Google is separating Gemini’s chat-history setting from its model-training permission, allowing users to preserve conversations without automatically contributing future interactions to AI datasets. Previously, turning off history was the only way to opt out of training, which also removed the ability to resume an earlier chat.

The new training control remains opt-out rather than opt-in. It appears under Activity in Gemini settings and through Google Account’s Data and Privacy pages. The rollout is global but may take several weeks to reach every account.

Changing the setting applies only to future activity. Inputs, responses, uploaded files, tool calls, location and connected-app data already used for training are not removed retroactively. Google says it may retain some Gemini activity for as long as three years for other purposes, even when it does not add that material to training datasets.

There is also a trade-off for people who disable chat history entirely. Google currently holds those conversations for 72 hours, with 24 hours visible to the user. Under the updated policy, retention will extend to two months so the company can operate the service, process feedback and investigate abuse. The separate switch gives users a clearer training choice, but it does not create a zero-retention mode or erase data that has already entered Google’s systems.