How to Move ChatGPT History to Claude, Memory and Conversations
By Matthew Firth, founder of ModelBrain · Nov 1, 2026 · 4 min readOur own proposal: move your ChatGPT memory with Claude's import and leave your conversations in a store Claude can query, because the import carries a summary and not the history. We indexed a real ChatGPT export of 1,462 conversations into 152,760 searchable chunks in 118 seconds on a Mac, with no failures, and nothing had to be uploaded by hand.
Two things move, and they move differently
Claude has a memory import for what ChatGPT knows about you. In Claude, open Settings, Capabilities, Memory and choose Start import. Claude gives you a prompt, you run it in ChatGPT, and you paste the summary it writes back into Claude. Anthropic describes this on its help page. Menu labels shift between app versions, so follow the labels you see. What arrives is a summary of you, not your conversations.
Your conversations are the other half. In ChatGPT, open Settings, Data Controls, Export Data and confirm. OpenAI emails a download link, and the link expires after 24 hours, so download the zip when the email lands.
| What you want | Where it comes from | How it gets to Claude |
|---|---|---|
| Preferences and facts about you | ChatGPT memory | Claude memory import, one pasted summary |
| Past conversations | The export zip | Upload files to a Claude Project, or index them in a store Claude can query |
| Custom instructions | ChatGPT settings | Copy into Claude preferences by hand |
What is in the zip you are moving
Our real export was 916 MB and held 1,463 conversations in fifteen numbered files, conversations-000.json to conversations-014.json, plus chat.html and about 834 image files. Each conversation is a tree of edits and regenerated answers, which our export walkthrough explains. Loading a file of that size into a Project by hand means choosing shards and deciding what to leave out.
Our own proposal: index the history, then let Claude ask for it
Moving 1,463 conversations into a chat window copies them once and then leaves them as files. Putting them in a memory store that Claude reaches over MCP leaves them searchable by any assistant you use. We ran that on the same export with ModelBrain 0.6.6 on a Mac on October 27, 2026. It indexed 1,462 conversations into 152,760 chunks in 118 seconds with no failures, and skipped the one conversation flagged do-not-remember, as it should.
| Step | What happens |
|---|---|
| Export from ChatGPT | Zip arrives by email, link valid 24 hours |
| Index the zip | ModelBrain reads the shards directly, keeps the date on every chunk, skips do-not-remember conversations |
| Connect Claude | Claude calls ModelBrain over MCP, so nothing is re-uploaded |
| Ask | Claude recalls from the 152,760 chunks when a question needs them |
The store stays on your machine. Current download is 0.6.6 for macOS, Windows and Linux. The same index serves any assistant that speaks MCP, which is the part a one-time import cannot do. For the file layouts see the three export formats compared.