Google Extends Gemini Temporary Chat Retention From 72 Hours to Two Months

Google is extending the retention period for temporary Gemini chats, and for chats with history turned off, from 72 hours to around two months. TechCrunch reported the change on 17 November 2026. It will roll out globally in the Gemini app over the coming weeks.
Until now, the 72-hour window covered temporary chats and chats with Keep Activity off. Keep Activity is the Gemini setting that determines whether conversations are saved to your account history. Under the revised policy, those conversations are retained for around two months instead. Users typically choose temporary chats because they expect them to be short-lived, so a window measured in weeks rather than days is a substantive change, not a rounding adjustment. Any workflow that treated temporary chats as ephemeral will need its assumptions revisited.
Alongside the retention change, Google is replacing the single Keep Activity setting with two separate controls. One governs chat history. The other governs whether conversations are used to train AI models. Google says temporary Gemini conversations are not used for model training.
The defaults depend on each user's earlier choice. Users who had disabled Keep Activity will find both new settings turned off by default. Users who had it turned on will keep chat history enabled, and will be asked separately whether their chats may be used for model training.
The broader context here is the decoupling of retention from training eligibility. Until now, one toggle bundled the two questions together. Splitting them gives each its own control, and leaves previously opted-out users with both settings off. For anyone who manages Gemini use for a team, or advises on data handling for chat-based tools, this matters. How long a conversation is kept and whether it may inform model development are now separate decisions, and each should be reviewed on its own.
In my view, the practical question for administrators and heavy users is less the length of the window than whether the new controls behave as their labels suggest once the change reaches individual accounts. Settings screens and documentation are where that gets tested, and the rollout over the coming weeks will be the first real opportunity to check. Anyone who built processes around the 72-hour figure should re-read the new policy rather than assume the old behaviour carries over.
On balance, granular consent controls are a constructive direction. A single bundled switch forced an all-or-nothing decision. Two independent settings let a user decide how long Google keeps a chat, and whether that chat contributes to training, as distinct questions. The longer retention will concern some users, while the separation gives others more room to choose. Which of those effects matters more will depend on how each person works with the tool.


