There is no direct import. Export your ChatGPT data, convert conversations.json into Markdown files, then bring the relevant ones into Claude as project knowledge or pasted context. Custom instructions and memories must be rewritten by hand.
Short answer
There is no direct import. Export your ChatGPT data, convert conversations.json into Markdown files, then bring the relevant ones into Claude as project knowledge or pasted context. Custom instructions and memories must be rewritten by hand.
Step by step
Export from ChatGPT: Settings, Data Controls, Export Data. Confirm the email OpenAI sends, then download the ZIP within 24 hours.
Convert conversations.json: Turn the archive into one Markdown file per conversation so the content is readable and selectable rather than a nested JSON tree.
Pick what is worth moving: Most history is disposable. Identify the handful of conversations that carry ongoing context: specifications, style guides, long running projects.
Create a Claude Project: Upload the chosen Markdown files as project knowledge. Claude reads them as context for every conversation in that project.
Rewrite your instructions and memories: Copy your ChatGPT custom instructions into Claude's project instructions, and turn anything ChatGPT remembered about you into an explicit written brief.
Nobody ships an importer between AI assistants, and nobody is going to. Migration is therefore a manual job, but it is a much smaller job than it first appears once you accept that transferring everything is the wrong goal.
| Custom instructions | Yes, manually | Copy the text into Claude project instructions |
| Saved memories | Manually | Read them out of ChatGPT and rewrite as a brief |
| Uploaded files | Yes | They are in your ChatGPT export |
| Custom GPTs | No | Rebuild as a Claude Project with instructions |
| Conversation continuity | No | Claude has no knowledge of the prior chat |
Do not migrate everything
The instinct is to bring three years of history across. Do not. Claude's project knowledge works best with a focused set of documents; dumping hundreds of unrelated transcripts makes retrieval worse, not better.
A realistic split: five to twenty conversations that carry live context go into project knowledge. Everything else goes into a local archive you can search when you need it, and never gets uploaded anywhere.
Rebuilding memory as a brief
ChatGPT's memory is a list of facts it inferred about you. Open Settings, Personalization, Manage memories and read it. Most of it will be noise, but the useful parts, your role, your stack, your writing preferences, your recurring projects, become a short document.
That document is better than the original memory feature, because it is explicit, reviewable, and portable to whatever you use next. Paste it into Claude's project instructions.
Custom GPTs become Projects
A custom GPT is instructions plus knowledge files plus a name. A Claude Project is the same three things. Copy the instruction text across, upload the same files, and you have a working equivalent in about five minutes. Actions and API tool calls do not transfer and need rebuilding differently.
The archive step
Before any of this, export from ChatGPT and convert the result. conversations.json stores messages as a branching tree with parent and child pointers, so it is not something you can read, search, or upload usefully in raw form. Converted to Markdown you get one clean file per conversation, which is both what Claude wants as project knowledge and what you want as a permanent local archive.
ChatExports does that conversion in the browser with no upload, and reads Claude's export format too, so when you later add a third assistant to the mix everything lands in the same folder structure. Switching platforms then costs an afternoon rather than the fear of losing years of work.
Frequently asked questions
Can I import my ChatGPT history directly into Claude?
No. There is no import feature. You upload converted conversation files as project knowledge instead.
Will Claude remember conversations I had in ChatGPT?
Only what you explicitly provide as project knowledge or paste into a conversation. There is no shared memory between platforms.
How many files can a Claude Project hold?
Projects have a context budget rather than a fixed file count. A focused set of documents performs far better than a bulk upload.
Should I delete my ChatGPT account after switching?
Export and convert your history first. Once you have readable local copies, deleting costs you nothing.