I’ve been comparing the ChatGPT-to-Claude switch to a breakup for a few weeks now. And the responses told me something: a lot of you already know you want to make the move. You’ve been spending more time with Claude. The conversations are better. The work is better. You just haven’t packed your bags yet.
That’s the hard part of any transition. The new relationship is clearly working, but your entire history, your context, your patterns, your projects, all of it still lives at the old place. Starting from zero feels wasteful. So you stay longer than you should, splitting time between two apartments, pretending that’s sustainable.
It isn’t. And you don’t have to.
There are three levels to this migration. Pick the one that matches how deep you were with ChatGPT, and how clean you want your fresh start to be.
Level 1: Grab the essentials (2 minutes)
This is the overnight bag. Claude built a direct import for exactly this situation.
Go to your Claude account. Click your profile picture, then Settings, then the Capabilities tab. You’ll see a button that says “Import memory from other AI providers.” Click it.
Claude gives you a prompt to copy. Take that prompt over to ChatGPT and run it. ChatGPT will spit out a structured summary of everything it knows about you, your preferences, your work patterns, the context it’s built up over time.
Copy that output. Paste it back into Claude’s import field. Done.
One thing worth knowing: before you run that prompt in ChatGPT, go into ChatGPT’s settings and make sure all memory sources are turned on. Profile, personalization, “reference saved memories,” “reference browser memories,” “reference chat history.” All of it. And switch to a reasoning model before you run the prompt. The thinking models do a better job pulling from deeper in your history. They take more passes through your account and surface things the default model skips.
Once Claude has that import, it formats everything into clean, categorized memories that regenerate nightly as you keep using it. Read through them. If something’s off, click the edit icon and correct it. You’re teaching a new partner your preferences. A little patience early pays off.
Level 2: Move the furniture (10 minutes per project)
If you used ChatGPT Projects, you had context, files, and custom instructions organized by workstream. Those don’t come over in a basic memory import. You need to recreate them.
The process is simple. Go back to ChatGPT, open a project you want to migrate, and run that same import prompt inside the project chat. The reasoning model matters even more here because project-level context is often buried in conversation threads, not stored as explicit memories.
Take what ChatGPT produces and go to Claude. Create a new Project with the same name. Paste the extracted context into the project instructions. Upload any files you had attached in ChatGPT. Those files become Claude’s knowledge base for that project, and Claude can reference them in every conversation within it.
If you had five projects in ChatGPT, budget about an hour. Most of that time is just copying, pasting, and uploading files you already have on your computer.
Level 3: The full move (30 minutes, mostly waiting)
This is where you bring everything. Every conversation, every file, every pattern from years of daily use.
Go to ChatGPT Settings, then Data Controls, then Export Data. Confirm the export. Depending on how much history you have, the download link arrives by email anywhere from a few minutes to 24 hours later. My own export was over a gigabyte, which tells you something about how much context we hand these tools without thinking about it.
Download the zip file. Inside you’ll find a file called chat.html. That file contains every conversation you’ve ever had with ChatGPT.
Here’s where it gets interesting. Open Claude’s desktop app and switch to Cowork mode. Cowork lets Claude interact directly with files on your computer, and it doesn’t have the upload size limits you hit in the browser. Point it at your exported folder or upload that chat.html file directly.
Then give Claude a prompt that asks it to analyze your full history and build an abstraction. Not a raw data dump, but a structured profile: how you think, how you communicate, what you’re working on, how you make decisions, what you don’t tolerate. The output reads like a cognitive brief, and honestly, it’s a strange mirror to look into. Years of daily conversations distilled into a few pages of patterns you might not have articulated yourself.
Copy that brief. Go to Claude’s Settings, Capabilities, Memory, and paste it in with a note asking Claude to add it to your working memory.
The reason you want the abstraction rather than the raw export is practical. Claude can always access your full ChatGPT files locally through Cowork when you need something specific. But for everyday conversations, the compressed profile gives Claude what it actually needs: a working model of who you are and what you care about. Signal, not noise.
The thing people get wrong
The mistake I see is treating this like a data migration when it’s really a context migration. Dumping thousands of conversations into a new tool doesn’t make that tool useful. What makes it useful is when it understands your decision-making patterns, your communication preferences, the projects you’re actively running, and the kind of answers that waste your time.
That’s what these three levels build toward. Level 1 gets you the essentials so Claude stops feeling like a stranger. Level 2 rebuilds your workspaces so you can actually do project work. Level 3 gives Claude the deep context that took years to build in ChatGPT.
You don’t need all three. But if you’ve been splitting your time between two tools because you can’t quite let go of the history you built, now you know: the history travels. The context transfers. And the new place is ready whenever you are.
Pick a level. Move your stuff. And stop paying rent in two places.