Claude now has a built-in way to import memory from ChatGPT and other AI products. It is useful if you want Claude to know your preferences, background, and recurring work. It doesn't move your ChatGPT conversation threads.
That distinction gets lost in migration guides that promise to move "everything." Anthropic's documented flow asks ChatGPT to produce a block of remembered information. Claude reads that block, extracts key details, and stores them as memory entries. Your active thread, its corrections, its files, and its unfinished work need a different handoff.
Claude's importer moves a profile of you. A handoff brief moves the state of one piece of work.
If you are switching products, use Claude's memory import for the broad background. Use a focused handoff brief for the conversation you want to continue. Keep the source files somewhere you control. Those three moves solve three different problems.
Three different jobs get called a "transfer"
Migration guides often mix memory import, full account exports, project uploads, and one-chat handoffs under promises to move "everything." They solve different problems.
An account export keeps a copy of old chats. A memory import teaches Claude recurring information about you. A one-chat handoff carries the current state of work. Calling all three a transfer makes people expect a memory importer to recreate threads or mistake an archive for something Claude can continue.
What Claude's ChatGPT memory import actually does
Anthropic's memory import instructions (opens in new tab) describe a three-step flow:
- Claude gives you a prompt to run in the AI product you are leaving.
- That product returns a text block containing stored memories and learned context.
- You paste the result into Claude, which extracts key information into its own memory.
The prompt asks for response preferences, personal details, projects, goals, tools, corrections, and recurring topics. This is account-level context. It helps Claude learn that you prefer short answers, work in Swift, live in Nevada, or keep returning to the same product project.
It doesn't recreate the conversation where you debated three designs, rejected two, corrected a bad assumption, and left the third half-finished. Claude starts with memory entries, not the old thread.
OpenAI's documentation draws the same boundary from the ChatGPT side. ChatGPT can use past chats, saved memories, custom instructions, files, and connected apps as separate sources. Its memory guide (opens in new tab) says memory doesn't retain every detail from every conversation. It also says saved memories are stored separately from chat history.
"Context" covers four different containers. Treating them as one is how migration instructions become confusing.
Memory tells the next AI who you are. A handoff tells it what happened in this piece of work. Source material tells it what to trust.
Choose the transfer method based on what you need
Memory import, a handoff brief, and a data export each have a specific job.
Use Claude's memory import when you are changing your default AI product and want familiar preferences to follow you. It is the closest thing to a profile migration.
Use a handoff brief when you want to continue one active ChatGPT conversation in Claude. The brief should carry goals, decisions, constraints, source material, open questions, and the next action.
Use the ChatGPT data export when you want an account archive. OpenAI says the downloaded ZIP can include chat history and other account data. Keep that ZIP for backup. It is a noisy place to start one specific task, and Anthropic's documented memory importer doesn't ask for it.
Attach or link project files separately. A memory system is a poor home for exact requirements, datasets, contracts, source code, or the latest draft. Give the next model the files it needs rather than hoping an old chat summarized them correctly.
How to move your ChatGPT memory to Claude
Claude's current web and desktop flow starts in Settings > Memory > Start import. Anthropic says memory import is available on its Free, Pro, Max, and Team plans. Some accounts with an older memory interface can still show the control under Settings > Capabilities.
Claude supplies the export prompt. Run it in ChatGPT, read the result, remove anything you do not want to carry over, and paste the cleaned text into Claude's import box. Claude turns the text into individual memory entries that you can review.
Review the first import instead of treating it as a perfect copy. Anthropic says Claude's memory focuses on work-related topics and can skip personal details that are unrelated to work. The importer is also marked experimental. After importing, ask Claude what it learned and correct the entries that are wrong or stale.
This part should take a few minutes. It is the right tool for preferences and recurring background. Stop there if that is all you need.
How to continue one ChatGPT conversation in Claude
Run the following prompt inside the ChatGPT conversation you want to move. A blank chat cannot see the corrections and decisions inside the thread you care about.
Prepare a handoff for another AI assistant that will continue this work.
Return concise Markdown with these sections:
1. Goal
2. Current state
3. Decisions already made, with the reason for each
4. Constraints and preferences that still apply
5. Source material: files, links, data, and exact facts to trust
6. Open questions or uncertain facts
7. Next action
Use the latest correction when this conversation contains conflicting information.
Do not revive rejected ideas or propose new work. Do not invent missing facts.
Keep names, numbers, dates, quotations, and URLs exact.
This produces a working brief instead of a transcript summary. A normal summary tends to preserve the topic and lose the decisions. The next model then repeats work you already did or casually reopens a path you already rejected.
Seven sections are enough to carry the work without dumping the whole transcript into the new chat.
Read the brief before you paste it into Claude. Check the parts that are expensive to get wrong:
- names, numbers, dates, and links
- decisions that changed later in the conversation
- old assumptions that were corrected
- source files the summary mentions but does not contain
- private details the next product doesn't need
Then paste the brief into a new Claude chat and attach the source files. Before asking Claude to continue, use one short check:
Restate the goal, decisions, constraints, missing sources, and next action.
Do not start the work yet.
That reply tells you whether the handoff landed. Fix any misunderstanding while the state is still compact. Once the new chat starts producing work, a bad premise becomes harder to spot.
Keep project truth outside AI memory
The durable fix is a portable working record. In a February 2026 Hacker News discussion (opens in new tab), one user described the same problem across Claude, Cursor, ChatGPT, and Perplexity: each tool lacked the context from the others. Replies recommended shared Markdown files for progress, requirements, decisions, and reasoning. One commenter kept full notes plus shorter summaries for different parts of the work.
The Markdown approach works because vendor memory can remember how you like to work while your files record the work itself.
A portable workflow survives a new model, a new product, or a memory reset.
My rule is straightforward:
- Source files hold facts. Requirements, data, code, drafts, and links live outside the chat.
- A decision log holds choices. Record the options you rejected and why.
- A handoff brief holds current state. It is small enough to inspect and move.
- AI memory holds preferences. Tone, recurring background, and stable working habits belong there.
This structure also makes failures cheaper. If a memory import misses one preference, you can add it. If a handoff misses a decision, you can check the log. If a model mangles a number, you can open the source file.
A one-minute ChatGPT-to-Claude checklist
Before leaving ChatGPT:
- Use Claude's importer for broad memory and preferences.
- Generate a handoff brief inside each active thread you want to continue.
- Review the brief for corrected facts, final decisions, and missing sources.
- Save project files and a short decision log somewhere you control.
- Paste the brief into Claude, attach the sources, and ask Claude to restate the state before it continues.
You do not need to move every old chat to keep working. Move the profile, the current state, and the sources through the containers designed for each one.
Stop moving your work every time the best model changes
Why are you moving from ChatGPT to Claude, or from Claude back to ChatGPT? One product is better for what you need today. That will change. A new model will ship, pricing will move, or another provider will handle your work better.
I'd be remiss if I didn't mention Cumbersome, the iPhone and Mac product I built for exactly this problem. It connects directly to OpenAI, Anthropic, Google AI Studio, Z.ai, gateways, and compatible APIs. Pick the provider and model you want, then switch inside the same conversation without packing up the thread and moving to another chat product. You can also edit the conversation context before switching, so a dead end does not follow you to the next model.
Cumbersome is free. You use your own API keys, pay the providers directly, and keep model choice separate from the chat client. The next time the best option changes, you can change the model instead of changing products.
Bless up! 🙏✨