Yes: if the Keep Activity setting is on, Google says your Gemini chats can be used to train its generative AI models, and a subset can be read by human reviewers. The claim went around Hacker News in a "Tell HN" post titled "Gemini trains on chat history, even for subscribers (personal accounts)". The poster, who wrote that they had assumed a paid plan meant they were not "the product", pointed at Google's own help pages. We read those pages. The mechanism is documented, and it is stricter than many users expect.
This post separates what Google's pages say from what Hacker News commenters reported, and shows how to change the setting.
TL;DR: the questions people ask #
| Question | Answer from Google's Gemini Apps Privacy Hub (last updated October 26, 2026) |
|---|---|
| Is my chat used to train Gemini? | Yes, if Keep Activity is on |
| Can a human read it? | A subset of chats is reviewed by humans, including trained service providers |
| How long are reviewed chats kept? | Up to three years, disconnected from your account |
| What if I delete my activity? | Chats already reviewed are not deleted that way, per the Hacker News poster quoting Google; the hub says reviewed chats are retained up to three years |
| How do I opt out of training? | Turn off Keep Activity, or use temporary chats |
| Does opting out remove all storage? | No. Chats are still kept with your account for 72 hours |
| Does a paid plan change it? | The pages we read do not say a personal paid plan exempts you |
Agents that act on your behalf need rules. See how explainx helps secure your agents →
What Google's help page actually says #
The Gemini Apps Privacy Hub is the primary source. Several lines matter.
On human review, it says: "A subset of chats are reviewed by human reviewers (including Google's trained service providers) to help improve Google services." Reviewers assess whether responses were low-quality, inaccurate or harmful and suggest better ones. Chats are also reviewed to keep Gemini safe and to address violations of Google's terms, including its generative AI prohibited use policy. "Reviewed chats are retained for up to three years."
On the setting, the page says that with Keep Activity on, Google uses your activity "to provide, develop, and improve its services (including training generative AI models)", with human reviewers' help. Data reviewed by service providers "is disconnected from your account and saved for 3 years".
With Keep Activity off, "your future chats won't appear in your Activity, and won't be used to train our AI models, unless you choose to send Google feedback." They are still saved for 72 hours so Gemini can respond, process feedback and "protect Google, its users, and the public". Temporary chats "are not used to train Google's AI models."
And the warning that triggered the thread: if the setting is on, "don't enter data that's confidential or that you wouldn't want a reviewer to see or Google to use to improve its services."
Consent bowls with a pebble being placed, picturing a user deciding whether chats may be used for Gemini training
Does paying change anything? #
This is the part the thread's title claims, and the evidence is narrower than the headline. The Hacker News poster says that with a personal Google AI subscription there is "no way to opt out" except turning off chat history or using temporary chats. That matches the hub: the control is the Keep Activity setting, and we found no line on the page that ties it to plan tier.
What we could not verify is how consumer plans differ from business or education products. Google's pages describe Workspace-connected features, and the separate Connected Apps page says data from connected apps, such as Gmail, Drive or Photos, is used to personalize Gemini and to "improve Google services, including training generative AI models, for everyone". Admin-managed Workspace accounts follow different terms, which we did not review. If you use Gemini through work, ask your administrator rather than relying on a consumer page.
The trade-off: memory versus privacy #
Keep Activity does two jobs. It feeds training and review, and it keeps your chat history and powers features that remember you. Turning it off therefore costs something. Chats vanish from your Activity after the 72-hour window. A Hacker News commenter, advenn, wrote that if you want persistent memory you have to enable Keep Activity, and that this is why they do not use Gemini for anything serious or personal.
Another commenter, richrichardsson, reported a worse case on a Pixel phone: with the setting off, voice requests to the assistant to set a reminder did not work, and the assistant said it could not act on the phone unless chat history was enabled. That is one user's report, not something we tested, and Google may have changed behavior since. It does show that the setting reaches beyond the web chat.
What developers said on Hacker News #
The thread had about 72 points and 15 top-level comments when we read it. These are users' views, not verified facts.
- "Everyone does this."icantevenhold assumed all providers use chats no matter what they say, and233mhz advised acting as if everything you type is stored forever and could be reused if terms change.jbrooks84 made a similar point.
- Training is not the only use.fweimer argued that even without training, Google's terms let it use content to operate services and build new features, so the training exclusion is "a distraction".
- Human review is expected.viraptor said they trust no provider, but that reviewing a subset of chats is normal where a provider could be liable, and that the poster may be reading the terms too harshly. Others disagreed with that tone.
- Consent for what is typed.realsarm noted that people paste private material into Gemini and NotebookLM without a second thought.
- Dismissals and jokes. Several comments were jokes or said they simply use other chatbots. We left those out.
Not one commenter disputed the hub quotes. The disagreement was about whether this is surprising, not about what the pages say.
How to change the setting, step by step #
- Open Gemini settings and find the Keep Activity control (menu names change, so use Google's help page to locate it).
- Turn Keep Activity off to stop future chats from being used to train models and from entering human review for improvement. Remember that feedback you submit is still used.
- Use a temporary chat for anything sensitive when you keep Activity on.
- Delete past activity if you wish, but note that deleting activity does not remove data in other Google services, and reviewed chats follow the three-year rule.
- Check audio and Gemini Live separately. The hub has its own question on whether Google uses your audio and Live videos and screenshares to improve its services, so read that answer too.
A checkbox beside an exit door, showing the opt-out route for Gemini chat training
What this means for builders and teams #
If you build with or around Gemini, treat the consumer app as a place where content may be seen by a reviewer. Do not paste customer data, credentials, source code under NDA or unreleased research into a personal account with Keep Activity on. Use the API or a managed Workspace plan for work, and read the data terms that apply to it.
For agents that act on your behalf, the stakes rise. Connected Apps, remote browsing and screen automation all pass more personal data through the model. If you run agents with real access, a guard that blocks dangerous actions is a sensible layer, and [AgentBeam](https://agentbeam.com/?ref=explainx) is the agent security platform from the explainx.ai team built for that.
Where this sits in the wider pattern #
The question of "paid but still used for training" is not unique to Google. We covered a very similar issue when OpenAI said it will train on personal Daybreak sessions even with the opt-out off, and Hacker News users reported that a Codex "improve the model for everyone" setting came back on. Google is also facing a European question about training data in the EU court adviser's opinion on Gemini and press publishers. And Gemini's plan structure keeps shifting, as when free and AI Plus users lost Flash and Pro models on October 9.
The common thread: a subscription buys capacity and features. Whether it buys privacy depends on a setting, and that setting often sits in a place most people never look.
What we verified and what we did not #
- Verified: the quoted passages in Google's Gemini Apps Privacy Hub (last updated October 26, 2026), including Keep Activity, human review, three-year retention and the 72-hour window.
- Not verified: whether any plan tier, region or account type changes these terms. The hub does not say, and we did not test accounts.
- Reported by users, not tested: the Pixel reminder behavior, and the claim that deleting activity does not remove reviewed chats (the hub says reviewed chats are retained up to three years).
Check the live pages before you decide, because Google edits them.
Related reading #
Details are accurate as of November 4, 2026. Google updates its help pages often; check them before you rely on this summary.