Move Your ChatGPT Memory to Claude Without Losing the Important Context

Claude now has an official memory-import flow. The transfer is still a copy-and-paste process: Claude gives you an extraction prompt, ChatGPT returns a block of remembered context, you review that block, and Claude turns the approved text into its own memory entries.

This is useful, but it is not an account connection. It does not recreate your ChatGPT conversations, copy uploaded files, rebuild Projects or custom GPTs, or guarantee that every personal detail will be retained. Think of it as a carefully reviewed handoff note between assistants.

The shortest safe path is:

1. Open Claude's memory import and copy the prompt it provides.
2. Paste that prompt into ChatGPT and let it produce your remembered context.
3. Remove secrets, stale facts, and anything you do not want to send to Anthropic.
4. Paste the cleaned text into Claude, import it, and verify the resulting memory.

First, Know What You Are Moving

“Memory” can refer to several different things. Only some of them belong in this transfer.

OpenAI's current Memory FAQ describes ChatGPT memory as an evolving synthesis of information from chats and other sources. It also warns that the visible memory summary may not show everything ChatGPT remembers. That is why the transfer should be treated as a useful snapshot, not a complete database export.

Use the Import Entry Your Claude Account Shows

Anthropic is rolling out a new memory experience, so two valid interface paths currently exist.

• If Claude shows Settings > Memory, choose Start import there.
• If memory still appears under Settings > Capabilities, find the Memory section and choose Start import. Some legacy accounts may also see an Import memory to Claude card on the home screen.

Anthropic's memory import and export guide currently says imports are available on web and Claude Desktop for Free, Pro, Max, and Team plans. Its stated availability line does not include Enterprise. If you use a managed workspace, follow your administrator's policy rather than moving work context through a personal account.

Do not worry if a tutorial shows a different menu from yours. Settings > Memory and Settings > Capabilities represent different rollout states, not two different migration methods.

Extract Context from ChatGPT

Inside Claude's import flow, copy the extraction prompt supplied by Anthropic. Paste it into a normal ChatGPT conversation. The prompt asks ChatGPT to collect remembered instructions, preferences, personal details, recurring projects, tools, corrections, and other durable context into one copyable result.

Use the provider-supplied prompt instead of an old prompt copied from a video or social post. Anthropic can update its import contract as the feature changes.

Before trusting the result, compare it with the memory controls available under ChatGPT Settings > Personalization > Memory. Look for three kinds of problems:

• Missing context: a current project, preferred output format, or recurring constraint is absent.
• Stale context: an old role, location, tool, or goal is no longer true.
• Contradictory context: two instructions describe different versions of the same preference.

You do not need to make the output exhaustive. A smaller, current handoff is usually more useful than a long archive full of obsolete detail.

Review the Text Before It Leaves ChatGPT

Copying the extraction result into Claude sends that information to a different provider. Treat the text like a document you are about to share outside its original system.

Remove or generalize:

• passwords, API keys, access tokens, recovery codes, and private links;
• client names, customer records, internal URLs, unpublished financials, or contract details;
• health, legal, identity, payment, or family information that Claude does not need;
• old preferences that would produce the wrong behavior;
• facts about other people who did not consent to the transfer.

Anthropic explicitly recommends customizing the extraction to exclude sensitive information. A practical rule is simple: if a detail would not belong in a reusable onboarding note for a new collaborator, it probably does not belong in imported memory.

For company work, check policy before pasting anything. A Team plan being technically supported does not mean every employer permits cross-provider transfer of internal context.

Import and Verify What Claude Retained

Return to Claude's import window, paste the cleaned block, and choose Add to memory. Claude will extract key information into memory entries. In the newer experience, review those entries in the memory panel. Legacy accounts may show imported edits under a management view.

Anthropic labels the feature experimental and says imports may not always be incorporated successfully. It also says Claude memory is designed to prioritize work-related context, so unrelated personal details may be left out. On the legacy experience, updates can take up to 24 hours to appear.

Use a short verification conversation instead of assuming success. Ask Claude to explain what it learned, then test a few observable behaviors:

• What response format do I prefer?
• Which current project am I working on?
• Which tools or languages should you assume I use?
• What should you avoid doing in your answers?
• Which imported facts are uncertain or contradictory?

Correct errors in Claude's memory controls or ask Claude to change/remove the relevant entry. Do not repeatedly import the same block to fix one omission; duplicates can make the resulting context harder to manage.

Keep a Full ChatGPT Export as a Separate Backup

If your goal includes preserving old conversations, request a ChatGPT data export in addition to the memory handoff. OpenAI's data-export guide says the downloaded ZIP can include chat history and other relevant account data.

That ZIP is an archive, not the documented input for Claude's built-in memory importer. Store it securely and use it to retrieve an old conversation when needed; do not assume Claude will reconstruct thousands of chats from it.

OpenAI currently makes self-service settings export available for Free, Go, Plus, Pro, and eligible Edu workspaces. Business, Enterprise, and Healthcare workspaces use organization-managed access instead. OpenAI says an export can take up to seven days to arrive and its download link expires after 24 hours, so download and inspect it before closing or merging an account.

If Import Memory Is Missing or Fails

If the import button is absent, check Claude on the web or the current Claude Desktop app and look under both memory paths. Confirm that the account is on a plan named in Anthropic's current availability guidance. Managed workspace policy can also affect what you see.

If the button exists but the result is weak:

1. Review the extracted text and remove contradictions or excessive history.
2. Group only the durable facts that will change future answers.
3. Import once, wait for the applicable experience to update, and inspect the entries.
4. Add an essential missing item manually through Claude's memory controls.

If memory import remains unavailable, create a short local Markdown handoff containing your current role, active projects, working preferences, important constraints, and information Claude should not assume. Paste only the relevant part when starting a new Claude project or conversation. This fallback requires more upkeep, but it keeps you in control of the exact context being shared.

A Transfer Is Complete When the Behavior Is Right

The goal is not to maximize the number of imported entries. The transfer is useful when Claude can apply the few facts that actually change your work: the right context, the right response style, the right constraints, and the right current goals.

Keep the ChatGPT archive if you need historical lookup. Keep the reviewed handoff if you want a portable baseline. Keep Claude's memory small enough to inspect. These three assets serve different jobs, and separating them is what makes the move dependable.