Google Trains Gemini on Paid Personal Accounts by Default
Google AI subscribers must turn off chat history to prevent Gemini from using their conversations for model training and human review.
TL;DR
- Gemini personal subscribers' saved chat history is used to train Google AI models unless activity logging is turned off [^1].
- Disabling activity logging stops training but deletes chat history after 72 hours, leaving users with a trade-off between privacy and retention [^1].
Background
Large language model providers rely on continuous user interactions to train and refine generative models. Many users expect that paying for a monthly subscription exempts their data from being used for model training or human review. However, data retention policies vary widely across platforms and subscription tiers, leaving users to parse support documentation to understand whether their conversations remain private or become training material.
What happened
A post on Hacker News highlighted that personal Google AI subscribers cannot opt out of model training while keeping their chat history intact [^1]. If the "Keep Activity" setting is enabled, chat transcripts are used to train generative AI models and improve Google services [^1]. Google documentation confirms that a subset of these chats undergo human review [^1].
Google explicitly warns users in its documentation: "Please don’t enter confidential information that you wouldn’t want a reviewer to see or Google to use to improve our services, including machine-learning technologies." [^1]. Although Google states that human-reviewed chats are disconnected from user accounts, those specific transcripts are not erased when a user deletes their activity and are retained for up to three years [^1].
The only way to prevent Google from using Gemini chats for model training is to turn off "Keep Activity" in the application settings or use temporary chat sessions [^1]. However, turning off "Keep Activity" means users lose any chat sessions after 72 hours [^1].
Why it matters
Paying for a software subscription traditionally implies a commercial agreement where the user is not turned into the product [^1]. For Gemini personal subscribers, paying is not enough to stop Google from using conversation data for training [^1]. Users face a direct trade-off between preserving long-term chat history and protecting their privacy [^1].
This policy poses challenges for individuals handling sensitive information, personal notes, or proprietary ideas. Because human reviewers retain selected chats for up to three years even after activity deletion, confidential data entered into Gemini remains stored outside the user's control for extended periods [^1].
Practical example
Say Alex pays for a personal Google AI subscription to assist with everyday tasks and research. Believing that a paid plan keeps conversations private, Alex enters notes containing proprietary ideas into Gemini [^1].
Because Alex has the default "Keep Activity" setting turned on, Google uses those chat transcripts to train its generative AI models [^1]. A subset of Alex's chats is selected for human review [^1]. Google disconnects the chats from Alex's account, but retains the reviewed transcripts for up to three years [^1]. Even if Alex later deletes the chat activity, those human-reviewed transcripts are not deleted [^1].
To prevent future chats from being used for training, Alex must turn off "Keep Activity" in Gemini's settings [^1]. However, doing so causes Gemini to delete chat sessions after 72 hours, forcing Alex to choose between keeping chat history and preventing AI training [^1].