How to move a ChatGPT project to Claude — without re-explaining everything
You've been working in ChatGPT for weeks — the goal, the constraints, the voice, the decisions already made all live in that chat history and in whatever project instructions and memory ChatGPT holds for you. Now you want Claude on it. There's no import button between vendors, and pasting a whole conversation into a fresh chat mostly transfers noise. What actually transfers a project is a brief. Here's how to build one once and reuse it every time you switch — in either direction.
Why copy-pasting the chat doesn't work
- Chat history is mostly noise. Dead ends, retries, and small talk outnumber the decisions. A fresh model has to dig for what matters — and often anchors on the wrong parts.
- It doesn't fit. Long conversations blow past what you can reasonably paste, so you trim arbitrarily and lose the standing rules you set in week one.
- ChatGPT's memory stays behind. Whatever it has learned about your project serves ChatGPT chats. That's true of every vendor's memory — it briefs its own model, not the competition's.
- You'll do it again next week. One heroic paste doesn't survive the next model switch. The re-explaining tax comes back every time.
The manual method (works, but you'll repeat it)
If you just need to switch once, this is legitimate:
- Ask ChatGPT to summarize the project: goal, decisions made, constraints, current state, next step.
- Edit the summary by hand — models flatter their own transcripts, so cut what you know is wrong or stale.
- Paste the edited brief at the top of a fresh Claude chat and state the next task below it.
The problem isn't quality — a hand-edited brief works fine. The problem is that it's a 15-minute ritual you'll repeat for every switch, and the brief itself lives nowhere. Next month it's stale and you start over.
The portable-record method
Promptmkr turns that one-off brief into a standing asset:
- Create a project and write the brief into it once: goal, key context, standing rules, what not to do, current state. This is the same distillation as the manual method — done a single time, in a record you own.
- Prime Claude in one click with the free Chrome Companion (or copy-paste the primer from the web app). The primer is generated from the record and formatted for the target model — structured XML for Claude, Markdown for ChatGPT — so the new chat starts oriented instead of cold.
- Log where you stopped. When the session ends, update the project's current state and next move. That takes a minute and replaces the 15-minute ritual.
- Switch back whenever. Tomorrow ChatGPT gets the same one-click briefing, updated to today's state. The record is neutral — it doesn't care which model wins the week.
The free plan includes one project and three one-click primers — a one-time allowance, not monthly, but enough to run a real switch in both directions before paying anything.
What belongs in the brief (and what doesn't)
{{variables}}. Every prompt you ever tried. Save the ones that earned reuse. What not to expect
- There is no automatic import of a ChatGPT conversation or its memory. The distillation step is real work — you do it once instead of every switch.
- Promptmkr is not a chat app. Claude does the Claude work; Promptmkr hands it the briefing.
- Priming doesn't transfer files or uploaded documents — it transfers the working brief: goal, context, rules, state.
Do the distillation once
Write the brief into a project today; every model switch after that is one click.
Comparing tools instead? See why Promptmkr is a Claude Projects alternative that works in every model.