Does ChatGPT train on your data? What changed in 2026

The Morse teamPublished October 15, 20269 min read

Yes. If you use ChatGPT on a personal plan (Free, Go, Plus, or Pro), OpenAI may use your conversations to train its models unless you turn off "Improve the model for everyone" in Settings, then Data controls. Temporary chats and business accounts are not used for training, but opting out only covers new conversations, and it does nothing about ads, human review, or court orders, which is where most of 2026's news happened.

Everything below was checked against OpenAI's own pages on October 15, 2026. Where something is only alleged, we say so.

Does ChatGPT train on your chats by default?

For personal accounts, yes. OpenAI's page on how your data is used to improve model performance says: "ChatGPT, for instance, improves by further training on the conversations people have with it, unless you opt out." For "services for individuals such as ChatGPT, Sora, or Operator," it adds, "we may use your content to train our models."

Business products are the other way round. The same page says: "By default, we do not train on any inputs or outputs from our products for business users, including ChatGPT Team, ChatGPT Enterprise, and the API." OpenAI's data controls help page says the same of Business, Enterprise, Edu, and ChatGPT for Healthcare workspaces.

On a personal plan, paying for Plus or Pro does not change the training default. It only removes ads.

How do you stop ChatGPT from training on your data?

OpenAI's data controls page (updated in late September 2026) gives these steps.

On the web, signed in:

1. Open your account menu.
2. Select Settings.
3. Select Data controls.
4. Select "Improve the model for everyone", turn it off, and select Done.

On iOS and Android: open the sidebar, tap your profile icon to open Settings, select Data controls, and turn off "Improve the model for everyone". When you're signed in, the setting follows your account across devices.

Signed out: open Settings and turn the same switch off. OpenAI says "your choice is saved in that browser," so it resets if you clear the browser or switch devices.

Everywhere at once: OpenAI's privacy portal has a "do not train on my content" request, which also covers Sora. Sora's in-app setting is separate from ChatGPT's.

Four things to know about the switch:

• It isn't retroactive. OpenAI says "once you opt out, new conversations will not be used to train our models." Chats from before the switch aren't pulled back.
• Feedback overrides it. "If you choose to provide feedback, such as selecting thumbs up or thumbs down on a response, the entire conversation associated with that feedback may be used to train OpenAI models." That applies after you opt out.
• It doesn't delete anything. Your history stays where it was. Deleting is a separate step.
• Memory is a different switch. OpenAI calls memory and model training "separate settings."

Does ChatGPT train on temporary chats?

No, not while they stay temporary. OpenAI's temporary chat page says a temporary chat "is not used to improve OpenAI models while it remains temporary, whether or not you choose personalization." It also stays out of your history and doesn't create or update memories.

The fine print:

• It's still kept for up to 30 days. "OpenAI may keep a copy of a temporary chat for up to 30 days for safety purposes."
• Saving it changes the rules. If you save a temporary chat, it "becomes a regular chat" and follows your training setting.
• Personalized temporary chats can read your memories. You choose Personalized or Unpersonalized before the first message, and can't change it after.
• GPT actions leave OpenAI's rules behind. Data a GPT sends to a third party "is governed by the recipient's privacy policy," which may keep it longer than 30 days.

Some 2026 coverage said ads could still draw on temporary chats. OpenAI's ads help page says they don't: "Temporary Chats will not show ads."

Does ChatGPT train on deleted chats?

Deleting a chat removes it from your account right away, and OpenAI's chat retention page says it's scheduled "for permanent deletion from its systems within 30 days, unless the chat was already de-identified and disassociated from your account or OpenAI must retain it longer for security or legal obligations."

OpenAI's privacy policy explains the first exception: deleted data is removed within 30 days unless "it has already been de-identified and disassociated from your account when you allow us to use your Content to improve our models." In practice, if training was on when you had the chat, a de-identified copy may already sit in training data, and deleting the chat later doesn't reach it. Opt out before the sensitive conversation, not after.

The second exception, legal obligations, covers the court order described below.

Does ChatGPT sell your data?

OpenAI says no. Its privacy policy states: "We don't 'sell' Personal Data." The same section says that, depending on your choices, it "may share limited data with select marketing partners" to advertise OpenAI's own products on other sites, which some state laws treat as targeted advertising. You can turn that off with the marketing privacy control in your account settings or with Global Privacy Control.

Ads inside ChatGPT are a separate question. OpenAI's ads help page says: "We do not share your conversations with ChatGPT with advertisers, and we never sell your data to advertisers." Advertisers get "aggregated, non-identifying information," such as views and clicks. But the ads are chosen using your words. The ad system considers "the context and intent of your current conversation," and, with ad personalization on, "past chats and memory." Turning personalization off "doesn't remove ads."

OpenAI doesn't sell your chats, but on Free and Go it reads them to pick ads, and on any personal plan it may train on them unless you opt out.

