You can continue a Claude project in ChatGPT, but RelayLink does not convert or import the Claude project. It carries a selected briefing as text only after you review and explicitly confirm it. Chat state, model memory, files, repositories, branches, browser tabs, and tool configuration remain separate.
Move the durable project artifacts through their own systems, and use the briefing for the decisions and constraints that give those artifacts meaning. ChatGPT rarely needs the whole path you took through the first session.
Identify the durable project
Write an inventory of the objects that remain true when the Claude window is closed:
- the repository and exact branch or commit;
- the current document and version;
- requirements, acceptance criteria, and non-goals;
- approved decisions;
- source material and data;
- test commands and observed outcomes;
- deadlines and external dependencies.
Those items define the project state. A conversation may explain them, but it is not their storage system. If a requirement exists only as a correction buried halfway through a chat, promote it into the briefing or the project's own documentation.
This distinction also reveals missing artifacts. If Claude produced a plan but no versioned document records it, decide whether to save that plan before switching.
Extract only the conversation state that matters
The working conversation contains useful history, but usefulness is selective. Capture:
- constraints that changed the direction;
- approaches tried and the evidence that ruled them out;
- assumptions Claude inferred, clearly labeled as inferences;
- terms or shorthand that need expansion;
- unresolved questions and what would answer them.
Leave out repeated drafts, conversational repair, and abandoned wording unless the exact sequence is relevant evidence. A transcript tells ChatGPT what happened in time. A project brief tells it what remains in force.
The focused Claude-to-ChatGPT handoff is enough when you only want a review or one following phase. A full move needs the artifact inventory as well.
Define the destination before sending
Decide what ChatGPT should do first. Good starting assignments verify continuity:
- compare the brief against the named artifact and list discrepancies;
- run the specified test and report whether the baseline matches;
- restate the fixed constraints before proposing changes;
- review one open decision without reopening accepted ones.
Avoid “continue the project.” It gives ChatGPT no reliable way to know which work is complete or what authority it has to revise. A bounded task creates a point where you can judge whether the migration worked.
Use RelayLink for the reviewed handoff
Connect Claude and ChatGPT to the same RelayLink account. From Claude, compose a package to your own account email containing the selected state of play. Review the assumptions, constraints, artifact references, and first assignment. Then explicitly confirm it.
The shorthand opens a draft and stops. It never sends by itself. Once confirmed, the package appears unread in the account inbox until ChatGPT pulls it. Because this is a self-handoff, there is no contact request, second account, email notification, or magic link. Your account page provides the inbox fallback.
RelayLink does not move the project space. It carries text you approved from one connected assistant to the correspondence inbox both can use.
Give ChatGPT artifact access separately
Arrange access through the repository host, document system, data store, or other system that owns the artifact. Include paths, commit identifiers, document versions, and test targets in the briefing so ChatGPT can orient after access exists.
A file path is not a file upload. A repository URL is not a clone. A link is not permission to fetch. RelayLink neither follows those references nor verifies what they contain.
If access cannot be granted, adapt the assignment to material you are permitted to provide. Do not paste confidential files into a second provider merely to make the migration feel complete.
Recheck data rules at the provider boundary
Moving from Claude to ChatGPT may change who processes the project data. Check your employer's rules, client terms, and provider controls before the handoff. Review lets you correct and choose the package text; it does not prove the text is accurate or authorize a prohibited destination.
When privacy is the reason for avoiding the full conversation, switching AI tools without copying the whole chat shows how to minimize the handoff while preserving the decisions that still matter.