The import, step by step
Anthropic built a direct route for this. There is no file export or API involved: you ask your old assistant to write out what it remembers, and paste the result into Claude.
- Open claude.com/import-memory (or the help article on importing memory) and copy the prompt Anthropic provides.
- Paste that prompt into ChatGPT. The prompt asks for stored memories and learned context in one code block, with dates where available; check the output for omissions and errors.
- Read the output before you move it. Remove anything sensitive you do not want in Claude, such as health details, passwords or client names.
- In Claude, go to Settings > Your Claude and click Start import. In older versions of the app this was an “Import memory to Claude” card on the home screen.
- Paste the text and click Add to memory.
- Claude turns the text into separate memory items. Check them in the memory panel (Settings > Your Claude > Manage) and edit or delete what is wrong.
The import works on Free, Pro, Max, Ultra, Team and Enterprise, on the web and in Claude Desktop. Memory has to be enabled on your account for it to be used; on Team and Enterprise that depends on what the owner allows.
What comes across and what does not
| Comes across | Does not come across |
|---|---|
| Saved memories that ChatGPT writes out in response to the prompt | Your chat history (that is a separate process) |
| Work-related preferences and facts | Personal details Claude may leave out, since the import focuses on work |
| Anything you removed before pasting |
Anthropic labels the feature experimental and still in development. This flow imports text into memory; it does not recreate your project files, instructions or custom assistants, so plan to rebuild those as Claude Projects. See how Claude Projects work.
The misunderstanding: memory is not your knowledge
The import moves what ChatGPT chose to remember about you: a list of preferences and facts. It does not move the reasoning in your conversations, the documents you uploaded, or the way you built prompts over months. People often switch expecting Claude to “know everything ChatGPT knew” and find it knows a summary. Treat the import as a starting point, then give Claude the real material through Projects. How Claude then uses that memory is explained in does Claude have memory.
What this means for a team
The import is per person. If a team moves from ChatGPT Business to Claude Team, each member imports their own memories, and none of it becomes shared team knowledge. Some members will skip it. And if part of the team keeps using ChatGPT, the two memories start to drift apart from day one: what someone teaches Claude on Monday, ChatGPT never hears about. Switching tools by copy and paste is something you redo every time the tool changes.
Where Bombos fits
Bombos removes the need to move memory between tools. It is a plugin that gives Claude, ChatGPT, Copilot, Cursor and your own agents one shared memory, with team knowledge, rules learned from corrections, and what already sits in mail, documents and repos. Switch tools and the memory stays. One person can try it first for their own agents, in our cloud or on their own server.
Related questions
Does importing move my ChatGPT conversations too?
No. The memory import only covers saved memories. Moving chat history is a separate process with its own help page.
Which Claude plans support memory import?
Free, Pro, Max, Ultra, Team and Enterprise, on the web and in Claude Desktop.
How long does the import take?
The current help article says updates appear shortly after import; its section on the legacy memory experience allows up to 24 hours.
Can I get my memory out of Claude again?
Yes. Go to Settings > Your Claude, click Manage next to Your memory and copy it to a file, or ask Claude to write out its memories word for word.
