Claude can’t import your ChatGPT chat history. Its memory import brings over what ChatGPT knows about you, not the conversations themselves. To take every conversation with you, export your data from ChatGPT, then either upload the chats you need to a Claude Project, or split the whole history into a folder of files that Claude reads and searches on your computer.
Which way fits depends on how much history you have. A few important chats fit in a Project. Years of chats usually don’t, and they work better as a folder.
Memory and conversations are two different moves
Claude’s memory import moves your preferences and context, not your chat logs. Anthropic’s memory import guide describes it this way: copy the prompt from claude.com/import-memory, run it in ChatGPT, then paste ChatGPT’s answer into Settings → Memory → Start import and click Add to memory. It works on the Free, Pro, Max and Team plans, on the web and in Claude Desktop.
That covers things like your role, your projects and how you like answers written. It doesn’t cover the conversations themselves: the drafts you worked through, the decisions you made and the reasons behind them. For those you need the export.
If you only want Claude to know you, the memory import is enough and you can stop here.
Step 1: Export your data from ChatGPT
OpenAI’s export guide gives these steps:
- Sign in to ChatGPT and click your profile icon in the bottom left.
- Open Settings → Data Controls.
- Under Export Data, click Export, then Confirm export.
- Wait for an email from OpenAI with a Download data export link, and download the ZIP.
The link expires 24 hours after the email is sent, so download it soon. OpenAI
says the export can take up to 7 days to arrive; check your spam and
promotions folders for mail from noreply@tm.openai.com.
Inside the ZIP, the two files that matter are:
chat.html, every conversation on one page you can open in a browser.conversations.json, the same conversations as data, with titles, timestamps and every message.
There are a few smaller files too, such as user.json for account details.
This breakdown comes from third-party guides
(ai-chat-importer,
tactiq); OpenAI’s own
page only says the ZIP includes your chat history and account data.
Step 2a: A few chats? Upload them to a Claude Project
If you need a handful of conversations, a Claude Project is the simplest place for them. Every chat in the Project can use its files.
Anthropic’s upload limits for Project files are 30MB per file and no cap on the number of files, but the total content must fit within Claude’s context window. HTML, JSON and TXT are all accepted. On paid plans, Projects use retrieval and can hold up to about ten times more.
So in practice:
- Open
chat.htmlin your browser, find the conversations you care about, and copy each into its own text file. Upload those. - Don’t upload all of
chat.htmlorconversations.jsonif you’ve used ChatGPT for a long time. A long history can pass the 30MB file limit, and even below it, years of chats won’t fit in context.
This keeps the important chats close at hand. The rest stay in the ZIP.
Step 2b: Lots of chats? Turn them into a folder
For your whole history, split conversations.json into one Markdown file per
conversation and keep them in a folder. Claude then opens only the
conversations a question needs, instead of loading everything at once.
You don’t have to write the script yourself. Unzip the export into a folder, open it in Claude Code (or Codex), and ask:
This folder has my ChatGPT export. Write and run a script that reads
conversations.json and saves each conversation as a Markdown file in
chats/, named YYYY-MM-DD-title.md. Put the title and date at the top,
then each message with "You:" or "ChatGPT:" before it. Each conversation
stores its messages as a tree (the "mapping" field); follow the branch
that ends at current_node so edited messages don't appear twice.
Skip empty and system messages. Don't change conversations.json.Check a few files against chat.html before you rely on them. The export
format isn’t documented by OpenAI and can change, so the script may need a
fix or two.
Once you have the folder, Claude can work with it in a few ways:
- Claude Code reads the folder you open it in, and
claude --add-diradds more folders to a session. - Cowork in Claude Desktop can connect the folder on a paid plan.
- Search is the weak spot. Claude finds files by name or by grepping for exact words. If you remember the idea but not the words, it can miss the chat.
There’s more on each option in how to give Claude access to a folder.
Where StashBase helps
StashBase is the writing workspace I build. It’s useful once your history is a folder of Markdown files:
- Browse and search the whole history. Open the folder as a project. Keyword search works across every conversation with no account. Search by meaning finds chats by idea rather than exact words; it stays off until you add an OpenAI or OpenRouter key.
- Ask Claude about it. Pick Claude Code or Codex in Chat and ask things like “What did I decide about the pricing page last spring?” Through MCP, Claude Desktop and other agents can search the same project.
- Write from it. Old chats become source material for new drafts in the same project, and the agent’s edits show up as suggested changes you accept or reject.
- Optional Wiki Pages. An agent task can turn the folder into topic pages that link back to the original chats.
StashBase doesn’t read the ChatGPT ZIP directly. Split the export first, as
in Step 2b, then open the chats/ folder.
Which way should you choose?
| What moves | Limit | Good for | |
|---|---|---|---|
| Memory import | What ChatGPT knows about you | No conversations | Getting Claude to know you on day one |
| Claude Project | Chats you pick and upload | 30MB per file; total must fit in context | A handful of important chats |
| Local folder | Every conversation, one file each | One-time split; search by exact words unless you add an index | Years of history you want to search and reuse |
If you’re switching for good, do two of these: run the memory import, then keep the full history as a folder for the day you need an old conversation.