How do I turn off ChatGPT training on my data?

#Setting or Privacy Portal: do you need both?

No. OpenAI's Privacy Portal offers a request called Do not train on my content. According to OpenAI, the setting and the portal request each work on their own for ChatGPT conversations and Codex tasks: "You don't need to do both."

The difference is in how they work. The setting is a switch you can turn back on whenever you like. The portal request is meant as a permanent opt-out: you select Make a Privacy Request, verify your account, choose Do not train on my content and get an email once it has been processed.

Either way, OpenAI describes the opt-out as applying to new conversations. It does not undo training that has already happened.

#Exceptions: feedback, Codex and Temporary Chat

• A thumbs up or down. If you rate an answer, OpenAI may use the entire conversation behind that rating to train its models, even after you opted out. If you want a conversation kept out of training, leave the thumbs alone.
• Codex. On a personal plan, the ChatGPT setting also covers Codex tasks. Codex also has a switch of its own: Include environments in the Codex settings decides whether extra context from your Codex environments may be used for training. The ChatGPT setting and the Privacy Portal leave that switch as it is.
• Temporary Chat. For a single conversation, open a new chat and select Temporary. While it stays temporary, the chat does not appear in your history, does not create or update memories and is not used for training. OpenAI may keep a copy for up to 30 days for safety purposes. If you save it, it becomes a regular chat and follows your training setting (OpenAI on temporary chat).
• Archived chats. Archiving a chat does not take it out of training. Your setting keeps applying to it.

#Which ChatGPT accounts are used for training?

On the personal plans, Free, Go, Plus and Pro, OpenAI may use your conversations for training until you turn it off. ChatGPT Business, Enterprise and Edu work the other way round: OpenAI does not use content from those workspaces for training unless the organisation has signed a separate agreement allowing it, and a thumbs up or down does not change that. The details are on OpenAI's enterprise privacy page.

So the account type matters more than the setting. What else changes between the two, from a data processing agreement to what an admin can see, is covered in free versus business AI accounts. And "no training" is not the same as "not stored": see what "we don't train on your data" actually means.

#Deleting a chat: what it does and does not do

Turning off training does not delete anything. If you want a conversation gone, delete it separately. ChatGPT then removes it from your account straight away, and OpenAI schedules it for permanent deletion within 30 days, unless the chat was already de-identified and disconnected from your account, or OpenAI has to keep it longer for security or legal reasons. What else stays behind after you delete, from logs to backups, is in what deleting an AI chat really does.

What deleting does not do is retrieve data that has already been used to train a model. Once "baked into" a trained model, information is no longer in a box you can empty. Turning off training and deleting chats mainly help from now on.

#The right to erasure under the GDPR

As a European user you have the right to erasure (Article 17 GDPR), also known as the right to be forgotten. You can submit a request through the provider's privacy portal, for example to delete your account and stored conversations. For ChatGPT that is the same Privacy Portal. It is meant for personal accounts: if your ChatGPT account is provided by your employer, the organisation handles privacy requests about that data, not OpenAI. Keep in mind that erasure primarily covers stored data and future processing, and that providers may retain certain data where there is a legal basis to do so.

#Gemini: turn off Keep Activity

In the Gemini app with a personal Google account, the setting is Keep Activity, under Gemini Apps activity. When it is off, Google does not use your future chats to train its AI models, unless you send feedback. Chats are still kept for 72 hours so that Gemini can respond and Google can protect its services; that protection can involve human reviewers. Chats a reviewer has already looked at are kept for up to three years and stay when you delete your activity. Temporary chats are not used for training either. This follows the Gemini Apps Privacy Hub, updated 19 October 2026. More in Gemini privacy risks at work.

#Claude: turn off model improvement

On Claude Free, Pro and Max, including Claude Code on those accounts, go to Settings, then Privacy, and turn off Help Improve our AI models (Anthropic's instructions). New chats and coding sessions are then left out of future training; data already used in training stays there. With the setting on, Anthropic keeps your data for up to five years, otherwise 30 days (Anthropic, 28 August 2025). Incognito chats are not used for training, even with the setting on. Team, Enterprise and the API fall under Anthropic's commercial terms and are not used for training by default. More in Claude privacy risks.

#Copilot: personal account or work account

With a work or school account (Microsoft Entra ID), Microsoft states that prompts, responses and data accessed through Microsoft Graph are not used to train the foundation models behind Copilot (Microsoft Learn). Microsoft has renamed Microsoft 365 Copilot to Microsoft Copilot, so check which account you are signed in with rather than the app name.

With a personal Microsoft account, the Copilot app has its own training choice under Privacy: Training on conversation activity and Training on voice conversations. Opting out covers your future conversations (Microsoft's privacy controls). That page applies only to the version of the app from before the update of 18 August 2026. For the updated app, Microsoft states that prompts, responses and file contents are not used to train foundation models, and its privacy settings no longer list a training switch (Microsoft on Copilot activity history). More in Copilot privacy risks.

#Why opting out does not solve the problem

Turning off training is sensible, but it does not address the core risk at work. Two reasons:

1. Opting out governs what the provider may do, not what you share. Even without training, a sensitive detail can be stored, read by a connected app or travel further than intended via a shared link. With ChatGPT at work, the risk usually lies in what gets pasted.
2. A personal account with training switched off is not an approved work environment. For customer, patient or case data, a managed business environment with a data processing agreement is the starting point, not a toggle in a free account.

#Checking training settings for one tool does not cover the next one

Turning off training in ChatGPT settles ChatGPT. It says nothing about the next AI tool someone on the team opens. For more than 1,000 tools, BeeSensible's catalogue records whether each one trains on input, alongside its risk rating, so the question does not have to be answered from scratch every time a new tool appears.

#The data you do not share needs no opt-out at all

Opting out limits what happens to your data once it has been sent. It is more effective to keep sensitive data out in the first place, and to know what you can and cannot share with AI. BeeSensible highlights sensitive details while you type in browser-based AI tools, so you can mask or remove them before you send. What you never send needs no opt-out, no deletion and no erasure request.

Further reading: our AI data leakage overview and the article on data sovereignty and AI prompts.