AI
Google Separates Chat History From AI Training in Gemini
Users can now preserve their Gemini conversations without automatically contributing to Google's AI model training.
Ars Technica
Google is rolling out a significant adjustment to how it handles data in its Gemini chatbot—decoupling your conversation history from its use in training machine learning models. The move gives users finer control over what information feeds into Google's generative AI systems while still letting them access past conversations.
What's Changing#
Previously, opting out of AI training required turning off chat history altogether, eliminating the ability to revisit earlier conversations. Under the new framework, these two functions operate independently. Users will find separate toggles in Gemini's settings—one for preserving chat history and another for blocking use of those chats in model training.
The training opt-out appears as an opt-in mechanism: Google defaults to using your data to improve its models, but you can disable it by toggling the control under Activity in Gemini settings. Alternatively, navigate through Data and Privacy > My Activity > Other activity > Gemini Apps Activity in your Google account.
Important Limitations#
This change only applies going forward. Any Gemini interactions you've already had—including text inputs, responses, uploaded files, tool calls, location data, and connections from other apps—remain part of Google's training dataset. The company retains such data for up to three years to refine its AI capabilities.
Google has indicated it attempts to filter out personal information from Gemini activity logs, though the company offers no absolute guarantee of success in doing so.
Why It Matters#
Gemini's integration across Google's ecosystem means frequent, involuntary contact with the technology for many users. The new controls represent a step toward letting users access Gemini's core functionality without blanket consent to being part of Google's AI development pipeline.

