There is no direct transfer between OpenAI and Anthropic. Export your ChatGPT data from Settings, then Data Controls, then Export Data, convert the conversations.json into Markdown or plain text files, then attach the conversations that matter to a Claude Project. Claude reads them as project knowledge rather than as chat history.
Short answer
There is no direct transfer between OpenAI and Anthropic. Export your ChatGPT data from Settings, then Data Controls, then Export Data, convert the conversations.json into Markdown or plain text files, then attach the conversations that matter to a Claude Project. Claude reads them as project knowledge rather than as chat history.
Step by step
Request your ChatGPT export: In ChatGPT, open Settings, then Data Controls, then Export Data, and confirm from the email OpenAI sends. You will receive a ZIP within minutes to a few hours, and the download link expires after roughly 24 hours.
Convert conversations.json into readable files: The JSON inside is a tree of message nodes, not transcripts, so Claude cannot make much of it as-is. Convert it into one Markdown or text file per conversation, which is the format Claude ingests most cleanly.
Select what is actually worth moving: Do not move everything. Most chat history is disposable. Filter down to the conversations that carry standing context: your preferences, ongoing projects, decisions and their reasoning, and reference material you keep returning to.
Create a Claude Project: In Claude, create a Project and upload the selected files as project knowledge. Every conversation inside that Project can then draw on them, which is the closest equivalent to Claude having known you all along.
Write a context primer: Add one short document summarising who you are, what you work on, and how you want Claude to respond. This replaces the accumulated implicit context that made your ChatGPT account feel personalised.
People ask for this expecting an import button. There is not one, and there is unlikely to be one, because OpenAI and Anthropic have no incentive to make leaving easy. What does exist is a reliable manual path, and done properly it produces a Claude account that knows more about you than your ChatGPT account did, because you get to curate rather than accumulate.
| Custom instructions | Yes | Copy the text into Claude's project or profile instructions |
| Your standing preferences | Yes | Rewrite as a primer document |
| Saved memories | Partly | Must be read out of ChatGPT and rewritten manually |
| Uploaded files | Yes | They are in the export ZIP; re-upload the ones you need |
| Custom GPTs | No | Configurations are not in the export; rebuild as Claude Projects |
| Conversation branches | Not usefully | Abandoned edit branches are in the JSON but add noise |
The honest summary: your content moves, your configuration has to be rebuilt, and the rebuilding is a couple of hours at most.
Why the raw export fails as an import
conversations.json stores conversations as a linked tree of message nodes, because ChatGPT supports editing and regenerating any turn. Every version you abandoned is still in there. Handing that file to Claude wastes most of the context window on structural JSON and dead branches before Claude reaches anything you said.
Converted per-conversation Markdown is dramatically better: it is linear, it is labelled by speaker, it is small enough to attach selectively, and Claude parses it without any preamble from you.
Doing the conversion
ChatExports reads conversations.json directly, rebuilds the message tree into normal linear conversations, and exports each one as a separate Markdown, text, Word, or PDF file. You can filter by date or search across the whole history first, which is how you find the twenty conversations worth moving out of two thousand that are not.
Everything is parsed locally in your browser. Your entire ChatGPT history is the single most revealing file your account will ever produce, and it should not pass through a third party's server on its way to another AI vendor.
The curation step is the whole point
The instinct is to move everything. Resist it. Claude's project knowledge has a size limit, and more importantly, feeding it two years of half-finished conversations makes its answers worse, not better.
A good migration is usually:
One primer document: who you are, your role, your domain, your response preferences
Five to twenty reference conversations that contain decisions or material you still rely on
Any project-specific history, kept in a separate Claude Project per project
That is a sharper starting point than the ChatGPT account you left, which had all of the above buried under everything else.
Running both for a while
Most people do not switch cleanly, they run both for a few months and drift. That is fine, but it means your history is now split across two vendors, and a year from now you will want it in one place. Exporting from both sides into the same folder of Markdown files, on the same schedule, is the version of this that does not turn into a mess. Our cross-platform migration guide covers doing this across more than two AIs.
Frequently asked questions
Can I import ChatGPT data into Claude directly?
No. There is no direct import, and neither vendor offers a migration tool. The working path is to export from ChatGPT, convert the JSON into readable per-conversation files, and upload the ones that matter into a Claude Project as project knowledge.
Will Claude remember my ChatGPT conversations after I upload them?
Within a Project, yes, in the sense that Claude can read them for every conversation in that Project. It is not the same as native memory, but for practical purposes it gives Claude the context your ChatGPT account had accumulated.
How much of my ChatGPT history should I actually move?
Far less than you think. A primer document plus five to twenty genuinely useful conversations beats a full dump, because project knowledge has size limits and irrelevant history degrades the quality of Claude's answers.
Can I upload conversations.json to Claude as-is?
You can, but it works badly. The file is a tree of message nodes including every branch you abandoned, so most of the context window goes on structure and dead ends rather than on what you actually said.
Do my custom GPTs transfer to Claude?
No. Custom GPT configurations are not included in the OpenAI export at all, only conversations you had with them. Rebuild each one as a Claude Project with the instructions and reference files it needs.