A post on Hacker News titled "Tell HN: Gemini trains on chat history, even for subscribers (personal accounts)" went up at about 2:10 AM PT on November 4 and had 644 points and 168 comments by about 9:30 PM PT. Its author, jameslk, wrote that he learned "too late" that paying for Google's AI plan did not keep his chats out of training. Most of the argument since has been less about whether that is true and more about whether it is surprising, fair, or different from what other AI companies do. This roundup collects what people said on Hacker News and X, and checks the main claims against Google's own help pages. Reddit cannot be read from our tools, so it is not covered.
The Hacker News thread is the center of this story. X was much quieter, and most of what we found there came from very small accounts.
What Google's pages say
The claims in the thread match Google's Gemini Apps Privacy Hub, which shows a last-updated date of October 26, 2026, with the Privacy Notice section dated October 13. With a setting called Keep Activity on, Google says it uses saved activity to provide and improve its services, "including training generative AI models," with help from human reviewers. Chats that reviewers read are retained for up to three years and are not deleted when you delete your activity. Google says chats are disconnected from the account before they go to service providers, and it asks users not to enter confidential information.
Google's activity page says Keep Activity is on by default for people 18 or older. With it off, Google says future chats are not used to train its models unless you send feedback, but they are still kept for 72 hours. Temporary chats are not used to train either. Turning the setting off also makes some connected apps, such as Google Workspace, unavailable. The hub says Gemini chats are not used to show ads, and that work or school accounts can fall under different data terms.
The part that stirred people up is the bundle. On a personal account, the setting that keeps your chat history is the same setting that allows training.
The original complaint and the "everyone does this" reply
Thread author jameslk made the case in a follow-up comment that this goes beyond legal housekeeping. He wrote that Google is "claiming to use chat transcripts for both training and for human review to improve training, not just for liability," that "They don’t define “subset”. It could be most chats," and that the setting applies to "free and paid user subscriptions, by default."
Several people said the headline should not shock anyone. giwook wrote "Is this really surprising to anyone?" and said he had assumed Google was "hoovering up anything I type into its AI." WarmWash wrote that "Anytime you aren't paying full price for something, you are paying the rest with something else." 233mhz advised treating anything you type as stored, "no matter how much money you pay."
Others pushed back on reading the terms as sinister. viraptor wrote "I don't trust any of the providers, but at the same time I think you're reading into the terms too much for wrong reasons," and said human review and three-year retention make sense for chats that could create liability: "They're being retained for legal purposes." swatcoder replied that even if that is the legal team's intent, users can still decide they "may want to use Gemini less/differently or look at other providers."
Is Google the odd one out
This was the most contested factual point. reasonableklout wrote that "most of the products today like Claude Code / Chat allow you to opt-out of training without also disabling important features like chat history." therealmarv wrote that Google "should offer history + not training (every other big AI company is offering that)," and that "The only way to get that currently is to use Gemini in Google Workspace." Another user, Havoc, pointed to Gemini enterprise, which he said has "a note under each chat session explicitly saying it’s not used to improve the models." That fits Google's line that work and school accounts can have different terms, though we did not check the enterprise terms themselves.
fny argued the human review part is not unique to Google: "This is true of all the frontier model providers even if you disable activity tracking." We did not verify that against other providers' policies, so treat it as a commenter's claim. fweimer took a wider view, quoting Google's terms on using content to operate the services and writing "The training exclusion has always been a distraction."
For readers who use other assistants, our guide to ChatGPT data controls covers the equivalent switch there, and our piece on whether DeepSeek is safe to use covers a different set of privacy trade-offs.
Complaints about features that depend on the setting
A second thread of complaints was about what stops working when Keep Activity is off. richrichardsson wrote that on a Pixel phone "Gemini is completely useless without enabling this setting," and said that after he asked it why a reminder never fired, the assistant told him "I can't do anything with your phone unless you enable chat history." Google's Privacy Hub does say some connected apps, including Phone, Messages and WhatsApp, stay available on Android with the setting off, so what exactly breaks may depend on the feature and device. devsda called it a pattern: "Google has this pattern of all or nothing approach."
Who said they were leaving
A few people said they were cancelling or already had. kekqqq wrote "I kept paying for the subscription for a year and a half, but I think I will stop now." EbNar wrote "This is one of the reasons why I unsubscribed from Gemini Pro." wat10000 wrote that "The Qwen 3.8 I'm running on a spare computer is definitely not training on my inputs," as the case for local models. These are individual comments, not a measure of how many subscribers will act.
What X said
X was thin. Our searches turned up about a dozen small accounts posting near-identical explainers with the same opening line, that paying for Gemini does not stop Google from training on your chats, followed by steps to switch off Keep Activity. The ones we found had under 100 views each at about 9:30 PM PT. A few added a line of their own. AviPilcer wrote "Private should be the default." sabislabi wrote "Private stuff? Temporary chat first." The most viewed post we found that discussed the story was keithrull's AI roundup post, with 620 views and 10 likes, which listed "Gemini trains on paid chats unless you opt out" as one of three items.
The sharpest complaint came from TKelnar, who wrote that "Every other AI company offers the option to turn off model training on user data." That post had 4 likes and 110 views. He sent the same text at least nine times between about 2:56 AM and 12:10 PM PT, replying to different Google and Gemini accounts, including this one. He also wrote that "As of November 4th, you changed the terms." We could not confirm a change on that date. The Google pages we read carry last-updated dates of October 13 and October 26, and the Hacker News author linked Google's own support pages. We did not find a reply from Google on X or in the thread.
What to take from it
The facts are not in dispute: on a personal Gemini account, saved chats can be used for training and human review unless Keep Activity is off or you use a temporary chat, and the Google pages we read describe no exception for paid plans. The disagreement is over whether Google handles it worse than rivals, since several commenters say other assistants let you keep history and opt out of training, and a few say the human review part is common everywhere. If you use Gemini for anything private, the options Google lists are to turn Keep Activity off, start a temporary chat, or shorten the auto-delete period from its 18-month default. Each of those costs something, such as lost history or unavailable connected apps. Check the current setting in your own account, because Google's pages are updated often.
Sources
- Hacker News: Tell HN, Gemini trains on chat history, even for subscribers
- Gemini Apps Privacy Hub (Google)
- Manage and delete your activity in Gemini Apps (Google)
- TKelnar on X: complaint about model training options
- AviPilcer on X: opt-out explainer
- sabislabi on X: temporary chat advice
- keithrull on X: AI roundup post
