Gemini trains on chat history, even for subscribers, HN post says

A Hacker News post published on 4 November 2026 claims that saved Gemini chat history from personal Google AI subscribers can be used to train generative models, with no separate opt-out for paid accounts. The claim matters because many users assume a subscription changes how their conversations are handled, and the post sets out a different picture based on Google's own support pages.

The post, titled "Tell HN: Gemini trains on chat history, even for subscribers (personal accounts)", was submitted by user jameslk and had gathered 8 points and one comment at the time captured in the source. Its author says the issue was learned about too late, after assuming that paying for a personal subscription would keep chats out of training. The single comment in the thread, from r_lee, was brief and sarcastic in tone, offering no additional technical detail.

What did the Hacker News post claim about Gemini and training?

According to Hacker News, the central claim is narrow and specific: for a personal account with a Google AI subscription, saved chat history in Gemini can be used for training, and the only ways to avoid that use are to turn off chat history or to use temporary chat sessions instead of saved ones. The author presents this as a surprise rather than as a change to a previously published policy, and frames the paid subscription as insufficient on its own to change the default handling of saved activity.

The post does not claim that every conversation is automatically reviewed or that every subscriber is affected in exactly the same way. It focuses on the setting that controls whether activity is kept. When that setting is on, the author argues, the activity falls within the scope of data Google says it may use to improve its services, including by training generative AI models. When the setting is off, saved history is not kept in the same way, but the user loses features that depend on that history being retained.

That distinction is important for readers trying to assess the post. This is not a report of a new model launch or a new product feature. It is a user-reported reading of existing support documentation, shared in a discussion forum, and it should be treated as an interpretation of those pages rather than as an announcement from Google. The value of the post is that it draws attention to the connection between a paid plan, a saved-history setting, and training use — a connection that may not be obvious from the subscription page alone.

How does a personal subscription affect the settings?

According to the Hacker News post, a personal Google AI subscription does not, by itself, remove saved chats from training use. The author says they had expected payment to be enough to avoid being, in their words, turned into the product, and discovered otherwise when they looked at the relevant settings and support explanations. In other words, the subscription status and the activity setting appear to be controlled separately.

For a reader checking their own account, the practical question is therefore not only whether they pay, but which activity setting is active for their personal account. The post points to the "Keep Activity" setting in Gemini's settings as the control that needs to be changed. Turning that setting off is presented as the way to disable training on chat sessions, while leaving it on means activity is kept and, under the support wording quoted in the post, may be used to improve Google services for everyone, including through training generative AI models.

The post also mentions temporary chat sessions as an alternative way to avoid saved history. That option is different from changing the underlying setting: instead of relying on a persistent saved history, the user relies on a session that is not saved in the same way. The post does not provide a full comparison of temporary sessions and saved sessions, and readers who want to verify the current behaviour for their account type should check the support pages linked from the original thread and the settings in their own Gemini account, since account types and regional availability can differ.

What do the support pages say about human review and retention?

The Hacker News post cites three Google support explanations to support its reading. The first concerns use of Gemini app activity when the Keep Activity setting is on. The second concerns human review, and the third concerns how long reviewed chats are retained. Taken together, according to the post, these pages describe a system in which some chats may be seen by human reviewers, reviewed material may be disconnected from the account in Google's handling, and reviewed chats may be kept for a period even after activity is deleted.

On human review, the post says a subset of chats are reviewed by human reviewers. It also notes Google's statement that reviewed chats are disconnected from the user's account, while stressing that Google still warns users not to enter confidential information that they would not want a reviewer to see or Google to use to improve its services, including machine-learning technologies. For readers, the key point is the combination of disconnecting from the account and still warning against confidential input: the post treats that warning as the practical guidance to follow, regardless of how the data is linked internally.

On retention, the post says chats reviewed by human reviewers are not deleted when a user deletes their activity, and are instead retained for up to three years. That is a longer retention period than many users might expect from a delete action, and it is one of the details that makes the post notable. It also explains why the author emphasises the warning about confidential information: deletion of activity and retention of reviewed material are described as separate processes, so deleting a chat does not necessarily end the retention of a reviewed copy.

What is the trade-off if you turn off Keep Activity?

According to the Hacker News post, turning off Keep Activity is the step required to disable training on chat sessions, but it comes with a clear trade-off. Once the setting is off, the user will lose any chat session after 72 hours. That means the convenience of a long-lived, saved history is exchanged for a shorter retention window and for keeping those sessions outside the training use described in the support pages.

For occasional users, that trade-off may be acceptable. For people who rely on Gemini as a long-term working memory — returning to earlier threads, building on previous answers, or keeping a record of decisions across a project — losing sessions after 72 hours could be a significant cost. The post does not suggest a middle option for personal subscribers, such as keeping full history while opting out of training, and it presents the choice as binary: keep activity and accept the described use, or turn it off and accept the loss of saved sessions.

Readers should also separate this personal-account discussion from workplace or education accounts, which may be governed by different terms. The post is explicitly about personal accounts, and it does not describe how institutional settings, administrator controls, or other Google services handle similar data. Anyone using Gemini for client work, research, or internal company material should check the terms that apply to their specific account type rather than relying on this post alone.

How can you explore this topic in Grafy?

There is no Grafy app that changes Gemini's settings for a Google account, and this story does not describe a Grafy feature for managing another provider's data controls. What a reader can do in Grafy is use the tools to compare how different models and services describe their handling of user data, and to keep notes or research separate from a single provider's chat history. Grafy Chat lets you chat with many AI models side by side, which can be useful for comparing answers and for asking the same privacy or policy question to more than one model before deciding where to put sensitive material. Reader can help you read, search and question PDFs and papers with AI, including saving your own research notes and source material in a workspace you control rather than in a single chat thread.

The broader pattern of training use being reintroduced or clarified after users report it is not unique to this post. Grafy News has previously covered a related report in which users said OpenAI appeared to re-enable training on Codex chats — see OpenAI appears to re-enable training on Codex chats, users report. Reading the two reports together gives a more complete picture of a recurring issue for developers and researchers: subscription status, saved history, and training use are often governed by separate settings, and the details are usually in support pages rather than in product marketing.

For creators, developers and researchers, the practical lesson from the Grafy News Desk is to treat chat history as data until you have verified otherwise for your account type. Check the activity setting, avoid entering confidential information in any service whose support pages warn against it, and keep a separate record of important prompts, outputs and decisions — for example in Grafy — so that a change in retention or training policy does not erase your work.

Frequently asked questions

Does a paid personal Google AI subscription automatically opt Gemini chats out of training?

No. According to the Hacker News post, a personal Google AI subscription does not by itself opt saved chats out of training. The post says training use depends on the Keep Activity setting, so subscribers with that setting on should check their Gemini settings rather than assuming payment alone changes how saved history is used.

How can a personal subscriber stop Gemini from using saved chats for training?

According to the post, a personal subscriber can turn off Keep Activity in Gemini's settings, or use temporary chat sessions instead of saved ones. Turning Keep Activity off means chat sessions are lost after 72 hours, so the user trades long-term saved history for keeping those sessions outside the training use described in the support pages.

Are chats reviewed by human reviewers deleted when activity is deleted?

No. According to the support page cited in the post, chats reviewed by human reviewers are not deleted when activity is deleted. Instead, they are retained for up to three years, which is why the post emphasises Google's warning not to enter confidential information that a reviewer should not see.

Source: Hacker News — https://news.ycombinator.com/item?id=50363461 (published 2026-11-04)

---

Source: Hacker News — front page — Tell HN: Gemini trains on chat history, even for subscribers (personal accounts)