Gemini Trains on Paid Subscribers' Chats With No Opt-Out

Google Gemini uses saved chat history to train its generative AI models by default for personal accounts, including paid subscribers, with no opt-out available short of disabling the assistant's memory entirely. The policy, surfaced in a Hacker News thread that drew 318 upvotes on November 4, 2026, points to Google's own support documentation and has reignited debate over transparency in consumer AI products.

How the Training Policy Works

According to Google's disclosure, Gemini Apps activity is used to improve Google services for everyone, including by training generative AI models, if the Keep Activity setting is on. Keep Activity defaults to on for adults. Turning it off stops future chats from being used to improve Google AI, but the same toggle also deletes stored conversations after 72 hours. That 72-hour window eliminates long-term memory for the assistant, effectively neutralizing one of the product's core features.

The policy applies to personal accounts regardless of subscription tier. Paid Gemini subscribers face the same data-use terms as free users. Google's commercial offerings, including the Gemini API, Vertex AI, and Workspace tiers, operate under separate data agreements that exempt customer data from training.

Human Review and Retention

A subset of Gemini chats goes to human reviewers for quality assurance. Google's support page warns users: 'Please don't enter confidential information that you wouldn't want a reviewer to see.' Those reviewed copies are retained for up to three years, even after a user deletes their activity. The company does not disclose what percentage of conversations are selected for review or how reviewers are instructed to handle sensitive information.

No Separate Opt-Out for Training

The central complaint from Hacker News and privacy advocates is that Google offers no toggle to stop training without also disabling chat memory. Users who want to preserve long-term context in their Gemini conversations must accept that those conversations will be fed into training data. Users who want to block training must sacrifice the assistant's ability to remember past interactions beyond 72 hours.

Critics note that this binary choice is not surfaced clearly at checkout for paid personal subscribers. The policy lands the same week Google switched Gemini to opt-in for schools after backlash over consumer-style data handling defaults in educational settings.

Availability and Policy Status

Officially launched on November 4, 2026, this policy disclosure was already in effect but gained renewed attention through the Hacker News post linking to Google's own terms. Google has not announced plans to introduce a separate opt-out toggle or change the Keep Activity behavior.

What This Means

Google's approach to training data in Gemini personal accounts reflects a broader tension in consumer AI: companies want longitudinal conversation data to improve models, but users expect control over how their inputs are reused. The all-or-nothing toggle puts paid subscribers in the same data-use bucket as free users, with no middle ground for those who value both privacy and assistant memory. As AI products mature, the pressure to offer granular data controls will likely intensify, especially in markets with strict privacy regulation.