ChatGPT remembers things about you — your projects, your preferences, the way you like answers shaped. At some point you’ll want that memory out: to back it up, to read what it has actually stored, or to move to another assistant. Here’s how to export your ChatGPT memory, and the more important point underneath it — why a one-time copy is not the same as owning your memory.

How to export your ChatGPT memory

There are two layers worth pulling out, and they live in different places.

1. Your saved memories

These are the discrete facts ChatGPT has chosen to remember about you.

Open ChatGPT and go to Settings → Personalisation → Memory.
Select Manage memories to see the list of everything it has saved.
Read through it. You can edit or delete individual entries here, but there is no native “download all memories” button — so to keep a copy, select the text and save it somewhere you control.

2. Your full data export

For everything else — your conversations, custom instructions and account data — use the formal export:

Go to Settings → Data controls → Export data.
Confirm the request. OpenAI emails you a download link, usually within minutes to a few hours.
The archive arrives as a .zip containing HTML and JSON files. Your conversations are in conversations.json; your custom instructions and saved context are in the account files alongside.

Keep that archive somewhere safe. It is the closest thing to a snapshot of what ChatGPT knows about you.

Why the one-time copy isn’t enough

Here’s the catch. What you’ve just produced is a photograph of your memory at one moment — not the living memory itself.

In March 2026, the major vendors made this easier: you can now copy your memory out of one assistant and import it into the next. That’s a genuine improvement on being locked in. But it has three structural limits that no export button fixes:

It’s a copy, not the original. The authoritative memory still lives on the vendor’s servers, in the vendor’s format. Your export goes stale the moment you keep using ChatGPT.
It only moves once. Importing your memory into a new app is a one-way, lossy migration. Do it again next month and you’re reconciling two divergent copies by hand.
It leaves the persona behind. An export carries what the assistant knows about you. It does not carry who the assistant is when it’s with you — the voice, the conventions you agreed, the things it refuses. That relationship is often the most valuable thing you’ve built, and a data dump throws it away.

So the question isn’t really “can I move my memory?” After March 2026, everyone can. The question is “whose memory is it?”

What owning your AI memory looks like instead

A one-time export answers “can I get a copy out?”. An AI memory layer answers a better question: “can my memory just be mine, everywhere, all the time?”

Instead of a snapshot you periodically dump and re-import, you hold one persona in a signed, local-first file you own. It isn’t a copy of the vendor’s memory — it is your memory, and every AI you use reads from it:

One file you hold. Your persona lives in a git repo you control; your corpus is a local SQLite database on your machine. Nothing is hosted by default.
Present everywhere at once. The same memory in ChatGPT, Claude, Gemini, local models and 30-plus editor and chat surfaces — not four exports drifting apart.
Both dimensions travel. Your context and the assistant’s character move together, because they live in the same file.
An exit that’s always open. Open format, exportable any time, self-hostable. Leaving is a feature, not a fight.

The test that proves it

There’s a simple, un-fakeable test for whether a memory is genuinely yours. Cultivate an assistant under Claude for a week, then point a local Llama model at the same persona file. If it greets you the same way, holds the same refusals and remembers what you were working on — that continuity is the thing you own. A vendor export can’t give you that, because it only ever spoke one app’s language.

So go ahead and export your ChatGPT memory — it’s worth having a copy. Just don’t mistake the photograph for the thing itself. If you want a memory that’s actually yours, you want an AI memory layer — one persona, owned by you, present across every model you use. That’s what aiperson is built to do. When you’re ready, you can own your AI memory today.

How to export your ChatGPT memory (and why a one-time copy isn't enough)