Breaking up with ChatGPT?

Take your stuff with you.

A lot of people are making the switch to Claude right now, myself included, and this isn't just adding another AI to the mix. We're breaking up with the other one.

And like most breakups, it's hard because ChatGPT already knows me. My tone, my work, how I think. Starting over means losing months of context I didn't realize I'd built up.

Thankfully, I don't have to start from zero.

Anthropic built a page for this at claude.com/import-memory, which helps, but the prompt they provide doesn't pull everything out reliably. Here's a two-part process that does a better job.

Wait, why are people leaving?

Two reasons.

The first is what happened with the Pentagon. The US Department of Defense told Anthropic to remove its restrictions on Claude being used for mass surveillance and autonomous weapons — or lose the government contract. Anthropic refused. They got blacklisted and replaced with OpenAI within days.

You can have your own opinion on that. But when a company walks away from hundreds of millions rather than comply, that's not a PR move. That's a decision with real consequences, and it changes how you think about who you're working with.

The second is that Claude has gotten genuinely better. Writing, analysis, reasoning, day-to-day thinking — for most of the work people actually do, it's now the stronger tool. The recent updates have been significant.

The process

Open a new ChatGPT chat and paste in Prompt 1 below. It pulls your stored memories and custom instructions in one clean block. Follow up in the same chat with Prompt 2, which surfaces the patterns from your actual conversation history — how you write, what you work on, what you regularly ask for.

Before you transfer anything, edit it. Most people skip this and then wonder why Claude still feels generic. Paste everything into a document, go through it, and cut what doesn't matter — old one-off tasks, finished projects, anything you asked about once and never came back to. Keep your tone, your workflows, your recurring projects, how you like answers structured. That's the stuff that makes an AI actually useful to you specifically.

Once it's clean, go to claude.com/import-memory, paste it in, and you're done. Give it up to 24 hours to fully integrate, then start a new chat and ask "What do you know about me?" to confirm it worked.

One thing to know: Claude's memory requires a paid plan, starting at $20/month for Pro. Free users can still paste the profile at the start of any chat — it works for that session, it just won't persist.

The prompts

Prompt 1: Extract memories and instructions

Copy and paste this into a new ChatGPT conversation:

I'm moving to a different AI assistant and I need to export everything you know about me. Please provide ALL of the following in a single, clearly structured response:

1. STORED MEMORIES

List every single memory you have stored about me. Write them out verbatim, exactly as they appear in your memory. Don't summarise or paraphrase — I want the raw entries.

2. CUSTOM INSTRUCTIONS

Reproduce my full custom instructions exactly as written — both:

- "What would you like ChatGPT to know about you?"

- "How would you like ChatGPT to respond?"

If either field is empty, say so.

3. PREFERENCES AND CONTEXT

Based on our interactions, list any preferences, recurring context, or personal details you've picked up that might not be in my stored memories. This includes things like:

- My profession and industry

- Tools or platforms I use regularly

- Writing style preferences

- How I like information structured

- Topics I frequently ask about

- Any specific formatting preferences

Be thorough. I'd rather have too much than miss something important.

Prompt 2: Summarise chat history patterns

Paste this as a follow-up in the same chat:

Now I need you to go deeper. Based on the full history of our conversations — not just your stored memories — provide a summary of:

1. WRITING STYLE PROFILE

How do I write? What's my tone, sentence length, vocabulary level, and any distinctive patterns in how I communicate?

2. RECURRING TASKS

What types of tasks do I most frequently ask you to help with? List them in rough order of frequency.

3. PROJECTS AND WORKFLOWS

What ongoing projects or workflows have come up repeatedly? Include any context about what stage they're at or how I approach them.

4. OPINIONS AND PREFERENCES

What opinions, preferences, or strong views have I expressed that would help a new AI assistant understand how I think?

5. THINGS TO AVOID

Based on our interactions, is there anything I've pushed back on, corrected you about, or asked you not to do? List any "don't do this" patterns.

Format everything clearly with headers. This will be pasted into another AI's memory system, so write it in a way that's useful as a reference document rather than a conversation.

Make it work better.

If ChatGPT doesn't give you much, memory was probably switched off, or you haven't used it long enough to build anything up. In that case, write your own profile from scratch. A hand-written one tends to be more accurate anyway — you know your context better than any AI does.

If you use Custom GPTs, those won't be in the export. Recreate them as Claude Projects — paste your GPT's system prompt into a Project and it works the same way.

If you want everything, you can export your full ChatGPT history via Settings → Data Controls → Export Data. OpenAI emails you a zip file. Most people don't need this, but if you've been using it for years and want a full picture, it's there.

The whole thing takes about ten minutes. After that, Claude knows your tone, your work, your preferences — and you're not spending the first week re-explaining yourself.