Google Gemini: Temporary Chats to Be Stored for Two Months
Google is overhaulng how its Gemini AI assistant handles transient data and user privacy. The company has announced a major extension of the storage duration for temporary chats and a restructuring of privacy controls, allowing users to separate conversation history from AI model training settings.
For users who rely on the 'Temporary Chat' feature to keep conversations out of their permanent history, the window of anonymity is shrinking. Previously, these chats were deleted from Google's servers after a maximum of 72 hours. Under the new policy, which Google plans to roll out globally in the coming weeks, such conversations will remain linked to the user's Google account for approximately two months.
Security and Contextual Reasoning
Google justifies this extended retention period by citing the need for enhanced security measures. According to the company, the longer storage period is intended to help investigate technical malfunctions, ensure service security, and better detect potential instances of abuse. While the data stays on the servers longer, Google maintains that Gemini will still use only the last 24 hours of conversation content as context to answer queries. Furthermore, the company explicitly stated that content from these temporary chats will not be used to train its generative AI models.
Granular Privacy Controls
Alongside the storage changes, Google is decoupling its privacy settings. Previously, the ability to save chat history and the permission to use data for AI training were closely linked. The new update allows users to manage these two functions independently. This means users can opt to keep a record of their regular conversations without automatically granting permission to use that data for improving Google's AI models.
The rollout logic depends on the user's current settings: those who have already disabled activity storage will find both new options turned off by default. For users with history currently enabled, the chat history will remain active, but the permission for AI training can now be managed separately. However, a caveat remains for those providing voluntary feedback: if a user rates a response or reports a problem while using a temporary chat, the associated content—including the last 24 hours of dialogue and any attached files—can be shared for review. This specific data may be inspected by trained human reviewers and could be stored for up to three years under certain circumstances.