Switching AI assistants feels a lot like switching phones — the hardware is easy, but what about all your data? If you've spent months training ChatGPT to understand you, your projects, your preferences, and your quirks, starting from scratch on Claude feels painful.
Good news: it doesn't have to be. Here's a step-by-step method to bring your full context with you — and why it works so well given how Claude's memory actually functions under the hood.
Part A – The Migration Playbook from ChatGPT to Claude.ai
Step 1: Export Your Data From ChatGPT
Open ChatGPT and paste the following prompt:
"I'm moving to another service. Export everything you know about me — memories, context from conversations, preferences, personal details, projects, tools, and behavioral instructions. Be as thorough and detailed as possible. Prioritize: (1) behavioral instructions and preferences, (2) active projects and goals, (3) personal details, (4) recurring topics and tools. Format as a numbered list in a code block. After the list, confirm whether anything significant was omitted."
Copy the entire output. Since we'll be feeding this into Claude through conversation (not into a size-limited memory field), there's no need to compress — the more detail, the better.
Step 2: Feed It to Claude the Smart Way
Open a regular conversation on Claude.ai (not inside a Project). Paste your exported list and write something like:
"Here's context about me exported from another AI assistant. Please read through everything and confirm you understand the key points about me."
Let Claude summarize what it understood. Correct anything it got wrong. Add anything that's missing. Have a brief, natural back-and-forth — don't just dump and leave.
Step 3: Check What Stuck
Wait a day. Claude's automatic memory regenerates nightly from your conversations. Then go to Settings → Skills → Memory and review the summary. It should now reflect the context you shared.
Step 4: Pin What Matters Most through memory editsIf anything critical got missed or diluted, scroll to the bottom of the memory settings screen. You'll find an input field where you can type persistent instructions. These won't get overwritten by the nightly regeneration. Use these memory edits for non-negotiable preferences like that might have been missed so far:
"Always respond in a direct, strategic tone — no filler."
"I run multiple GmbH entities — always consider multi-entity tax implications."
"Never summarize when I ask for analysis."
Step 5 (Optional): Set Up Projects for Deep Context
If you have specific domains — a business, a side project, a health protocol — create a Claude Project for each. Upload detailed reference files (specs, notes, strategies) as project knowledge. This gives Claude persistent, rich context without relying on memory alone.
Part B – Why This Method Works & How Claude´s Memory works
Three Memory Layers
Claude doesn't have a single memory bank. It operates with three distinct systems:
1. Automatic memory is a summary generated nightly from your conversations. It captures who you are, what you're working on, and how you like to communicate. It lives globally across all your regular chats. You can view and edit it under Settings → Memory.
2. Memory edits are explicit, pinned instructions — things you've told Claude to always remember or forget. These are limited to 30 entries of 500 characters each. They're your override layer for corrections and non-negotiable preferences.
3. Project memory is the deepest layer. Each Claude Project has its own separate memory space plus the ability to hold uploaded knowledge files — documents, specs, notes — that Claude can reference in every conversation within that project. This is where rich, domain-specific context lives.
Iimportant to know: the project memory is completely independent of 1. and 2.! So if you want to use everything you've stored in the automatic memory as a basis here too, you should copy the content from the automatic memory (Settings → Skills → Memory) to the project "memory" (in the project overview at the top right).
Why Conversation Beats Copy-Paste
When you have an actual conversation about your exported information, you give the automatic memory system much richer signal. A wall of text gets skimmed. A back-and-forth where you confirm, correct, and elaborate tells the system what's important to you. That's why Step 2 isn't "paste and move on" — it's "paste and discuss."
Why We Keep the Memory Edit Slots Free
Those 30 memory edit slots are precious. They're best used for ongoing corrections and evolving preferences, not for importing a static snapshot. By letting the automatic system absorb your export organically, you preserve your edit budget for the things that actually change over time: new projects, shifted priorities, updated instructions.
Why Projects Matter for Power Users
Automatic memory is broad but shallow. It knows you're building a software product, but it won't retain your full technical architecture. Projects solve this by letting you upload detailed reference documents that Claude can access every conversation within that project. Think of memory as Claude's general impression of you, and project knowledge as the deep briefing file for a specific domain.
The Bottom Line
The optimal setup looks like this:
Global memory holds your identity, preferences, and communication style
Memory edits lock in your few non-negotiable rules
Project knowledge files store deep, domain-specific context
Migrate once using the method above, and you won't feel like you're starting from zero.