Migrate ChatGPT Pro → Claude

Codified from Gmail draft 19dc1047fd2b7dc2 · Apr 24 2026

Honest answer first

There is NO automatic migration. OpenAI and Anthropic don't share data. Anyone who tells you otherwise is selling something.

What you CAN do, in 30–45 min: extract the 3 things that actually matter (memories, custom instructions, recurring projects) and seed Claude with them. Skip the conversations dump — it's 95% junk.

The 80/20 — what's actually worth migrating

KEEP:
1. ChatGPT "Memories" (the saved facts about you)
2. Custom Instructions (the "how to respond to me" block)
3. 5–10 recurring chat threads you reference often (named projects, recurring workflows)

SKIP:
- The full conversations.json export. It's 50–200MB of one-shot questions you'll never re-read. Importing it makes Claude slower at finding signal, not better.

Step 1 — Export your ChatGPT data (5 min, then wait for email)

1. ChatGPT → bottom-left profile icon → Settings
2. Data Controls → Export Data → Confirm Export
3. OpenAI emails you a download link within 1–24 hours (usually <2 hr)
4. Download the ZIP when it arrives. Extract to a folder.

What's in the ZIP:
- conversations.json — every chat ever (skip)
- chat.html — browseable version (use this to scan, not import)
- user.json — your account metadata (skip)
- message_feedback.json (skip)

Step 2 — Pull your ChatGPT Memories (5 min)

1. ChatGPT → Settings → Personalization → Memory → Manage
2. Copy every memory entry into a single text file. One per line.
3. Save as: chatgpt_memories.txt

That's your gold. These are the persistent facts ChatGPT learned about you over time.

Step 3 — Pull your Custom Instructions (2 min)

1. ChatGPT → Settings → Personalization → Custom Instructions
2. Copy both fields:
   - "What would you like ChatGPT to know about you?"
   - "How would you like ChatGPT to respond?"
3. Save as: chatgpt_custom_instructions.txt

Step 4 — Identify your recurring threads (10 min)

Open chat.html from the ZIP. Scan the chat list. For any thread you've returned to 3+ times or that represents a real "project," note:

• Title
• Topic in 1 line
• Key facts/context that aren't already in CONTEXT.md

You're looking for maybe 5–10 of these. Examples:
- MIS strategy iterations
- Eden Gardens planning sessions (if any pre-Claude)
- STR Orlando setup decisions
- Any consulting framework you built

Copy the meaningful chunks into a single markdown file: chatgpt_recurring_projects.md

Step 5 — Seed Claude (10 min)

A) Memories → Claude memory
Open a Claude chat. Paste your chatgpt_memories.txt and say:

Add these as memory edits, one per line. Skip anything stale or conflicting with what you already know about me.

Claude will use the memory_user_edits tool.

B) Custom Instructions → Claude user preferences
Settings → Profile → Personal Preferences. Paste the cleaned ChatGPT custom instructions. Edit for tone — they were written for ChatGPT.

C) Recurring projects → Claude Projects
For each recurring project, decide:
- Already exists as a Hook Street Project? → upload the markdown to that Project's knowledge base
- New project? → create one, upload markdown
Skip anything that's stale.

Step 6 — Done state

You're done when:
- Claude knows the persistent facts ChatGPT knew (memories migrated)
- Claude responds in your preferred style without you re-explaining (custom instructions migrated)
- Recurring projects have their own knowledge base in Claude (no need to re-explain history)
- conversations.json is in a folder somewhere you can grep if you ever need a specific old thread (don't import it)

Time budget

• Step 1: 5 min active + 1–24 hr wait
• Step 2: 5 min
• Step 3: 2 min
• Step 4: 10 min (this is the longest — actual scanning)
• Step 5: 10 min

TOTAL: ~30–45 min active

When to do this

Trigger: you find yourself re-explaining the same context to Claude that you already explained to ChatGPT 6 months ago. Until then, it's not urgent.