What ChatGPT actually knows about you (and how to delete it)
The short version: ChatGPT keeps what it has learned about you in a memory summary at Settings › Personalization › Memory. You can read and edit it yourself. Deleting an entry there does not remove what it knows, because the same information also lives in your chats, archived chats, uploaded files, and connected apps.
There is a trick going around. You ask ChatGPT to tell you what it secretly knows about you, and it comes back with something specific enough to be unsettling. The reels always end the same way: a screenshot, a stunned caption, a prompt pack in the comments.
The trick works. But the interesting word in that sentence is not knows. It is secretly, and it is doing a lot of work it has not earned. What ChatGPT knows about you is written down on a page you can open in about four seconds, and the part worth your attention is not what it says. It is what happens when you try to delete it.
It is the same shape as the viral prompt packs we took apart in there are no secret ChatGPT codes: a real effect, sold with a word that makes it sound like a hidden feature rather than a documented one.
Open the page it is reading
Go to Settings › Personalization › Memory. That is the whole secret. You are looking at a summary ChatGPT writes about you, and it updates itself with new context as you chat. The top of the page tells you when it was last changed.
You can edit it in place: type a correction into the box at the bottom, or highlight a specific line to fix just that. It is a document, and you have the pen.
Then ask it, because the page is not the whole story
This is the first thing the viral version leaves out. OpenAI states plainly that the memory summary will not include everything ChatGPT remembers from your chats, and that if you want to know whether it remembers something, you should ask it in the chat.
Does your memory contain anything about [topic]? Answer only from what you actually remember about me, and say so if you have nothing.
There is also a quieter control most people never notice: tap the book icon under any response and ChatGPT will show you which sources shaped it, including custom instructions, past chats, files, and memories.
Deleting a memory does not delete what ChatGPT knows. In OpenAI's own words, fully removing something means deleting every source where it appears: past chats, archived chats, files, the memory summary, and any connected apps that hold the same information. Clear one and the others can put it back.
Clear all five sources, in this order
Order matters, because the earlier ones tell you what to look for in the later ones. Read the summary first, then go hunting.
- The memory summary. Settings, Personalization, Memory. Edit individual lines, or use the three-dot menu and choose Delete and turn off memory to clear the lot.
- The chat it came from. Saved memories are stored separately from your chat history, so deleting the conversation does not remove the memory, and deleting the memory does not remove the conversation. Both, every time.
- Archived chats. Archiving is not deleting. Settings, General, Archived chats. Anything in there is still a live source.
- Files you uploaded. A resume, a bank statement, a medical letter. The attachment is its own source, separate from the message you sent it with.
- Connected apps. Email, calendar, and drive connections can carry the same information straight back in. Disconnect anything you are not actively using.
Know what the off switch does not do
Using Delete and turn off memory does not delete your past chats. And if you switch memory back on later, ChatGPT may build new memories out of the conversations still sitting in your history, including old ones. A reset that leaves the chats in place is not a reset. It is a pause.
One more piece of small print worth knowing: OpenAI says it may keep a log of deleted saved memories for up to 30 days for safety and debugging.
Use Temporary Chat for anything you would not want summarised
This is the setting that should be in every one of these viral reels and never is. A Temporary Chat does not use your existing memories, does not create new ones, does not appear in your history, and is deleted from OpenAI's systems after 30 days.
It is not a privacy guarantee, and OpenAI notes such chats may still be reviewed for abuse monitoring. But if you are about to paste something you would not want folded into a running profile of you, that is the mode to be in.
Turning off memory does not stop your chats being used to improve the model, and turning off training does not stop memory. Training lives at Settings › Data Controls › Improve the model for everyone, it applies to your whole account on every device, and your chats stay in your history either way. Decide on both, not one. Note also that turning off memory and personalization does not disable safety features, which may still use limited safety-relevant context in rare, high-risk situations.
None of this is a ChatGPT quirk, either. Claude keeps the same kind of running notes, and deleting the conversation leaves them behind in exactly the same way — the difference is that Claude puts them in an editable list you can open. We went through that one in what Claude remembers about you, and how to wipe it, including the five things to check the first time you open it.
If you want the same treatment applied to the other ChatGPT feature the reels keep overselling, we did that with agent mode, explained honestly: what it genuinely does, and the safety rules the viral version leaves out.
And if the account you are clearing belongs to someone under 18, the settings page looks different now. OpenAI moves those accounts into a separate experience on its own, and one of the ten settings a parent can reach is the memory switch covered above. We went through all of it in what ChatGPT for Teens actually changes, including the four signals that decide who counts as a teen.
Questions people actually ask
Does the viral prompt actually work?
Yes, and that part is not a trick. If memory is on, ChatGPT has been keeping a running summary of what it considers important about you, so asking it to describe you produces something specific. What the viral version gets wrong is the word secretly. It is a documented feature with a settings page you can open right now.
Can ChatGPT really tell what I am in denial about?
No. It is reading a context file and writing plausible prose about it, the same way it does for any other prompt. It can be uncannily accurate about things you told it and confidently wrong about things you did not. Treat the output as a mirror of your own chat history, not a diagnosis.
If I delete a memory, is it gone?
Not necessarily. OpenAI's help centre says that to fully delete something ChatGPT may know about you, you have to delete every source it appears in: past chats, archived chats, files, the memory summary, and any connected apps. Saved memories are also stored separately from your chat history, so deleting the conversation does not remove the memory made from it.
Does turning memory off delete anything?
No. Turning it off stops new memories from forming. OpenAI notes that this does not delete your past chats, and that if you turn memory back on later it may create new memories from chats still in your history, including older ones. Clear the chats first if you want the reset to hold.
What if I just do not want any of this?
Use Temporary Chat. It does not use your existing memories, does not create new ones, does not save to your history, and OpenAI deletes it from their systems after 30 days. It is the closest thing to an off switch that still lets you use the product.
Is memory the same as ChatGPT training on my chats?
No, they are two separate settings. Memory personalises your answers. Training is controlled under Settings, Data Controls, Improve the model for everyone. You can turn either one off without touching the other.
Want this built for you?
We write these memos because we build this stuff every day. If you want it working in your business instead of sitting on your reading list, that is literally our job.