Gemini will keep chats for about two months even with history off, up from 72 hours. Review where client data goes
Turning off history in the Gemini app will soon mean Google keeps those chats for about two months, so the rules for client data need a second look.
What happened
TechCrunch reports that Google is extending how long it keeps temporary Gemini chats and chats made with history turned off to around two months. Google's Gemini Apps Privacy Hub previously listed 72 hours. The change is rolling out globally in the Gemini app over the coming weeks. Google says the longer period is to respond to users, process feedback and protect its services, and that temporary conversations aren't used to train models. For comparison, TechCrunch notes OpenAI may keep ChatGPT temporary chats for up to 30 days and Anthropic keeps Claude incognito chats for 30 days by default.
At the same time, Google is splitting chat history and AI training into separate settings, which used to share one "Keep Activity" switch. Ars Technica explains that training is opt-out: you need to find the toggle under Activity in Gemini settings. The change applies only to future chats. Ars adds that data already used for training can include inputs, responses, uploaded files, tool calls, location and data from connected apps, that Google says it may keep that data for up to three years, and that reviewed chats already in training data won't be deleted.
Both articles describe the Gemini app. Neither covers the Gemini API or Google Workspace business accounts, so don't assume these numbers apply there.
My take
The likeliest exposure comes from everyday use. Staff paste a client's contract, a lead export or a support thread into whichever chat app is open, often with history off because they believe that means the data is gone. After this change, that belief is wrong for Gemini by about two months.
I'd write a short rule that names which AI tools may receive client data and under which account type, then confirm the retention and training terms for that exact account rather than relying on consumer defaults. For recurring work, move it out of chat apps and into an automation that calls a model API under documented terms, with the client data limited to the fields the task needs. That also gives you a log of what was sent, which a chat window never does.
More posts
- Open OCR models now read 1,000 pages for under $0.10, and better serving saved more than switching modelsNov 20, 2026
- Gemini 4 Neon costs $3 in and $15 out per million tokens and tops Zapier's AutomationBench at 56.3%, though it uses more output tokens per taskNov 20, 2026
- Gemini API's Tool Retrieval loads tool schemas only when the model asks for them, so agents with hundreds of tools send fewer input tokensNov 20, 2026
