What ChatGPT remembers, and what it takes to remove something

Memory is the feature most likely to move information between contexts you thought were separate. For anyone working across more than one client, project or employer, that is worth ten minutes of understanding rather than a setting you meet by accident.

Where it comes from

Memory lives under Settings → Personalization → Memory and can be switched off at any time.

It is assembled from more than your typing. OpenAI describes it as remembering context from your chats, your files, and your connected apps. So a document you uploaded once and a connector you authorised months ago are both feeding it.

The summary is only a view

You can read what ChatGPT remembers in the memory summary, and you can edit it by typing into it or highlighting a line to correct.

The documentation is unusually direct about the limit of that view: the summary "will not include everything that ChatGPT remembers based on your chats". If you want to know whether something specific is in there, the advice is to ask in the chat rather than to scan the page.

There is a related tool worth knowing. The book icon below a response shows which sources personalised it, including custom instructions, past chats, files and memories, and tapping a memory explains why it was used. OpenAI notes that this too "may not show every factor or source that shaped a response". Memory sources are not included when you share a chat.

Removing something takes a sweep

This is the part that matters and the part nobody expects.

The second half of the trap follows from the first. Choosing Delete and turn off memory does not delete your past chats. If you turn memory back on later, ChatGPT may build new memories from the chats that are still there, including old ones. The feature reconstructs itself from the material you left behind.

The clean surface, if you want one, is Temporary Chat. It does not use or create memories at all.

Business workspaces, and one irreversible step

Try this

Ask in a chat: what do you remember about me? Then open the memory summary and compare the two. The gap between them is the point of this page.

What goes wrong

Reading the memory summary as the complete record. It is a partial view and the documentation says so.

Deleting a memory entry and considering the matter closed. The source chat, the archived copy and the uploaded file are all still there.

Turning memory off to clear it. That leaves your chats in place, and turning it back on can rebuild memories from them.

Merging a personal workspace into an employer's during signup. It cannot be undone and it hands over every prior conversation.

Assuming project boundaries contain context by default. They do not, unless project-only memory is switched on.

How to check it worked

After a sweep, open a new chat and ask about the thing you removed. If it comes back, you missed a source, and the likely candidates are an archived chat or a connected app rather than the summary you already cleaned. Then check the book icon on a personalised response and see which sources it names.