time, administrator account, and the exact name of the disabled option. Save a screenshot of the toggle in the Off position alongside a copy of your Workspace subscription agreement. This record simplifies answering "how was the setting configured then" during a GDPR inspection or security audit. In your email to the IT team, include a link to Google documentation and this note: training on prompt contents was disabled on the specified date; this does not affect privacy settings in other services. Do not send passwords or active session tokens from the console.

AIAI-generated image.

Gdy coś nie wyszło

The first issue is a missing Gemini section after clicking Google Workspace. In practice, this usually means a license without this service or an account lacking settings management privileges: check under Admins > Admin roles to see if your role includes permissions for services; if not, ask another administrator to grant them. The second issue is a grayed-out toggle that cannot be changed. It is often inherited from an organizational unit higher up; open the Organizational units view on the left, select your specific company unit, and click Override or temporarily adjust inheritance. The third issue is a change that looks saved, but reverts to On after a refresh: first clear your browser cache and sign in via incognito mode, then check if a sync rule from another Google Workspace management tool is running. To revert the change, return to the same field, switch the toggle to On, and click Save; the rollback takes effect immediately in the console, though logs may display the old value for a few hours.

Disabling training in Gemini closes one of the easiest paths through which prompts and uploaded files become part of a third-party vendor's model. For a company of 100 people, where each employee enters dozens of snippets into an AI tool daily, a single toggle frees the IT team from repeatedly explaining policy and creates a clear audit trail. The risk assessment remains with the administrator and legal counsel, but the configuration itself takes 15 minutes. It pays to keep the scale in mind: UODO outlines breach reporting deadlines, and EU fines reach 20 million EUR or 4% of turnover; that is not a threshold to test over a setting you can disable before your first cup of coffee.