Human Reviewers Read Paid Gemini Chats, and Google Keeps Them 3 Years After Deletion
Gemini chats on paid personal plans like Google AI Pro and Ultra are used for model training and human review whenever Keep Activity is on. Reviewed chats survive deletion for up to three years, and switching training off erases your chat history after 72 hours.
- Gemini chats train Google's models by default, and a paid Google AI plan changes nothing.
- Human reviewers read a subset, and reviewed chats stay up to three years after you delete.
- Switching off Keep Activity is the only opt out, and it wipes your chat history in 72 hours.
Paying for Gemini does not keep your conversations out of Google's training data, and some of them are read by human reviewers. That reading of Google's own documentation hit 560 points on Hacker News on November 4, posted by a subscriber who said they assumed a paid plan meant not being the product. The post quoted the Gemini Apps Privacy Hub directly.
None of this is a policy change. Google announced it in August 2025, when it renamed the "Gemini Apps Activity" setting to "Keep Activity." The privacy notice was last updated on October 13, 2026, and the hub on October 26. What surfaced this week is that the hub draws no line between paid and free plans, and that most subscribers did not know it.
The notice splits on a setting, not a plan
Google AI Plus, Pro, and Ultra appear in the hub exactly once, in the list of data Google collects: "Subscription information: if you have a paid subscription to a Google AI plan, subscription related information." Nowhere does a paid plan change how chats are handled.
What changes the handling is one switch. With Keep Activity on, Google uses your activity to improve its services "including to train generative AI models," and "human reviewers (including trained reviewers from our service providers) review some of the data we collect for these purposes." Reviewers judge whether a response was inaccurate or harmful and propose a better one. Those proposals go back into the model.
The single real carve-out in the document is the work or school account. If you reach Gemini through a Google Workspace account your employer pays for, the Workspace privacy hub applies instead, and it promises no human review and no generative AI training without the customer's permission. A personal account with a Google AI subscription on it sits outside that promise.
Retention works like this:
- Activity is auto-deleted after 18 months by default. You can set 3 months, 36 months, or keep it indefinitely.
- Chats that went through human review are retained for up to three years, disconnected from your account before they reach a service provider.
- Those three years are not affected when you delete your activity. Google's stated reason is that the copy is already detached from the account.
That screenshot is Google's own, from the August 2025 announcement. The in-product text says keeping activity "helps improve Google services, including AI models," and adds that with the setting off, chats are still saved for 72 hours.
What opting out costs
Gemini has no switch that turns off training alone. To stop it you turn off Keep Activity, and then there is no chat history at all. New conversations do not appear in your activity list and sit in your account for only 72 hours. That is what the Hacker News author meant by losing your chats after three days.
The hub also lists what stops working with the setting off.
| Item | Keep Activity on | Keep Activity off |
|---|---|---|
| Model training | Used | Not used, unless you send feedback |
| Human review | Improvement and safety | Safety review continues |
| Chat history | 18 months by default | Gone after 72 hours |
| Workspace Connected Apps (Gmail, Drive) | Available | Unavailable |
| Tasks (remote browser, remote computer) | Available | Unavailable |
| Smartwatch calling and messaging | Available | Unavailable |
Source: Gemini Apps Privacy Hub, updated October 26, 2026
Two exceptions apply even with the setting off.
- Feedback. A thumbs up, a thumbs down, or a reported problem collects "the last 24 hours of your chats" along with any files and Connected Apps content in them. That bundle is reviewed by specially trained teams and retained for up to three years.
- Safety review. With Keep Activity off or in a temporary chat, Google still uses your chats "to respond to you and help protect Google, our users, and the public, including with help from human reviewers."
The reliable way out of training is a temporary chat. Temporary chats are kept out of your history, out of personalization, and out of model training, and they expire after 72 hours. You pick them per conversation, with Keep Activity left on.
If you connected Gmail or Photos
If you use the personalization features that wire Gmail, Drive, or Photos into Gemini, the Connected Apps documentation is worth a second read. Google says it does not train on your inbox or photo library wholesale. It trains on the summaries, excerpts, and inferences produced while answering you.
The same page adds a qualifier: if an email or file is short, or especially relevant to the prompt, the summary may be the email or file itself. A short email can enter training verbatim rather than as a summary. Google says it applies privacy-reducing techniques such as differential privacy before training. Disconnecting an app or deleting the original message does not retract data already sitting in your Gemini activity.
ChatGPT and Claude let you keep the history
The sharpest difference from the competition is the price of refusing. ChatGPT and Claude both let you turn off training while keeping your conversations.
| Personal account | Where training is turned off | History after opting out | Training and review retention |
|---|---|---|---|
| Gemini | Keep Activity off | Gone after 72 hours | Reviewed chats up to 3 years |
| ChatGPT | Data Controls > Improve the model for everyone | Kept | Not stated in public docs |
| Claude | Model improvement setting in privacy controls | Kept | Up to 5 years de-identified, if allowed |
Source: Gemini Apps Privacy Hub, OpenAI help center, Anthropic privacy center
OpenAI's help article says turning the control off leaves your history intact and excludes only new conversations from training. Anthropic revised its consumer terms in August 2025 to make Free, Pro, and Max users choose: allow training and conversations are kept de-identified for up to five years, decline and the existing 30-day retention stands. Both companies exclude incognito and temporary chats from training.
No region is exempt
The hub's only region-specific text does three things, and none of them is an opt-out. It names the provider (Google Ireland Limited in the EEA and Switzerland, Google LLC everywhere else), adds a legal-bases section for EU and UK data protection law, and tells Korea-based consumers that the Korean Location Terms of Service also apply. Singapore and the wider APAC region are not mentioned anywhere in the document, and no country is excluded from training or human review.
If you are in the EU or the UK, the GDPR rights the hub lists give you deletion and objection routes, but Google's position on reviewed chats is that they were disconnected from your account before a reviewer ever saw them. Under Singapore's PDPA you can withdraw consent, and the only mechanism Google documents for withdrawing it here is the same Keep Activity switch, at the same 72-hour cost. Why privacy of this kind is a design question rather than a policy one came up in our earlier piece on Meta AI's incognito mode.
If you use Gemini for work on a personal account, open the Gemini Apps Activity page and check whether Keep Activity is on. If you need the history and the Workspace connections enough to leave it on, start the conversations you would not want a reviewer reading, such as customer material or medical and legal questions, as temporary chats. For company documents, a corporate Workspace account is on better terms than any personal subscription you can buy.