Google to split Gemini chat history from model training, so you can opt out without losing your history
Google is ending the choice between privacy and chat history in Gemini. Right now, a single Keep Activity setting controls both whether your conversations are stored and whether the model is trained on them, so opting out of training meant wiping everything. In the coming weeks, Google will replace it with two toggles: one for history with a retention period, and one for training on text and attachments. To keep your conversations and opt out of training, you will have to change the settings manually.
AI-processed from ZDNet AI; edited by Hamidun NewsImagine you want a chatbot not to use your conversations for training, but you are not willing to lose the chat history that holds your drafts, work notes, and useful prompts. Until now, you could not do that in Gemini. Google has announced that this will change in the coming weeks, according to ZDNET.
What is wrong with the current setting
Right now, Gemini has a single toggle, Keep Activity. It handles two things at once: whether your chats are saved (and for how long), and whether they can be used to train AI. These functions cannot be separated. If you want Google not to train the model on your conversations, you turn off Keep Activity and lose access to past chats. If you want to keep your history, you have to accept training.
For an ordinary user, this is annoying, and for people who use Gemini every day, it is also inconvenient. A chat with a model often serves as a work log: code reviews, text variants, calculations. Few people are willing to delete it for the sake of privacy, and many simply left everything as it was.
How it will work after the update
Google will replace Keep Activity with two independent settings:
- the first controls history: you can keep or delete past chats and choose how long they are stored;
- the second decides whether your chats, including all text and attachments, can be used to train models.
In an email to Gemini users, the company explained the meaning this way: you can keep your history and separately manage whether your data is used to train the generative model.
There is a catch in the transition. The new settings will not be set based on your preferences; they will copy the current state of Keep Activity. If it was off, both new settings will be off. If it was on, both will be on. This means that anyone who wants to keep their history and opt out of training must open the Gemini settings after the update and turn off the second toggle themselves. This will not happen automatically.
How competitors handle it
ZDNET author Lance Whitney compares Google's approach with other major players. In his words, almost every AI service has a separate training setting that users must turn off manually, but most let you do so without giving up your history.
- In ChatGPT, it is the "Improve the model for everyone" option, which you simply switch off.
- In Microsoft Copilot, there are separate settings called "Training on conversation activity" and "Training on voice conversations."
- In Claude, training on data is off by default and can be managed through the "Help improve our AI models" option.
In all these cases, conversation history remains available. In the author's assessment, Google was the only major company that restricted users who wanted to opt out of training. He adds that Google has long had a reputation for not being particularly careful with privacy, and he calls the new step a move in the right direction.
What this means in practice
For individual users, the takeaway is simple: when the update reaches your account, check your Gemini settings and see where the training toggle is set. If you work with sensitive data (client documents, internal code, personal correspondence), make that decision deliberately rather than leaving a value inherited from Keep Activity.
For teams and product managers, this is also an example of how user privacy settings get framed. When data storage and its use for training are tied to a single toggle, people either give up the functionality or agree to something they did not want. Separate settings reduce this conflict.
There are limits to keep in mind. The material gives no exact launch date, only "in the coming weeks." It does not say how this will work in corporate and work accounts on Google Workspace, or whether the update applies in all regions. These details will need to be checked in Google's official announcements once the change begins rolling out.
- Nov 21, 2026Gemini Notebook Can Now Build Interactive Websites from Your PDFs and Notes
- Nov 21, 2026FBI: ARTEX AI Tool Breaches Networks in Under Three Hours
- Nov 21, 2026OpenAI Starts Rolling Out dots Background Agents to Some $20 ChatGPT Plus Subscribers
- Nov 21, 2026Google to Buy 1.99 Billion Emails from Bankrupt Spirit Airlines for AI for $10 Million
Key Questions Answered
What is wrong with the current setting?
Right now, Gemini has a single toggle, Keep Activity. It handles two things at once: whether your chats are saved (and for how long), and whether they can be used to train AI. These functions cannot be separated. If you want Google not to train the model on your conversations, you turn off Keep Activity and lose access to past chats. If you want to keep your history, you have to accept training. For an ordinary user, this is annoying, and for people who use Gemini every day, it is also inconvenient.
How it will work after the update?
Google will replace Keep Activity with two independent settings: the first controls history: you can keep or delete past chats and choose how long they are stored; the second decides whether your chats, including all text and attachments, can be used to train models. In an email to Gemini users, the company explained the meaning this way: you can keep your history and separately manage whether your data is used to train the generative model. There is a catch in the transition.
How competitors handle it?
ZDNET author Lance Whitney compares Google's approach with other major players. In his words, almost every AI service has a separate training setting that users must turn off manually, but most let you do so without giving up your history. In ChatGPT, it is the "Improve the model for everyone" option, which you simply switch off.
Build & Ship Production AI Agents Without Boilerplate Code
Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.