Moving to Claude: How to Migrate Without Starting from Scratch

Apr 2, 2026 · 9 min read

You have been using ChatGPT for months. Maybe a year. It knows your communication style, your projects, your preferences, the rules you have set for how it works with you.

Then you decide to try Claude. You open a fresh conversation. And you are back to square one. A generic assistant with no idea who you are.

Everything you built, every correction, every "don't do that, do this instead," gone.

This is not a reason to stay. It is a problem to solve. And it takes less time than you think.

Why Claude

A lot of people are making this move right now, and it is not random.

ChatGPT's output quality has been declining. The responses have become more verbose, more generic, more cautious. OpenAI's Pentagon deal eroded trust for many users. Meanwhile Claude has quietly become the stronger tool for the kind of work that matters: writing that sounds like you, reasoning that holds up, and skills that let you encode your actual methodology so the AI follows your process instead of guessing.

The switch is worth making. Here is how to do it without losing what you have built.

Two Paths

Path A is the quick route. Claude has a built-in import feature that pulls your context from ChatGPT in about five minutes. It captures the surface level: key facts, preferences, working style.

Path B goes deeper. A structured extraction that captures your voice, your frameworks, your specific AI instructions, and your Custom GPTs. This takes about thirty minutes but produces a reusable profile document that works with any AI tool, not just Claude.

If you just want to get up and running, start with Path A. If you want the complete migration, go straight to Path B. You can always do both: quick import first, deep extraction when you have the time.

Path A: Quick Import

Claude has a built-in memory import feature. It generates a prompt, you paste it into ChatGPT, copy the response back, and Claude commits it to memory.

Here is how:

Open claude.ai and click your profile picture (bottom left).
Go to Settings > Capabilities > Memory.
Click Import memory from other AI providers.
Claude generates a prompt. Copy it.
Paste the prompt into ChatGPT. Let it respond.
Copy the full response from ChatGPT.
Paste it back into Claude and confirm.

Done. Claude stores this in its memory and uses it across every future conversation.

This is a solid starting point. But it captures what ChatGPT chooses to summarise, which is not everything. Your voice patterns and custom instructions, the frameworks you have built, the specific way you want AI to work with you: those need the deeper approach.

Path B: Deep Migration

This is the thorough version. You end up with a structured profile document that captures everything your current AI knows about you.

Step 1: Extract What ChatGPT Knows

ChatGPT does not re-read your old conversations. It only has access to its stored memories and your Custom Instructions. This prompt captures what it remembers, which is a strong starting point but not the full picture.

Paste this into ChatGPT:

I'm migrating to a new AI assistant. I need you to create a
complete profile document that captures everything you know
about me.

Draw from my stored memories and Custom Instructions. Be
exhaustive. Include specific names, numbers, dates, tools,
terminology, and examples. Raw detail matters more than tidy
summaries.

Structure the document in three parts:

PART 1: WHO I AM
Everything relevant about my life, work, and world. Create
whatever sections make sense for me specifically. This might
include my role, expertise, projects, relationships, goals,
tools, industry, interests, or anything else that shapes how
I'd want an AI to understand me. Organize by what matters most.
Do not use generic sections that don't apply to me. Only include
categories where you have real information.

PART 2: HOW I COMMUNICATE
My writing and speaking style. Describe my tone (5-7 adjectives),
sentence patterns, vocabulary habits, and formatting preferences.
Include words or phrases I use often, and any language or patterns
I've told you to avoid. Give 2-3 short examples that capture my
authentic voice.

PART 3: HOW I WANT AI TO WORK WITH ME
Every instruction, rule, preference, or boundary I've set for how
you interact with me. Include standing rules, priority order for
my work, what to never do, and how to handle uncertainty. If I've
given you specific frameworks or methods to follow, include those.

Rules for this document:
- Write in third person ("[Name] is..." not "You are...")
- Do not skip anything. If a detail seems minor, include it anyway.
- Do not generalize. "Works in tech" is useless. "Senior DevOps
  engineer at Acme Corp managing AWS infrastructure for a 40-person
  team" is useful.
- If you know my preferred terminology vs terms I've retired,
  flag both.
- Include memories and Custom Instructions verbatim, woven into
  the relevant sections.
- At the end, list any areas where you have limited or no
  information so I know what to fill in myself.

