Google is separating chat history from AI training in its Gemini app, letting users keep conversation history without feeding that data into model training. Users will see two distinct toggles in Gemini's Activity settings instead of a single on-off switch that previously linked both functions.
Previously, the only way to block Google from using chat contents for training was to disable history entirely, which made retrieving past conversations impossible. The new setup lets users retain their chat history while opting out of AI training, or vice versa.
AI training will default to opt-out. Users must find the toggle under Activity in Gemini settings or navigate via Google Account > Data and Privacy > My Activity > Other Activity > Gemini Apps Activity to block training on their data.
The change applies only to future interactions. Historical data already fed into Gemini, including inputs, responses, uploaded files, tool calls, location, and data from connected apps, remains part of the training dataset. Google says it may retain this data for up to three years.
Users can manually delete specific chats on the same settings page. However, chats already reviewed and included in training data will not be retroactively removed. Even if users opt out of training, Google may use disabled-history data for other purposes for up to three years.
Data retention periods are also shifting. When users disable history, Google will now retain that data for two months instead of 72 hours. The company says this longer window helps "respond to you, process your feedback, and protect Google, its users, and the public" by identifying misuse.
The rollout is global but may take several weeks to reach all users.