ChatGPT memory does not appear in the data export, so it cannot be exported as a file. Open Settings, then Personalization, then Manage Memories to read the stored entries, or ask ChatGPT to list everything it remembers about you. Paste that into Claude as project instructions, alongside converted transcripts for the context memory does not capture.
Short answer
ChatGPT memory does not appear in the data export, so it cannot be exported as a file. Open Settings, then Personalization, then Manage Memories to read the stored entries, or ask ChatGPT to list everything it remembers about you. Paste that into Claude as project instructions, alongside converted transcripts for the context memory does not capture.
Step by step
Read your stored memories: In ChatGPT, go to Settings, then Personalization, then Manage Memories. This panel lists the discrete facts ChatGPT has saved about you. Copy the list out; there is no export button and it is not in the ZIP archive.
Ask ChatGPT to summarise what it knows: Prompt it with a request to write out everything it remembers about you, your work, and your preferences, as a document addressed to another assistant. This surfaces inferred context in a usable form, not just the saved entries.
Export the conversations behind the memory: Memory is a thin summary of a much richer history. Request your full data export from Settings, then Data Controls, then Export Data, so you also have the conversations the memories were distilled from.
Convert the export to readable files: Turn conversations.json into per-conversation Markdown, then pick out the handful that carry real standing context: your projects, your decisions, your working style.
Rebuild it inside Claude: Create a Claude Project. Put the memory summary into the project instructions, and upload the selected transcripts as project knowledge. Claude now has both the summary and the evidence behind it.
This is the request behind most ChatGPT-to-Claude switching, and it has a frustrating first answer: memory is the one thing OpenAI does not give you a copy of.
Memory is not in your export
People request the data export expecting to find their memories in it. They are not there. The ZIP contains conversations.json, user.json, message_feedback.json, model_comparisons.json, and your uploads. The memory store is not among them.
So the extraction is manual, and there are two passes worth doing.
The explicit pass. Settings, then Personalization, then Manage Memories shows the discrete entries ChatGPT has saved. Copy them verbatim. This is the literal memory store.
The implicit pass. Ask ChatGPT directly: have it write out everything it knows about you, your work, your preferences, and how you like responses formatted, as a handover document for another assistant. This captures the inferred context that never became a formal memory entry, and it is usually the more valuable of the two.
Claude's equivalent is Projects
Claude does not have a memory feature that maps one-to-one onto ChatGPT's. What it has is Projects, and for this purpose Projects are better, because they are explicit and you can edit them.
| ChatGPT | Claude equivalent |
|---|---|
| Saved memories | Project instructions |
| Custom instructions | Project instructions, or profile preferences |
| Per-topic memory bleed | Separate Project per topic |
The last row is an upgrade rather than a workaround. ChatGPT memory is global, so context from one part of your life leaks into unrelated conversations. Claude Projects are scoped, so your work context and your personal context stay apart unless you choose otherwise.
Memory alone is a poor handover
A memory list is a set of conclusions with the reasoning stripped out. "Prefers concise answers" and "works in fintech" tell Claude very little compared to the twenty conversations where you actually worked through something.
This is why the export still matters. ChatExports converts conversations.json into per-conversation Markdown files locally in your browser, and lets you search and filter the history so you can pull out the conversations worth uploading rather than dumping everything.
The combination that works: the memory summary as project instructions, plus a small set of real transcripts as project knowledge. Claude then has both the shorthand and the substance.
Verify before you rely on it
After setting the project up, ask Claude something that depends on the transferred context and see whether the answer reflects it. Migrations fail quietly here: the files upload, the instructions save, and nothing actually references them because the project knowledge exceeded its limit or the instructions were too vague to act on.
Keep the memory current
ChatGPT memory drifts as you use it, and your Claude project instructions will not update themselves. If you run both platforms during a transition, re-read the memory panel every couple of months and refresh the project instructions. Our guide on exporting ChatGPT to Claude covers the conversation side of keeping the two in sync.
Frequently asked questions
Is ChatGPT memory included in the data export?
No. The export ZIP contains conversations, feedback, model comparison data, account metadata, and uploads, but not the memory store. You have to read memories out of Settings, then Personalization, then Manage Memories manually.
Can I ask ChatGPT to write out everything it remembers?
Yes, and it is the more useful of the two methods. Ask it to produce a handover document for another assistant covering you, your work, and your response preferences. That captures inferred context the memory panel never recorded.
Does Claude have a memory feature like ChatGPT?
Not an identical one. Claude uses Projects, where instructions and uploaded knowledge apply to every conversation in that project. It is scoped rather than global, which avoids context from one area bleeding into unrelated chats.
How much history should I upload alongside the memory summary?
A small, deliberate set. Five to twenty conversations that contain standing decisions or reference material will outperform a full history dump, which mostly consumes the project knowledge limit with noise.
Will this work in reverse, from Claude to ChatGPT?
Yes, the same shape. Export from Claude under Settings, then Account, convert the JSON to readable files, and paste your project instructions into ChatGPT's custom instructions and memory.