How ChatGPT Memory Works — and How to Turn It Off

What "memory" means on each platform

The three systems aren't identical, but they follow the same basic pattern: an explicit set of facts you can view and manage, plus a broader layer synthesized automatically from your conversation history.

The practical upshot is similar across all three: mention your job, your writing preferences, or a project you're working on, and that context may carry into your next conversation whether you explicitly ask it to or not.

Memory vs. referencing a specific conversation

These get confused constantly, and they solve different problems.

Memory is a standing profile. It applies broadly, across every new conversation, built from a synthesis that you don't fully control — you can view and delete individual entries, but you don't choose which past chat a given detail came from.

Referencing a conversation is deliberate and scoped. You point the model at one specific past thread — a project, a piece of research, or a document you built together — and ask it to draw only on that context in a new chat. It doesn't touch your memory settings at all.

The difference matters most when memory gets something wrong. If ChatGPT has stored an outdated fact — a job you left, a city you moved from — that detail can quietly shape unrelated conversations for months until you go delete it. Referencing a specific chat doesn't have that failure mode: you're pointing at one exact conversation, not hoping the right details survived a background synthesis process.

Worth knowing: none of the three platforms share memory with each other. What ChatGPT knows about you stays in ChatGPT. If you want the same context available everywhere, you'll need to re-establish it on each platform separately, or maintain it outside all three.

How to check what's stored, and turn it off

1. ChatGPT — Settings → Personalization → Manage Memories. Individual entries can be deleted one at a time, or you can toggle off "Reference saved memories" and "Reference chat history" separately to disable each layer without deleting what's already stored.

2. Claude — Settings → Capabilities → Memory, then "View and edit memory" to see what's stored. Turning memory off gives two options: Pause, which keeps existing memory but stops building new entries, and Reset, which permanently deletes everything — that one can't be undone.

3. Gemini — Settings → Personal context. Delete individual saved facts from the list, or say "Forget that I mentioned [X]" directly in a chat. Turning off the Memory toggle here stops it from learning new things without erasing what's already saved.

On all three, disabling memory stops the system from retaining new information — it doesn't retroactively erase what's already there. If the goal is a clean slate, you need the separate delete or reset action, not just the toggle.