How to Import ChatGPT Memory into Claude: A Step-by-Step Guide

Yes, you can import ChatGPT memory into Claude, and there are two ways to do it. The first is Claude's built-in memory import, which Anthropic ships under Settings > Memory. The second is a manual, prompt-based transfer that works with any assistant. This guide covers how to import ChatGPT memory into Claude step by step, what carries over, what does not, and how to clean up afterwards. A short final section covers the developer version of the same problem: moving an application's memory layer between model providers without re-ingesting everything.

Can you import ChatGPT memory into Claude?

Yes. Anthropic's memory import flow lets you paste a text export from another AI provider into Claude, which then extracts it into memory entries. It is available on Free, Pro, Max, Team and Enterprise plans on the web and Claude Desktop, according to Anthropic. There is no direct ChatGPT-to-Claude connector; the transfer runs through text you copy yourself.

Two things set expectations. First, both products store memory differently. ChatGPT keeps a mix of saved memories, a memory summary and context derived from chat history, as described in the OpenAI memory FAQ. Claude saves memory as a set of individual topics and focuses on work-related context. Second, Anthropic labels memory imports "experimental and still in active development" and warns that Claude may not always incorporate every imported item. Treat the import as a strong head start, not a byte-for-byte copy.

If you want background on either side before you begin, see the guides on Claude memory and ChatGPT memory.

Step 1: Export what ChatGPT remembers about you

ChatGPT does not have an "export memory" button. You get the content out in one of two ways: read it from the memory settings screen, or ask ChatGPT to write it out.

Option A: The Manage memory screen

Open Settings, select Personalization, then Memory. Depending on your plan you will see controls such as Memory summary, Manage or Saved memories. The summary is a high-level view and, according to OpenAI, does not necessarily include every detail ChatGPT can reference. This screen is useful for checking what you are about to move, but copying from it line by line is slow.

Option B: The summarize-memory prompt

The faster route is to ask ChatGPT to write everything out in one block. Anthropic publishes a prompt written for exactly this purpose, and it is the same text Claude shows inside its import screen:

I'm moving to another service and need to export my data. List every memory you have stored about me, as well as any context you've learned about me from past conversations. Output everything in a single code block so I can easily copy it. Format each entry as: [date saved, if available] - memory content. Make sure to cover all of the following — preserve my words verbatim where possible: Instructions I've given you about how to respond (tone, format, style, 'always do X', 'never do Y'). Personal details: name, location, job, family, interests. Projects, goals, and recurring topics. Tools, languages, and frameworks I use. Preferences and corrections I've made to your behavior. Any other stored context not covered above. Do not summarize, group, or omit any entries. After the code block, confirm whether that is the complete set or if any remain.

Run it in a regular chat, not a temporary one, so ChatGPT has access to your memory. If the answer ends with "some entries remain," reply with "continue" until it confirms the set is complete. Then copy the code block into a local text or Markdown file. Keep this file: it is your backup, and the same text works for any other assistant later.

Before you paste anywhere, read the export and delete lines you do not want to carry over. The prompt pulls in personal details alongside work context, and this is the one moment where you control exactly what moves.

Custom Instructions and chat history are separate

Two things the prompt will not fully capture. Custom Instructions are direct guidance you typed into ChatGPT's settings about how it should respond; OpenAI treats them as distinct from memory. Open Settings > Personalization > Custom instructions and copy that text manually if you want it. Chat history is also separate. If you want your actual conversations, request a full data export under Settings > Data controls > Export data. According to OpenAI's export instructions, the ZIP arrives by email within up to 7 days and the download link expires after 24 hours. That archive is for your records; Claude's import does not accept it.

Step 2: How to import ChatGPT memory into Claude

With the exported text in hand, the Claude side takes about a minute.

The official import flow

1. Open Claude on the web or Claude Desktop and go to Settings > Memory.
2. Select "Start import." The screen shows the export prompt from Step 1 and a text box.
3. Paste the text you copied from ChatGPT into the box.
4. Click "Add to memory."

