Tell HN: Gemini trains on chat history, even for subscribers (personal accounts)
Google's Gemini is using all your saved chat history for AI training and human review, even if you're a paid subscriber, with no clear opt-out besides disabling history altogether. This revelation sparked a familiar HN debate about privacy, the true cost of 'free' (or even paid) services, and Google's ingrained data collection practices. Users are left to weigh convenience against contributing their conversations to the AI's ever-hungry maw.
The Lowdown
A Hacker News post revealed that Google's Gemini AI, including its paid subscription tiers, uses all saved chat history for training its generative AI models and for human review. This practice is enabled by default, and the only way for users to opt out is to completely disable 'Keep Activity,' which also means losing access to their chat history after 72 hours.
Key points from the story:
- Gemini app activity is explicitly used to improve Google services, including training generative AI models, if the 'Keep Activity' setting is on.
- A subset of these chats is reviewed by human reviewers, including Google's service providers.
- Chats reviewed by humans are retained for up to three years, even if the user deletes their activity.
- Google explicitly warns users not to enter confidential information that they wouldn't want a reviewer to see or Google to use for service improvement.
- The author expressed surprise, having assumed that a paid subscription would exempt their data from being used for training.
This means that users engaging with Gemini, regardless of their payment status, are implicitly consenting to their conversations being used to refine Google's AI and potentially being read by human eyes, unless they forgo the utility of persistent chat history.
The Gossip
Google's Grabby Gaze: Expectations vs. Reality
Many commenters expressed a distinct lack of surprise, citing Google's long-standing business model of data harvesting as the expected norm. The sentiment was often, 'What did you expect?' or 'If you're not paying top dollar, you are the product.' However, others voiced disappointment and a sense of betrayal, especially for paying subscribers who anticipated a higher level of privacy. The discussion highlighted a divide between those who are resigned to pervasive data collection and those who still hope for or demand better privacy practices, particularly when money changes hands.
Feature Frustration & Forced Forfeiture
A prominent theme was Google's 'all-or-nothing' approach, where disabling privacy-related settings like chat history cripples essential functionality across its ecosystem. Users reported that their Pixel phones become significantly less useful without Gemini's chat history enabled, and similar complaints surfaced about Google Assistant and YouTube's blank homepage without watch history. This coercive design was widely criticized as user-hostile, forcing individuals to choose between privacy and basic usability of services they pay for or rely upon.
Privacy Protections and Prudent Practices
Commenters explored alternatives and strategies for maintaining privacy in the AI era. Some noted that other major AI providers *do* offer opt-outs for training on personal chats or provide stronger privacy guarantees in their enterprise-tier services (e.g., Google Workspace, Azure OpenAI). The discussion also gravitated towards the ultimate solution: local, self-hosted LLMs as the only true way to ensure data sovereignty. Users shared tools and practices for keeping sensitive information away from cloud-based AI, underscoring a growing desire for personal control over data.