Language and display
BE THE BOSS OF AI.ADRIAN DUNKLEY ↗
← This newspaper edition

Gemini will let you keep chat history and opt out of training

Google announced separate controls for its Gemini app. Check what your account actually shows before you change how staff use it.

Cream accordion folds curve between transparent blue panels beside a shorter orange fold.
Paper Kept Apart · AI-generated artwork · AI Boss Ltd

On 16 November, Google said it would give users of the personal Gemini app separate controls for keeping their chat history and letting Google use it to train generative AI models. The company says it will start rolling out the change globally in the coming weeks. For a small office that keeps a useful thread open between shifts, the change means the team may be able to keep that conversation without contributing it to model training. Look for the settings in the account you use for work before changing staff guidance.

An empty history can still leave a copy with Google

The new training setting covers how Google uses chats to improve the models that generate responses. History controls whether you can return to a conversation through the app. Google says that after the update, turning History off will still leave future chats on its systems for around two months so it can respond, handle feedback and protect users. The Gemini Apps Privacy Hub also describes other processing, including for safety. Google says turning off Chat content applies to future chats. If you send feedback, the conversation may still be used for improvement. Work or school accounts can have different terms and administrator controls.

Check the account before anyone pastes a customer message

On the Gemini website, open Settings, then History. If your account shows separate options and your company allows it, keep History on and turn “Chat content including text & uploads” off. Start a new chat with this sample message: “The shop opens at 9 a.m. on Saturday. Rewrite this as a sign.” Reload the page. Check that the new chat appears in History and the Chat content control remains off. The screen cannot show how Google handled a specific conversation.

If the new options are absent, follow your existing workplace rules until the account changes. A training preference on a consumer account does not authorise uploading a guest's passport, booking record or complaint.

Where do staff go when the approved tool falls short?

Start the Shadow AI Risk Diagnostic with your manager or IT lead. Have your written AI rules and a list of approved work accounts to hand. Answer the questions on rules, access, staff behaviour and how you would spot a problem. In an illustrative run using the second option throughout, the tool returned 33 out of 100, labelled Emerging Exposure, and broke down the score by those four areas. Use the result to choose what to verify. Ask one team to show which account it uses for customer replies and what it does when that tool cannot finish a task. Compare the answer with the written rule. The score reflects the answers you enter. It cannot see staff accounts or check whether anyone followed a rule.

Sources and further reading

Sources checked 17 November 2026.