Export ChatGPT to Claude: what moves and how

You can’t export ChatGPT chats into Claude’s chat list. Anthropic’s help center says it plainly: importing conversation history from another AI provider isn’t possible. What you can do is move the two things that matter. Claude has a built-in memory import that takes what ChatGPT remembers about you. Your old chats move as files: get them out of ChatGPT as Markdown, then add the ones you still need to a Claude project, where Claude reads them in every chat inside that project. Below is what moves, what doesn’t, and the two routes to the files.

What moves from ChatGPT to Claude

Move ChatGPT memory and custom instructions by hand

Memory is the part Claude imports directly. In Claude, open Settings → Your Claude and choose Start import (in the older layout, the Memory section). Claude shows you a prompt written to pull your context into one chat. Run it in a new ChatGPT conversation, copy the answer, paste it into Claude’s import box and click Add to memory. Claude splits the text into individual memory entries you can review and edit. Anthropic calls the feature experimental and says Claude focuses on work topics, so personal details may not stick. The import is available on Free, Pro, Max, Ultra, Team and Enterprise plans.

Custom instructions don’t import, but they’re short. In ChatGPT, copy them from Settings → Personalization → Custom instructions. In Claude, open Settings → Your Claude and paste them into Instructions, which apply across all your chats. A custom GPT’s instructions go the same way, into a Claude project’s instructions instead.

Route 1: ChatGPT’s data export, then a Claude project

The official route costs nothing and works without an extension. It also exports your whole account, arrives days later, and gives you a data dump rather than readable files.

1. In ChatGPT, open your profile menu, go to Settings → Data controls, select Export under Export data, and confirm. OpenAI emails a download link up to 7 days later. The link lives 24 hours and only opens while you’re signed in to the account that asked, so download the ZIP as soon as it arrives. Details in OpenAI’s export article.
2. Open the ZIP. Your chats are in conversations.json, or in numbered files like conversations-000.json for a large account. chat.html is a viewer for reading, not for uploading.
3. Convert the JSON to one Markdown file per chat. The export stores each chat as a tree of message nodes, so use an open-source converter (search GitHub for “chatgpt export to markdown”), or Carry, below. The export explainer walks through the format.
4. In Claude, create a project and add the Markdown files you still need to its knowledge. Give the project a one-line instruction such as “These files are my past ChatGPT conversations; use them as background.”

Two limits catch people out. The export doesn’t include files you uploaded to ChatGPT, chats you deleted, or any way to pick one chat: it’s the whole account. And it’s a snapshot that starts going stale the day it arrives, so anything new means the whole wait again. There’s no Export button at all in Business or Enterprise workspaces. For the full picture of what’s in the ZIP, see the ChatGPT backup guide.

Route 2: Carry saves the Markdown for you

We make Carry, so weigh this accordingly. It’s an extension for Chrome and other Chromium browsers that saves each ChatGPT chat as a Markdown file in a folder you pick. For a switcher, that means: you pick which chats to upload, your uploaded files are kept as originals, there’s no 7-day wait, and each file updates in place as the chat changes. It reads chats through ChatGPT’s internal API, the same data the app loads, so a long chat saves with the middle intact, and it saves on its own every couple of hours while your browser is open. Carry never uploads your chats, and there’s no Carry account or API key.

1. Install Carry, connect ChatGPT, and pick a folder. The first sync saves your newest chats into carry-backup/chatgpt/, with uploaded files and files ChatGPT made for you in files/ beside them.
2. Wait for the sync to finish, then open the folder. Each chat is a .md file you can read, search or open in Obsidian.
3. Create a Claude project and add the files you want Claude to know about. When a ChatGPT chat changes, Carry updates its file.

Carry saves Claude chats and projects too, so the same folder holds both sides as you switch. The ChatGPT exporter page shows the saved files next to the official export.

Claude project limits to know

The numbers that decide how many chats fit, from Anthropic’s upload limits article:

Per project file

30 MB

Number of files

No fixed cap, but the extracted text must fit Claude’s context window; on paid plans retrieval kicks in past that, for up to 10x the capacity

Accepted types

PDF, DOCX, CSV, TXT, HTML, ODT, RTF, EPUB, JSON and XLSX, Markdown is plain text (if a .md file is refused, rename it .txt), and a small conversations.json can go up as-is

Per chat upload

500 MB per file, up to 20 files per message, if you’d rather attach to one conversation than a project

A Markdown chat is a few kilobytes, so the per-file limit never bites. Context does. Dozens of long chats fill a project fast, which is another reason to upload the chats you’ll actually reuse.

Common questions

Carry processes chats on your device and never uploads them. It downloads a signed compatibility file from carry.md and sends usage counts, which you can switch off; see privacy details. The ChatGPT save guide covers the export in more depth, and the Claude export guide the other side.