Does ChatGPT train on your conversations?

For consumer ChatGPT accounts (Free, Plus, and Pro), the default is yes. A setting called “Improve the model for everyone,” found in Settings → Data Controls, is switched on by default, and while it is enabled, the conversations you have, files you upload, and feedback you give can be used to train and refine OpenAI’s models.

Where to turn it off. In the ChatGPT app or website: Settings → Data Controls → Improve the model for everyone, then toggle it off. This applies going forward: new conversations after you disable it are excluded from training. It does not retroactively remove anything already used to train a model before you changed the setting, and OpenAI generally retains data for a period even after deletion for safety and abuse-monitoring purposes, so turning the toggle off is best understood as “stop future use,” not “erase past use.”

Temporary Chats work differently, and are opt-in. Starting a Temporary Chat (rather than toggling the account-wide data control) keeps that specific conversation out of your history, out of memory, and out of training, and it is automatically deleted after a defined retention window. The trade-off is that you have to choose it each time; it is not a persistent setting you turn on once. It is the more precise, session-by-session tool, while the Data Controls toggle is the persistent, account-wide one.

Business and enterprise tiers are excluded by default. ChatGPT Team, Enterprise, and Edu workspaces, along with API usage through OpenAI’s developer platform, are not used to train OpenAI’s models by default, and this is a contractual data-usage commitment rather than a toggle you need to manage yourself. If you use ChatGPT through a workplace or school account, you may already be covered without needing to change anything, though it is worth confirming with whoever administers that workspace, since some organizations still route through personal-tier accounts.

Worth checking either way. Because this is opt-out rather than opt-in, it is easy to have used ChatGPT for months or years without realizing conversations were feeding into training data. Checking the Data Controls setting takes under a minute, and using Temporary Chat for anything sensitive is a reasonable habit regardless of your account-wide setting.

For memory controls, connector permissions, and the rest of a reasonable ChatGPT privacy baseline, see the full ChatGPT privacy guide.