Most people treat the ChatGPT box like a private notebook. It isn't one, and it isn't a public billboard either. The truth sits in between, and once you know where, you can decide what goes in.
There are four groups who can end up seeing what you type. Two of them you can switch off in about three minutes. The other two you can't, and that is exactly why the checklist at the bottom of this guide exists.
The four, and what each one can actually see
1. ChatGPT itself (training)
On a personal account (Free, Go, Plus or Pro), a setting called Improve the model for everyone is on by default. While it is on, your conversations can be used to train future versions of the model.
Training doesn't mean a future version will repeat your chat word for word to a stranger. It means your words become part of what the model learns from. That is a decision you are allowed to make, and most people never knew they were making it.
Business, Enterprise and Edu accounts are different: they are not used for training by default.
2. Real people
OpenAI says a limited number of authorized staff and trusted outside service providers may access conversations. That happens to investigate abuse or security problems, to give support, for legal reasons, and (unless you opt out) to help improve the models.
That is not the same as "someone reads everything you type." It does mean a human could see a conversation, and some of those reasons apply even after you switch training off.
3. Anyone with the link
When you press Share, ChatGPT creates a public web page of that conversation. Anyone who has the link can open it, and it stays live until you delete it.
In summer 2025, some shared chats showed up in Google search results. Those were chats where people had ticked an option to make them discoverable. OpenAI removed that option, but the lesson stands: a shared chat is a web page, and you probably have old ones you've forgotten about.
4. A courtroom
In the copyright case between The New York Times and OpenAI, a court ordered OpenAI to hand over a sample of 20 million ChatGPT logs. A federal judge upheld that order in January 2026. The logs were de-identified, meaning names and identifying details were removed, and they are held under a protective order.
So this isn't "the newspaper is reading your chats." It is a reminder that chats are records, like emails, and records can be requested in legal cases. My guide Your AI Chats Can Be Used Against You in Court goes deeper on what that means for your own situation.
The two settings from the video
Setting 1: switch off training
- Click your name or profile picture, then Settings.
- Open Data controls.
- Find Improve the model for everyone and switch it off.
This stops new conversations from being used for training. It syncs to your account, so you only need to do it once, not on every device. On the phone app, it is the same path: your profile, then Settings, then Data controls.
Setting 2: delete your shared links
- Stay in Settings → Data controls.
- Next to Shared links, click Manage.
- Go down the list. For anything you don't want public anymore, click the delete icon.
Deleting a shared link takes the public page down. It doesn't delete your own copy of the conversation. If you have a lot of old links, my guide The Share Links You Forgot You Made covers Claude and other tools too.
Five more settings worth checking
3. Use Temporary Chat for anything sensitive
At the top of a new chat, choose Temporary Chat. Those conversations don't appear in your history, aren't used for training, and don't create memories. OpenAI can still keep a copy for up to 30 days for safety reasons, and then it is deleted.
Use it for one-off questions you'd rather not have sitting in your sidebar: a health worry, a money question, a family situation.
4. Look at what Memory has saved
- Go to Settings → Personalization.
- Open Manage memories and read the list. It is often more detailed than people expect.
- Delete anything you don't want kept, or switch off Reference saved memories and Reference chat history if you'd rather ChatGPT start fresh each time.
Deleting a chat does not delete a memory that came from it. They are separate, so check both. My guide The Memory Audit Prompts helps you read the list properly.
5. Clear out old chats
In Settings → Data controls, you can delete all chats, or delete them one at a time from the sidebar. Deleted chats are removed from your account right away and scheduled for permanent deletion from OpenAI's systems within 30 days, unless they must be kept for legal or security reasons.
Archiving is not deleting. An archived chat is just hidden from your sidebar.
6. Check the voice and video switches
If you use voice mode, look underneath Improve the model for everyone. You may see separate switches for audio and video recordings. Turn them off if you don't want your recordings used for training either.
7. See which apps are connected
Go to Settings → Apps (this used to be called Connectors). Every app listed there, such as Gmail, Google Drive or your calendar, is something ChatGPT can read. Disconnect anything you aren't using. You can connect it again later in one click.
What never to type into ChatGPT
The short version. The full checklist, and how to take private details out of a document before you paste it, is in 5 Things You Should Never Put Into ChatGPT or Claude.
- Passwords and login codes, including the six-digit codes your bank texts you.
- ID numbers: Social Security or Social Insurance numbers, passport and driver's license numbers, or photos of the cards.
- Full bank and card numbers, or screenshots that show them.
- Medical or legal documents with your name on them. Ask about the situation, and leave the file out.
- Confidential work: contracts, numbers that aren't public, anything you signed an agreement to keep private.
- Other people's private information: your kids' details, a friend's diagnosis, a family member's divorce.
- The details of an active dispute: names, dates and what happened. Take those to a real lawyer.
A prompt to check yourself before you paste
Before you answer anything, check the text below for private information: names, addresses, phone numbers, ID or account numbers, passwords, health details, or anything that identifies another person. List what you found and suggest a version with those details replaced by placeholders like [NAME] or [ACCOUNT]. Don't answer my actual question until I confirm the cleaned version.
One honest note on this prompt: by the time ChatGPT reads it, the text is already in the chat. It is a habit-builder, not a shield. The real version of this habit happens before you paste.
The three-minute version
- Settings → Data controls: switch off Improve the model for everyone, and the audio and video switches if you see them.
- Same screen: Shared links → Manage, and delete what you don't need.
- Settings → Personalization → Manage memories: read it, and delete what you don't want kept.
- Settings → Apps: disconnect anything you're not using.
- Use Temporary Chat for sensitive one-offs.
- Keep the "never type this" list in mind, because no setting replaces it.
Next step
Curious what ChatGPT has already pieced together about you from your chats? Try What ChatGPT Has Figured Out About You. And if you want to remove personal information it already holds, Delete Your Personal Information From ChatGPT walks through all five switches in order.
The one line to remember
Settings decide who else learns from your chats. Only you decide what goes in them.