What's the manual way to move ChatGPT context to Claude?

Export your ChatGPT data (Settings → Data Controls → Export). Inside the ZIP, open user.json / memory.json — that's the short list of pinned facts. Copy the bits you want Claude to know, paste them into Claude's custom instructions or a new Project's context. You'll lose the rich conversational history, but the structured preferences make the jump.

Will Claude understand ChatGPT's memory format?

Treat the format as 'plain text written by an AI'. Claude reads it fine; it just won't know it came from ChatGPT. If you paste it into custom instructions, prefix it with something like 'Here's context another AI had built up about me — use it as background.'

What about my full ChatGPT history? Can I dump that into Claude?

Technically yes, but Claude's context window will fill instantly. The conversations.json file is usually hundreds of MB for heavy users. You need to distill before you paste — extract the cross-cutting atoms (decisions, recurring people, preferences, themes) and paste those, not the raw transcripts.

Is there a tool that does this transfer automatically?

Konshus is close to it. The ChatGPT importer reads your export, extracts atoms, and assembles a portable persona. The same persona pastes cleanly into Claude, Gemini, or back into a new ChatGPT account — that's the point of doing the work once.

Even with perfect transfer, Claude will respond differently than ChatGPT did. Same context, different model behavior. The transfer preserves what you've told the AI; it can't preserve how the AI felt to talk to. That part is genuinely lost across providers.

A guide · ~4 min read

Can you transfer ChatGPT memory to Claude?

Q: Can you transfer ChatGPT memory to Claude?

Not directly. OpenAI and Anthropic don't talk to each other. There's no native migration tool. You can do it manually by exporting your ChatGPT data, pulling out the relevant context, and pasting it into a new Claude Project or custom instructions. The cleaner path is a vault that distills the export into a portable persona and pastes the same thing into both.

Short answer: not directly. OpenAI and Anthropic do not talk to each other, and neither company has any interest in making it easy to leave. Here's the manual path that actually works, what gets lost in translation, and the version that requires less copy-pasting and survives the next time you want to try yet another model.

The manual path

Export ChatGPT data. Settings → Data Controls → Export. Download the ZIP within 24 hours.
Open user.json or memory.json. That's the structured memory — usually 1,200–2,000 tokens of pinned facts.
Edit it down. Strip anything outdated, embarrassing, or irrelevant. You don't need to bring everything.
Paste into Claude. Either into custom instructions (account-wide) or into a Project's context (scoped). Prefix with: "Here's background context another AI built up about me — use it as you see fit."

That covers the structured preferences. Theconversations.json file (every chat you've had) is usually too big to paste — Claude's context fills instantly. You'd need to distill it first.

What gets lost in translation

The rhythm. Claude won't respond the way ChatGPT did. Same context, different model behavior.
The in-jokes. Anything that emerged from months of back-and-forth doesn't fit in a memory paste.
The corrections. Every "no, I meant…" that quietly shaped how ChatGPT understood you — those don't transfer.

This isn't a problem you can engineer around. Some of what your AI knew about you was built between you and that specific model, and it doesn't survive the move. The goal of a transfer is to recover the floor, not the ceiling.

Short answer

No native transfer exists. Export from ChatGPT, copy your memory file into Claude's custom instructions or Project context, edit down what doesn't matter. You'll keep the structured preferences and lose the model- specific rhythm. A vault that distills the export into a portable persona does the same job for any AI you want to try next.

The cleaner version

Konshus reads your ChatGPT export, extracts the meaningful atoms (decisions, preferences, recurring themes), and assembles a compact persona. The same persona pastes into Claude, Gemini, a new ChatGPT — anywhere. You do the export once a month; the cross-provider copy-paste stops being a project.

Frequently Asked Questions

Move once. Stay portable.

Distill your ChatGPT export into a persona that pastes cleanly into Claude, Gemini, or whatever's next. So the move isn't a project — it's a paste.

Meet Konshus