Google is quietly extending how long it holds on to conversations users may have assumed were fleeting. According to an update to the Gemini Apps Privacy Hub surfaced by TechCrunch on Monday, temporary chats and conversations created while chat history is turned off will now be retained for around two months — up from the 72-hour window documented in an archived version of the page dated November 12. The change will roll out globally in the Gemini app over the coming weeks.
The company's stated rationale is safety. As AI systems become more capable, Google says it needs longer retention to respond to users, process feedback, investigate potential policy violations and protect its services, and the updated hub also cites potential system failures as a reason to keep chats around. Google did not answer questions about what prompted the timing of the change or whether it affects who can access and review conversations, instead pointing to its documentation.
For users of rival assistants, the comparison is instructive. OpenAI says it may keep temporary ChatGPT chats for up to 30 days, and Anthropic retains incognito Claude chats for 30 days by default — both already longer than Gemini's old 72 hours. Google's new two-month ceiling makes it the most aggressive of the three on ephemeral conversations, though all three companies reserve the right to hold data longer for legal or safety reasons.
One genuinely new element softens the blow: Google is separating chat history from AI training. Until now, both were tied to a single "Keep Activity" setting. Users will now be able to keep conversations in their history without having them used to train Google's generative AI models, and vice versa. For people who previously had Keep Activity turned off, both new settings default to off; for those who had it on, history stays enabled and they can separately decide whether their conversations feed model training.
The Privacy Hub also states plainly that temporary conversations are not used to train AI models — a promise that matters more as assistants evolve from question-answering tools into agents that read files, send messages and act on behalf of users, expanding what a retained transcript can expose.
The practical takeaway is that "temporary" on consumer AI assistants now means weeks, not hours, and the retention ceiling keeps drifting upward as safety, abuse-fighting and incident-response teams claim a longer look-back window. Users who care should treat Gemini's ephemeral mode as a softer guarantee than its name implies — and check the new split settings, because the defaults differ depending on what their history toggle said before the update.
Comments (0)
Log in to join the discussion
Log InNo comments yet