KEY POINTS
- Gemini is getting separate settings for chat history and for the use of chats in AI training.
- Training stays opt-out, and the change applies only to future conversations, not data already used.
- With history turned off, Google will keep chats for two months instead of 72 hours.
Google is rolling out new privacy controls in its Gemini app that let people stop their conversations from being used to train the company's AI models without losing their chat history. Ars Technica reported on the change, which Google says is being deployed globally and may take a few weeks to reach every user.
Until now, the only way to keep chats, including attachments, out of the training pipeline was to switch off history entirely. That made it impossible to return to an earlier conversation later. The updated settings split the two functions into separate toggles, so a user can keep a searchable history while declining to contribute to training.
Training remains an opt-out arrangement. Users who do not want their chats used for that purpose must find the toggle themselves. It sits under Activity in the Gemini settings, and it can also be reached through a Google account under Data and Privacy, then My Activity, Other activity and Gemini Apps Activity. The history setting keeps its adjustable time limit, but changing it is no longer needed to block training.
The change is not retroactive. Anything already fed into Gemini is part of the training data, according to the report. That can include inputs, the chatbot's responses, uploaded files, tool calls, location and data from connected apps. Google says it may keep this information for up to three years to improve its models, and it has said before that it tries to filter personal information from activity history but cannot guarantee it.
Users can still delete individual items on the same settings page, whatever options they choose. Chats that have already been reviewed and added to training data will not be removed after the fact. Opting out also does not mean Google stops holding the data: it may use it in other ways for up to three years, just not as part of AI-training data sets.
There is a trade-off for people who disable history to avoid retention. At present, Google keeps that data for 72 hours, and the user can see it for 24 hours. Once the updated page is live, the period will extend to two months. Google says this allows it to respond to users, process feedback and protect the company, its users and the public. Ars Technica interprets this as giving Google a better chance of spotting abusive use of its AI even when history is off.
Despite the longer retention window, the outlet describes the new training control as a welcome improvement for privacy.