What happens to your chats with ChatGPT, Claude and Gemini?

When you type into ChatGPT, Claude or Gemini, your message travels to the company's computers, the model writes a reply, and the whole exchange is saved to your account so you can come back to it. What happens next depends on three things: whether you allow the company to use your chats to train future models (that is, to teach the next version from real conversations), how long the company retains (keeps) the data, and whether human reviewers may read some of it.

The rules changed a lot between 2023, when ChatGPT first added a switch to stop training on your history, and today. This is the state of things as of October 2026, for personal accounts. Work, school and enterprise plans follow separate contracts, covered at the end.

ChatGPT: one switch for training, a separate one for memory

In ChatGPT the training setting is called Improve the model for everyone, found under Settings, then Data controls. With it off, new conversations are not used to train OpenAI's models but still appear in your history. On a personal plan the same switch also covers Codex tasks. Turning it off does not delete anything; archiving a chat does not change its training status either.

Two exceptions are worth knowing. A temporary chat stays out of your history, does not create memories, and is not used for training, though OpenAI may keep a copy for up to 30 days for safety reasons. And if you press thumbs up or thumbs down on an answer, the whole conversation linked to that feedback may be used for training even after you opted out.

Memory, which lets ChatGPT remember facts about you across chats, is a separate control from training. Turning one off does not turn off the other.

Claude: you choose, and the choice sets how long data is kept

Since late 2025, Anthropic asks every user of Claude Free, Pro and Max (including Claude Code used from those accounts) whether their chats may be used to improve its models. The answer changes retention. If you say yes, new or resumed chats may be kept in a de-identified form for up to five years in training pipelines. If you say no, the usual rule applies: a chat you delete disappears from your history at once and from Anthropic's back-end systems within 30 days.

Incognito chats are never used for training, even if model improvement is on. Feedback sent with the thumbs buttons is kept for five years. Chats flagged by automated systems as breaking the usage policy may be kept for up to two years. Switching training off later stops the use of your chats in future training runs, but data already used in a model that has been trained stays there.

Gemini: one setting called Keep Activity

Google's controls sit in Gemini Apps Activity. With the Keep Activity setting on, your chats are saved to your account, used to improve Google's services including training generative models, and a sample is read by trained reviewers. Activity is deleted automatically after 18 months by default; you can choose 3 or 36 months, or no automatic deletion.

The detail that surprises people: chats that reviewers have looked at are disconnected from your account and kept for up to three years, and deleting your activity does not remove them. Google advises against typing anything confidential you would not want a reviewer to see.

With Keep Activity off, or in a temporary chat, your conversations are still held for 72 hours so Gemini can answer and Google can protect against abuse, but they are not used to train models unless you send feedback. Some features, such as connected Google Workspace apps, stop working when Keep Activity is off.

Side by side

How do you turn off training and limit what is kept?

Start with the training switch in each tool. In ChatGPT: Settings, Data controls, then turn off Improve the model for everyone. In Claude: Settings, Privacy, then turn off the toggle under Help improve our AI models; it works the same way on the web, desktop and mobile apps. In Gemini: open Gemini Apps Activity and turn Keep Activity off, or keep it on but shorten automatic deletion to three months.

Next, use the private mode for anything sensitive: temporary chat in ChatGPT and Gemini, incognito in Claude. Avoid the thumbs buttons on conversations you would not want stored longer. Delete chats you no longer need, and remember that turning off training does not delete what is already saved. Before pasting in a document, ask yourself whether you would be comfortable if a reviewer read it; if not, remove names, numbers and other identifying details first.

If the chats involve clients, patients or employees, a personal account is the wrong tool. By default OpenAI does not train on content from its Business, Enterprise, Edu and Healthcare workspaces, and Anthropic's business offerings, including Claude for Work and its API, sit under separate terms. Those plans also give an administrator control over retention and come with a written agreement, which makes it far easier to meet privacy laws such as Québec's Law 25.