What comes back will surprise you. ChatGPT has been quietly building a picture of who you are across every conversation. This prompt forces it to lay it all out in a format you can actually use.

Step 2: Extract Your Custom GPTs

If you have built Custom GPTs in ChatGPT, each one contains a system prompt and possibly uploaded knowledge files. Those represent real intellectual property and operational logic.

For each Custom GPT:

Open it and click Configure.
Copy the full system prompt (the instructions you wrote).
Download any uploaded knowledge files.

Save the system prompts into your profile document under Part 3. Keep the knowledge files for uploading into Claude Projects later.

If you have not built any Custom GPTs, skip this step.

Step 3: Fill the Gaps

ChatGPT will flag what it does not know at the end of its output. Read the full document and add anything important that is missing.

This is your chance to strengthen the profile. Add the things you never explicitly told ChatGPT but that shape how you work: your values, your non-negotiables, the words you would never use, the positioning that makes your work yours.

You know your context better than any AI does. The goal is a single document that could bring any AI assistant up to speed on who you are, how you work, and how you want AI to show up for you.

Step 4: Load into Claude

You have two options for where to put your profile.

Option 1: User Preferences. Go to claude.ai > Settings > Profile > User Preferences and paste your completed document. Claude reads it automatically in every new conversation. If it exceeds the character limit, prioritise Parts 2 and 3 (communication style and AI instructions). Upload the full document as a file when you need the deeper context.

Option 2: A Project. Create a new Project in Claude. Paste your profile into the Project Instructions, or upload it as a document. Every conversation inside that Project starts with your full context loaded. This is the better option if your profile is long or detailed. Projects have a higher context limit than User Preferences.

Step 5: Verify

Start a new conversation in Claude and ask:

Who am I and what matters most to me right now?

If the answer is accurate, you are done. If anything is off, refine that section and update your profile.

Migrating from Gemini

The same approach works. Paste the extraction prompt from Step 1 into Gemini instead of ChatGPT. Gemini draws from its memory of your conversations and saved preferences. The rest of the process is identical: fill the gaps, load into Claude, verify.

Full Conversation Export (For Those Who Want Everything)

The methods above capture what your AI remembers. If you want to capture everything, not just the selective memory, you can export your entire conversation history.

In ChatGPT, go to Settings > Data Controls > Export Data.
OpenAI emails you a zip file containing every conversation you have ever had.
Unzip it. The file you want is called conversations.json.

Start a new conversation in Claude, upload the file, and use this prompt:

I've uploaded my complete ChatGPT conversation history. Analyse
it and create a comprehensive profile of me using this structure:

PART 1: WHO I AM
Everything relevant about my life, work, and world. Create
sections based on what the data reveals. Only include categories
with real information.

PART 2: HOW I COMMUNICATE
My tone, sentence patterns, vocabulary, formatting preferences.
Words I use often and patterns I avoid. Include 2-3 examples.

PART 3: HOW I WANT AI TO WORK WITH ME
Every instruction, preference, or boundary I've expressed.

Write in third person. Be specific. Flag any areas where the
data is thin so I know what to add manually.

This takes longer but gives Claude the full picture rather than relying on ChatGPT's selective memory.

Keep It Current

Your profile document is not a one-time exercise. Update it when something significant changes: a new role, a new project, a shift in priorities.

Claude builds memory from your conversations over time, so the profile gives it a head start. Your ongoing interactions sharpen it from there.

If you want to go further, the Human Stack is the architecture that turns this single profile document into a complete system: five layers of context, organised by how fast they change, that power every AI interaction you have.

The profile you just built is Layer 1. There are four more.

What You Walk Away With

After completing this guide, you have:

A structured profile document that works with any AI tool, not just Claude
Your communication style captured so AI output sounds like you
Your rules, preferences, and frameworks preserved
A portable asset you can reuse whenever you switch tools or set up a new workspace

The migration takes thirty minutes. The profile you build lasts as long as you keep it current.