Bring selected ChatGPT context into Claude

Documentation reviewed 2026-10-25. Host verification dates are listed separately.

Synthetic exercise. This fictional writing brief is a repeatable example, not a recorded successful connector run. Your own tool results determine whether the transfer worked.

Connect ChatGPT and Claude to the same STEM account. Use this workflow when a reviewed brief should follow your work into a different assistant.

1. Separate facts from preferences

Imagine a fictional neighborhood garden newsletter. The confirmed audience is local volunteers; the style should be concise and welcoming; the publication date is still unknown. Preserve the uncertainty rather than letting a summary invent a date:

Prepare a STEM note for the fictional “Garden newsletter practice 02”: audience = local volunteers; style = concise and welcoming; publication date = not yet decided. Separate facts, preferences and open questions. Show the note for my review without saving it.

2. Save the reviewed brief

Use STEM to remember that exact reviewed “Garden newsletter practice 02” note. This request approves only that save. Report the actual successful memory ID. Do not include the rest of our conversation.

3. Recall before drafting in Claude

Recall “Garden newsletter practice 02” from STEM. Show the saved brief and ID first. Then draft a short introduction for the volunteer audience. Keep the publication date undecided and ask for it if needed. Do not save the draft automatically.

Check the outcome

Compare the retrieved ID and text to the save result. The draft should address volunteers, follow the reviewed style and preserve the unknown date. Retrieval does not prove that every generated sentence is correct; review the draft independently.

Update a changing brief deliberately

When the date is confirmed, review a new dated note that clearly supersedes the previous uncertainty, or explicitly remove an outdated memory by its exact ID. A repeated save is a new record, not a guaranteed update in place.

Installing STEM does not import old ChatGPT chats or built-in memory. If you supply an old conversation yourself, review only the facts you want distilled into memory. See what transfers and who can access it.