Google is changing how long Gemini chat history is kept when users have actually switched history off. Temporary chats and conversations with history turned off will in future be stored with the Google account for around two months. Until now, according to an archived version of Google’s privacy page, the period was 72 hours. At the same time, Google is separating two settings that were previously linked: saving chat history and using chats to train AI models. The changes will roll out globally in the Gemini app over the coming weeks.
Gemini chat history: What is changing
Google announced the changes on November 16, 2026, in a short blog post. The post makes two points. First, users will be able to keep their Gemini history without it being used to train generative AI models. How long history is saved can still be configured, and individual chats can still be reviewed and deleted as before.
Second, Google says it wants to make Gemini “safer”. If users turn the new History setting off, their future chats will be saved for around two months. According to Google, they are used to respond to the user, process feedback and “protect Google, its users, and the public.” The general justification is that AI systems are becoming more capable and that Google continues to invest in safety and security measures.
Until now, both functions were tied to a single setting called “Keep Activity,” as TechCrunch reports. Anyone who wanted to keep their history also agreed to its use for training.
What Google’s privacy notice says
The Gemini Apps Privacy Hub goes into more detail. It describes what temporary chats and chats with history turned off are used for during those roughly two months:
- Responding: Gemini uses the last 24 hours of the chat as context for its responses.
- Protection and resilience: Google stores the chats with the account so that they are available in case of a potential system failure, and uses them to protect Google, users and the public.
- No training without feedback: According to Google, temporary chats are not used to train AI models. The same applies with history turned off, as long as no feedback is submitted.
- With feedback: Anyone who submits feedback with history turned off also shares context, including the last 24 hours of chats along with content such as attachments and data from connected apps.
One point deserves particular attention: according to the Privacy Hub, stored chats are used “with help from human reviewers” to protect Google and its users even when history is turned off. Regardless of settings, Google uses chats to respond and for protection purposes, including with human review. Chats that human reviewers have looked at are also not deleted when users delete their history; instead, they are retained for up to three years.
For regular history, the familiar options remain unchanged: by default, Gemini auto-deletes entries after 18 months, and users can also choose 3 or 36 months or indefinite storage.
Which defaults apply
According to TechCrunch, Google is carrying over users’ previous choices. For those who had disabled Keep Activity, both new settings will be turned off. Those who had it enabled will keep their history and can now decide separately whether their conversations may be used for training.
Google did not answer TechCrunch’s questions about what prompted the change or whether it would affect how conversations are accessed or reviewed, and instead pointed to the blog post.
How it compares with OpenAI and Anthropic
At around two months, Google now keeps temporary chats significantly longer than its direct competitors. TechCrunch notes that OpenAI may keep temporary chats in ChatGPT for up to 30 days, while Anthropic retains incognito chats in Claude for 30 days by default. All providers reserve the right to keep data longer for legal or safety reasons or to investigate policy violations.
The jump from 72 hours to roughly 60 days is therefore large even by industry standards. Anyone who has seen temporary chats as a way to use Gemini without leaving lasting traces in their account should reconsider that assumption.
Assessment
Separating history from training is genuine progress. Until now, users had to choose between convenience and privacy: anyone who wanted to find earlier chats again also released them for training. That link is now gone.
The longer retention with history turned off, by contrast, goes in the other direction. Google justifies it with safety and protection against system failures but gives no concrete reasons why 72 hours are no longer enough. The wording that chats are used to protect “the public” and with human reviewers is broad. It fits a trend we are seeing at other labs as well: the more capable the models, the more providers justify monitoring and data retention with abuse prevention. OpenAI, for example, is currently bringing outside assessors into its own lab to evaluate the risks of new models.
Google’s blog post does not say whether different rules apply to users in the EU or to business accounts. It refers to a global rollout in the Gemini app.
What users can do now
- Check your settings: Once the new switches are available, it is worth reviewing the Gemini history settings. History and training can then be controlled separately.
- Do not treat temporary chats as ephemeral: Temporary chats will also be stored with the account for around two months. Truly confidential information still does not belong in an AI chat.
- Give feedback deliberately: With history turned off, submitting feedback to Google can mean that the last 24 hours of the conversation, including attachments, are used to improve Gemini.
- Keep an eye on connected apps: Data from connected services can be part of the context. Anyone linking Gemini to email, calendar or files should check which connections are really needed.
For Gemini itself, this is the second major change within a few days. Only last week, Google introduced a new agent model, Gemini 4 Hydrogen. The more tasks users hand over to such agents, the more important the question of how long the resulting data is stored becomes.




