How to Stop ChatGPT Training on Your Chats (and What Each Setting Does)

One switch does it, and it is not where most people look. Open Settings, select Data controls, and turn off Improve the model for everyone. OpenAI states that on personal workspaces with ChatGPT Free, Plus or Pro, data sharing is enabled by default, so if you have never touched this, it is on.

What that switch does not do is the part worth reading. It does not delete anything, it does not cover voice mode, it does not turn off memory, and there is one ordinary thing you can do afterwards that puts a whole conversation back into training anyway.

Key Takeaways

• The setting is called "Improve the model for everyone" and it lives under Settings then Data controls. Signed in, it applies to your account across web and mobile.
• It is on by default on Free, Plus and Pro. OpenAI says so in plain words. Business, Enterprise and Edu are the reverse: OpenAI does not train on their content by default.
• Turning it off keeps your history. OpenAI: it "does not delete or hide saved chats." Regular chats stay, archived chats stay archived, project chats stay in projects.
• A thumbs up or thumbs down overrides your opt-out for that conversation. OpenAI says the entire conversation attached to that feedback may be used for training even if you opted out.
• Memory is a separate setting, under Settings then Personalization. Opting out of training does not stop ChatGPT remembering things about you.
• Voice mode has its own control for sharing data for model training.
• Temporary chats are never used for training while they stay temporary, but OpenAI may keep a copy for up to 30 days for safety.
• Signed out, your choice is saved in that browser only, so it does not follow you to your phone or a private window.

The One Setting That Actually Stops Training

It is Improve the model for everyone, and here is where to find it on each platform. These are OpenAI's own steps from its data controls article.

When you are signed in, OpenAI says the setting "applies to your account across web and mobile devices," so you do this once and not per device. After you opt out, "new conversations and tasks won't be used to train OpenAI models."

Note the word new. The setting is forward-looking. It says nothing about conversations you have already had.

What Turning It Off Does Not Do

Four things, and each one catches people out. This is the table to read before you assume you are done.

There is one more worth knowing if you use ChatGPT's coding tool: OpenAI says Codex has a separate Include environments setting, and "Changing your ChatGPT setting or opting out through the Privacy Portal does not change that setting."

The Thumbs-Up Loophole

This is the single most useful thing on this page. If you opt out of training and then give a reply a thumbs up or a thumbs down, OpenAI may train on that whole conversation anyway.

Its wording, from how your data is used to improve model performance: "For individual plans (Free, Plus, and Pro), if you submit feedback on a response -- for example, by selecting thumbs up or thumbs down -- we may use the entire conversation associated with that feedback to improve our models, even if you have opted out of model training."

So the rating buttons are a per-conversation consent, not a quality signal with no strings. If you have opted out on purpose, the practical rule is simple: do not rate responses in conversations you would not hand over. Business, Enterprise and Edu work differently here, and better: OpenAI says response feedback on those plans "does not permit training on the associated conversation."

The same article names one more route most people never think about: "Conversations with OpenAI Support may be used to improve OpenAI services, including our models, if Improve the model for everyone is enabled in Settings."

Memory Is a Different Switch

Memory decides what ChatGPT knows about you. Training decides what OpenAI's models learn from you. OpenAI keeps them apart on purpose: "Memory controls how ChatGPT personalizes responses. Model-training controls determine whether eligible content can be used to improve OpenAI models. These are separate settings."

Memory lives under Settings, then Personalization, then Memory. Per OpenAI's memory article, the controls you may see there include Memory, Reference saved memories, Reference chat history, Memory summary, Manage and Saved memories, depending on your plan and region.

Two details with numbers attached. Turning off Reference saved memories also turns off Reference chat history, per OpenAI. And when you delete a saved memory, OpenAI "may retain logs of deleted saved memories for up to 30 days for safety and debugging purposes."

Our ChatGPT project memory explainer covers the project-level version of this, which has its own rules.

The Setting for One Private Conversation

Use a temporary chat when the issue is one conversation rather than your whole account. OpenAI's temporary chat article is unusually clear about what you get.

• It "is not used to improve OpenAI models while it remains temporary, whether or not you choose personalization."
• It does not appear in your chat history and does not create or update memories.
• You choose Personalized or Unpersonalized before your first message, and "You cannot change personalization after the conversation starts."
• OpenAI "may keep a copy of a temporary chat for up to 30 days for safety purposes."
• Saving it converts it to a regular chat, and from then on "the saved chat follows your account's personalization and model-improvement settings."

Start one by opening a new chat and selecting Temporary. The full set of controls is in our temporary chat explainer.

Where to Find All of This in One Place

OpenAI now has a Privacy Center that explains these settings rather than hiding them in a menu, and it is worth opening once. Per its privacy center article, it is "rolling out to signed-in ChatGPT Free, Go, Plus, Pro, and Business users on the web and in the ChatGPT apps for iOS and Android," and that rollout "does not include ChatGPT Enterprise, Edu, or ChatGPT for Healthcare."

To open it: on the web, account menu, then Help, then Privacy center. On mobile, Settings, then Privacy Center.

It does not change anything by itself. OpenAI: "Opening the privacy center does not change your settings, delete your data, or override your organization's policies." It groups your controls into personalized chats and ads, chat privacy, and your data, with a Manage button that jumps to the real setting. Our privacy centre explainer walks through each category.

Which Plan You Are On Changes the Answer

This is the part most how-to guides get wrong, because the default flips depending on what you pay for.

If you are on a work or school account and the toggle is missing or greyed out, that is why. OpenAI says available controls "depend on whether you are signed in, your plan, and your workspace settings."

What To Do, in Order

Do this if you want the short version. Settings, Data controls, turn off Improve the model for everyone. Then stop rating responses. Those two steps together are what the first one alone is usually assumed to do.

Also turn off memory if your worry is the app knowing you, not OpenAI's models improving. Settings, Personalization, Memory. The two settings do not substitute for each other.

Use a temporary chat for the one conversation you are actually worried about. It is the only option here that keeps a conversation out of both training and your history without changing your account.

Do not bother archiving things for privacy. OpenAI says archiving changes nothing about training. It is a tidiness feature.

Check your voice setting separately if you talk to ChatGPT rather than type.

If you are reconsidering which assistant to use over this, the best free AI chatbot in 2026 compares what each one does with your data, and Gmail's AI settings are the Google equivalent of this page.

Sources

• OpenAI, Data controls in ChatGPT, stamped "Updated: 2 days ago" and read live October 28, 2026, for the default state on Free, Plus and Pro, the per-platform steps, the signed-out browser scope, the saved-chat and archiving behaviour, the separate voice-mode control, the Codex Include environments setting and the memory-versus-training distinction.
• OpenAI, How your data is used to improve model performance, stamped "Updated: 8 days ago" and read live October 28, 2026, for the thumbs-up exception, the OpenAI Support conversations clause, the parental-controls route and the business-plan defaults.
• OpenAI, Memory in ChatGPT, stamped "Updated: 7 days ago" and read live October 28, 2026, for the Personalization menu path, the control names, the 30-day deletion schedule for chat-history memories, the 30-day log retention for deleted saved memories and the "Don't mention this again" scope.
• OpenAI, Temporary chat in ChatGPT, stamped "Updated: 9 days ago" and read live October 28, 2026, for the training exemption, the personalization choice being final, the 30-day safety copy and what saving a chat changes.
• OpenAI, Privacy Center in ChatGPT, stamped "Updated: 2 days ago" and read live October 28, 2026, for the plan rollout list, the Enterprise, Edu and Healthcare exclusion and the menu paths.