How-to guide

How to move a ChatGPT conversation to Claude (and back) without losing context

By Prashant Kumar, updated 22 October 2026. About 6 minutes to read.

You are deep into a ChatGPT conversation and want Claude to pick it up, or the other way round. Moving your account memory and moving one live chat are two different jobs. This guide covers the second: how to carry a single conversation, with its goal, decisions and code, into the other AI.

Memory import vs. moving one chat

Most guides about "switching from ChatGPT to Claude" are about memory: the facts the assistant remembers about you, such as your name, job and preferences. Claude has an official memory import for that, and guides like this one walk through it.

Memory does not bring a conversation with it. If you were halfway through debugging a function, planning a trip or drafting a report, Claude won't know where you left off. For that, you need to move the chat itself.

Option 1: a handoff prompt (no tools)

This works everywhere and costs nothing. It takes about two minutes.

1. Ask ChatGPT for a handoff summary. In the chat you want to move, send this prompt:

   Write a handoff brief so another AI assistant can continue this conversation.
   Include: my goal, the facts I gave you, decisions we made, what we tried
   and what happened (keep exact errors), the current state, open questions,
   and the next step. Keep code, file names and numbers word for word.
   Do not add anything that wasn't in our conversation.

2. Copy the reply. Check it quickly: is anything important missing, such as a constraint you mentioned early on?
3. Open a new chat in Claude and paste it, starting with a line like: "I'm continuing work from another assistant. Here is the brief. Confirm what you understand, then continue."
4. Re-upload files. Attachments never travel with text. Upload the same files again in Claude.

Why a summary and not the whole transcript? A long transcript fills Claude's context with repetition, costs more of your usage and can confuse the new chat. A short brief gives Claude what it needs and leaves room to work.

The weak spot: the summary is written by the AI you're leaving. It can drop details or quietly change code. Always check the code blocks.

Option 2: one click with mcosec.ai

mcosec.ai is a free Chrome extension built for exactly this. It doesn't ask the AI to summarise itself. It reads the conversation on the page and compacts it on your device.

1. Open the ChatGPT chat and click the mcosec.ai icon in Chrome's toolbar.
2. Press "Read this chat". Your prompts and ChatGPT's answers are read, only at that moment.
3. Pick how much to keep: Short, Standard or Detailed. Your first message (usually the goal), the latest messages and every code block are kept, while older answers and filler ("Sure! Here's…", "I hope this helps!") are shortened. A meter shows the tokens before and after, and you can edit the result.
4. Choose Claude and press "Copy and continue in Claude". Claude opens with the compacted chat already in the message box. Read it, then press Send yourself.

Nothing is sent to a server, and there's no account. Because the code is copied word for word from the page, it isn't rewritten by an AI along the way.

Option 3: the official import and export tools

• ChatGPT data export (Settings → Data controls → Export data) emails you a ZIP of all your chats. It's a backup, not a way to continue one chat: Claude can't open the ZIP as a conversation.
• Claude's memory import moves memories, not conversations (see above).
• Gemini added tools in 2026 to import ChatGPT and Claude chat history. That suits moving your whole history to Gemini, not handing one live chat to Claude.

Claude to ChatGPT

Everything above works in reverse. With the handoff prompt, send it in Claude and paste the brief into a new ChatGPT chat. With mcosec.ai, read the chat on claude.ai and choose ChatGPT as the destination. A common reason is hitting Claude's usage limit mid-task.

Tips for a clean handoff

• Say what you want next. End with the very next step ("now add tests for the parser"), so the new AI doesn't start by re-explaining everything.
• Keep exact errors and versions. "TypeError: x is undefined at app.js:12" helps far more than "it crashed".
• Re-attach files. No method moves attachments, so upload them again.
• Check the first reply. Ask the new AI to confirm the task and where it stands before it writes anything.

Questions