What changed in 2026?

Our timeline, built from primary sources and dated reporting:

Ads on Free and Go

Plus, Pro, Business, Enterprise, and Edu accounts have no ads, according to OpenAI. Free users can switch to an ads-free option, but OpenAI says it comes with "lower usage limits" and "reduced feature access," such as no image generation or deep research. OpenAI also says ads don't appear near "personal health, mental health, or politics" and that "ads do not influence ChatGPT's answers." None of this changes training: ads and training are governed by different settings.

Project Lily: what's reported, what's alleged, and what isn't proven

404 Media reported on September 14, 2026 that OpenAI is hiring hundreds of contractors to read real ChatGPT prompts, sometimes whole conversations, to rate and critique the replies. It reported that contractors don't see usernames, that OpenAI says it tries to remove personal information first, and that the company acknowledged sensitive details can still get through.

The proposed class action, as summarized by Decrypt, alleges that contractors hired through staffing firms summarize what a user wanted and score four candidate replies on a scale of one to seven, and that OpenAI's policies never clearly told users a person might read their chats. Decrypt also reports, citing 404 Media, that reviewers see a "user memories summary." The plaintiffs ask the court to make "Improve the model for everyone" off by default and to require opt-in consent before an outside reviewer sees a chat.

None of this has been tested in court. The docket shows OpenAI was served on September 21, with a response due October 13, and a stipulation filed by the parties on October 12. As of October 15, 2026, there's no ruling and no certified class. The four OpenAI pages we read for this article (the privacy policy, data controls, temporary chat, and the training page) don't contain the word "human" or "reviewer."

Court-ordered retention and the 20 million logs

In May 2025, a court in the New York Times copyright case ordered OpenAI to preserve ChatGPT output logs it would otherwise delete, including deleted chats. OpenAI's own update says that obligation "ended on September 26, 2025," but that it still stores "limited historical April–September 2025 user data" because the Times continues to demand it.

Separately, in January 2026 Judge Stein upheld the order to produce a sample of 20 million de-identified ChatGPT logs, finding users' "sincere" privacy interests protected by the smaller sample, de-identification, and a protective order, according to Jones Walker's analysis. In July 2026, TechCrunch reported that the Times accused OpenAI of withholding evidence about those logs; OpenAI called the allegations "blatantly false." For what this means for anyone's chats in a lawsuit, read whether AI chats can be subpoenaed.

"Deleted within 30 days" holds until a court says otherwise, and you won't be asked first.

What can you do about it?

• Turn off "Improve the model for everyone" now, before your next sensitive chat.
• Use a temporary chat for anything you wouldn't want trained on or kept in history, and skip the thumbs buttons.
• Keep identifying details out of what you type. De-identification removes names and usernames; it can't remove a story that only fits you.
• Delete what you don't need, knowing the exceptions above.
• Pick the tool by who you're hiding from. Our guide to using AI privately walks through the options.

How Morse keeps your chats private

On ChatGPT, your words sit on OpenAI's servers, tied to your account, where they can be trained on, read by reviewers, used to pick ads, and held for a court. Morse is built so your chats don't pile up anywhere but your own device, and you choose how private each chat is.

• Morse doesn't train AI on your content, sell your data, or show ads. There's no training switch to hunt for and no ad system reading your words.
• Your chat history stays on your device. Chats are stored in your browser, encrypted, and locked when you sign out. Morse's servers don't keep your messages or log what you write, so there's no server-side history to review or preserve, and Morse usually has very little to give if a court asks. Our privacy policy has the details.
• Deleting means deleted. Your device holds the only copy, so deleting a chat or clearing your browser removes it for good, with no 30-day copy waiting on a server. Morse can't recover it either, so keep anything you need.

Then pick the mode by what the chat holds:

• Anonymous, to use GPT, Claude, and Gemini without the provider knowing it's you. Morse's proxy removes your account, email, and IP address and sends the message under Morse's own account. OpenAI sees the words, with your identity stripped, so keep names and identifying details out, just as you would on ChatGPT.
• Private, the default, for chats no host should keep. Open-weight models like DeepSeek, Kimi, and GLM, plus Grok, run on hosts that agreed to zero data retention: they keep neither your prompts nor the replies. Your identity is stripped too, as in Anonymous.
• E2EE, for anything no one should read. Start the chat in E2EE and only a verified enclave can open it. Your browser checks the enclave's proof that it's genuine before anything leaves, encrypts the message for that enclave alone, and opens the reply on your device. Morse's servers relay text they can't read, and every reply has a lock that shows each check that passed. Here's how end-to-end encrypted AI chat works.

You can try a chat from the home page without an account, or start free, with 15 messages a day across GPT-6 Luna, Claude Haiku 5.5, GLM 5.3 Flash, and more. End-to-end encrypted chats come with Pro, at $19 a month.