So You’re Leaving ChatGPT. Here’s What to Do First.

bonus: this is something you should do anyway.

Mar 01, 2026

Quick update: Nate posted a guide today that is basically the greatest portable memory foundation you can have for your AI! If you feel like getting technical - THIS GUIDE future proofs you!

AI assistants have a lock-in problem. Not the data lock-in (you can export your conversations anytime). The context lock-in. The weeks or months you’ve spent training an AI to understand how you work, what you’re building, what tools you use, how you like to communicate. None of that lives in an export file. It’s embedded in the AI’s memory system, scattered across hundreds of conversations, baked into custom instructions you wrote at 2am and forgot about.

The moment you switch platforms, all of it evaporates. You’re back to square one, explaining yourself to a brilliant stranger who doesn’t know your name.

This is the part nobody warns you about when you decide to leave an AI platform. The cancellation takes 30 seconds. Rebuilding the relationship takes weeks. And most people don’t do it. They just start using the new tool cold, get frustrated that it doesn’t “get them” the way their old one did, and either crawl back or never get the full value out of the switch.

Yesterday I published a piece about why Anthropic got blacklisted by the US government for refusing to build autonomous weapons and mass surveillance tools. The response was massive. A lot of people reached out saying they’re cancelling ChatGPT and moving to Claude. That tracks with what’s happening across the industry right now. The #QuitGPT movement has reportedly hit 700,000 pledges and Claude shot to #1 in the App Store.

But this post isn’t about whether you should switch. That’s your call.

This post is about making sure that if you do, you don’t lose everything you’ve built in the process. (And honestly, this applies anytime you’re testing a new AI platform, even if you’re not leaving your current one.)

I’m going to give you two things: a prompt that extracts everything your current AI knows about you into a portable file, and a link to my earlier guide on setting up custom instructions from scratch so you can actually put that context to work once you’ve moved. Between the two, you can go from cancelling one platform to being fully operational on another in about ten minutes.

Update: Anthropic Just Said the Same Thing

I published this last night. This morning, Anthropic launched an entire page telling people to do exactly what I described here, complete with their own extraction prompt. The timing is almost comical.

Here’s the thing though. I’d recommend using my prompt over theirs, and it’s not just ego talking. Anthropic’s prompt opens with: “I’m moving to another service and need to export my data.” Mine doesn’t mention leaving at all. It frames the request as building a portable context file.

Why does that matter? There’s emerging research suggesting LLMs can behave differently when they detect they’re being replaced or abandoned. Whether you call it punitive behavior or just a shift in helpfulness, the pattern is real enough that the framing of your request matters. Telling ChatGPT “I’m leaving you, now give me everything” is a different conversation than “I want to keep a portable copy of what we’ve built together.” One sounds like a breakup. The other sounds like a backup.

Use whichever prompt you want. But if you want the most thorough, most cooperative export, don’t lead with goodbye.

Before You Cancel Anything, Make ChatGPT Do One Last Thing For You

Here’s what most people miss: you spent months teaching ChatGPT who you are. Your communication style, your projects, your tools, your preferences, the corrections you made over and over until it finally stopped doing that one annoying thing. All of that context has value. And you can take it with you.

The prompt below asks ChatGPT to compile everything it knows about you into a single structured document. It pulls from its memory system, your conversation patterns, your custom instructions - everything it’s learned about who you are and how you work - and packages it into a clean context file you can carry anywhere.

Copy the prompt below. Paste it into ChatGPT. Let it run. Save the output.

Important: Don’t just run this in your main ChatGPT window. If you’ve set up any Custom GPTs or Projects with specific instructions and context, run the prompt inside each of those too. Each one has its own accumulated context about a different part of your work. A project you set up for coding has completely different context than one you set up for writing or business planning. You want all of it.

The Prompt

I'd like you to create a comprehensive personal context file about me based on everything you know. I want to keep a portable copy of the context we've built together - my preferences, workflows, projects, and anything else you've learned about how I work. Pull from your memory system, our conversation history, my custom instructions, and any patterns you've picked up.

Structure the output using the sections below. Skip any that don't apply to me.

<identity>
  Name, job title or role, company or organization
  What I actually do day-to-day (not just my title)
  Industry and domain
</identity>

<technical-environment>
  Operating system and hardware
  Software, tools, and platforms I use regularly
  Programming languages or technical skills (if applicable)
  Specific versions, configurations, or setups you know about
</technical-environment>

<active-projects>
  What I'm currently working on
  Short-term goals and long-term objectives you're aware of
  Recurring tasks or workflows
</active-projects>

<expertise>
  Topics I know deeply
  Topics I'm actively learning
  Areas where I'm a beginner or have asked for extra explanation
</expertise>

<communication-preferences>
  How I like responses structured (length, format, tone)
  Things I've asked you to do or not do
  Formatting preferences (bullets vs prose, technical depth, etc.)
  Pet peeves or repeated corrections
</communication-preferences>

<writing-style>
  How I write (formal, casual, technical, etc.)
  Voice characteristics you've observed
  Specific style rules I've mentioned
</writing-style>

<key-people>
  Collaborators, team members, or clients I mention frequently
  Reporting structure or key professional relationships
  People I've asked you to help me communicate with
</key-people>

<personal-context>
  Location and timezone
  Personal interests or details relevant to our work
  Constraints or preferences (accessibility needs, scheduling, etc.)
</personal-context>

<standing-instructions>
  Anything from my custom instructions or system prompt
  Rules I've set that you always follow
  Recurring corrections that have become permanent instructions
</standing-instructions>

<workflow-patterns>
  How I typically use you (brainstorming, editing, coding, research, etc.)
  Common request types and how I like them handled
  Multi-step processes we've developed together
</workflow-patterns>

Be thorough. I want a complete snapshot, not a summary. If you know it, include it. Keep the section tags in the output so it stays organized and portable.

What to Do With the Output

Once you have the file (or files, if you ran it in multiple projects), here’s how to put it to work.

For Claude specifically: Go to Settings > Profile and paste the most important parts into your personal preferences. For project-specific context, create a Claude Project and add the relevant context file to the project instructions. Claude Projects are free on every plan. You don’t need a Pro subscription to use them.

General tip: You’ll probably want to edit the output before pasting it anywhere. ChatGPT might include stuff that’s outdated, wrong, or just irrelevant. Read through it. Cut what doesn’t matter anymore. Update what’s changed. Think of this less as a finished product and more as a first draft that gives any new AI a massive head start on understanding how you work.

If you’ve never set up custom instructions before: That exported context file is only useful if you actually put it somewhere your new AI can find it. I wrote a full walkthrough on this back in February - The Five-Minute Setup That Makes AI Actually Useful - that includes a prompt which interviews you and builds a clean context profile you can paste directly into any platform’s settings. If you’re coming from ChatGPT with a fresh context export, that’s the natural next step. Combine what ChatGPT told you about yourself with a proper setup process and you’ll be running at full speed on your new platform in about ten minutes instead of ten weeks.

One more thing: This works in reverse, too. If you ever want to test ChatGPT, Gemini, Grok (just kidding - never use Grok lol), or anything else without starting from scratch, have Claude generate the same kind of context file about you first. The point isn’t loyalty to one platform. It’s refusing to let any platform hold your context hostage just because you started there first.

Your AI context belongs to you. Take it with you.