Claude extracts key information from the paste and stores it as individual memory entries. You can review them in the memory panel. When the import finishes, a "See what Claude learned about you" button opens a new chat with the prompt "I updated my memory. What did you learn about me?", which is the quickest way to confirm what actually landed.

If you see Memory under Settings > Capabilities rather than its own Settings > Memory entry, you are on the legacy memory experience that a small number of Team and Enterprise organizations still use. The same "Start import" button sits in the Memory section there, imported items appear as memory edits under "Manage edits," and the updated memory can take up to 24 hours to show.

What Claude does with the pasted text

Claude's memory is designed to focus on work-related context: your role, projects, people and places in your work, communication and coding preferences. Anthropic states that Claude may not retain imported personal details unrelated to work. By default it also skips sensitive topics such as health, religion, politics or ethnicity unless you turn on "Include sensitive topics in memory" in Settings > Memory. Some categories are never saved regardless of settings, including government ID numbers, financial account numbers, criminal history and immigration status.

If a specific line from your export did not make it in, add it directly. In Settings > Memory, use the "Tell Claude what to change or remove" box, or simply tell Claude in a chat to remember it.

The prompt-based alternative

If the import button is not visible on your account, the manual method works in any Claude chat. Start a new conversation and paste the export with a framing instruction:

Below is what my previous AI assistant remembered about me. Please save the relevant work context, preferences and projects to your memory. Ask me before saving anything you consider personal or sensitive.

[paste the exported block here]

Claude saves memory as you chat, and you can tell it to "remember this" to save something directly. The result is the same kind of topic entries the import flow creates, with less automation.

Step 3: Review, edit and clean up Claude's memory

An import is a good moment to audit what Claude now holds. Everything lives in Settings > Memory under Topics. Select any topic to read it, use the edit icon to change it, or select "Delete" to remove it. Edits apply to every conversation from then on. You can also make changes from a chat: tell Claude what to remember, change or forget, and the update takes effect in your next conversation.

Three controls matter for cleanup:

• Pause memory keeps existing entries but stops Claude from using or adding to them. Chats held while paused are not added later.
• Reset memory permanently deletes all memories, including project memories. It cannot be undone, so export first if you might want the content back.
• Incognito chat (the ghost icon on a new chat outside a project) is not saved to history or memory, and is available on every plan. If you want a one-off chat that stays in history but does not touch memory, turn off "Memory" in the "+" menu before your first message.

One detail worth knowing from Anthropic's memory documentation: deleting a conversation does not remove memory entries generated from it. If something imported is wrong, delete the topic itself. Memory data is also included in Claude's data exports, so you can keep an archive of what Claude holds at any point.

To take a copy out of Claude later, ask it in a chat: "Write out your memories of me verbatim, exactly as they appear in your memory." Save the answer to a local file. That is Anthropic's documented export path and it is what you would paste into the next assistant.

What transfers and what does not

The practical summary: the import moves the who you are and how you work layer. It does not move conversations, files, or anything that ChatGPT inferred but did not write into the export.

Moving memory the other way or to other assistants

The export prompt is vendor-neutral. Running it in Claude produces the same kind of dated list, and Anthropic documents the shorter version above for pulling Claude's memory out verbatim. Paste that into ChatGPT and ask it to save the entries as memories, or paste it into Gemini, Copilot or any other assistant that offers a memory or personalization setting. Check each product's own memory controls afterwards to confirm what it kept; every vendor applies its own filters about what counts as worth remembering.

Keep the exported file. As long as you hold the plain-text version, you are not locked into any one assistant's memory store, and a quarterly re-export is a reasonable habit if you rely on memory heavily.

For developers: moving an application's memory layer between providers

The consumer problem above has an exact analogue in application code. If your agent's long-term memory lives inside one vendor's hosted feature, switching models means losing it or running a lossy copy-paste. The fix is the same in principle: keep memory in a store you control, with a text or structured export, and connect models to it rather than the other way around.

Three design choices make that work. First, store memory outside the model vendor: a database, a knowledge graph or a dedicated memory service that any model can query. Second, expose it through a protocol several clients already speak. MCP memory servers do this: the same remember and recall tools serve Claude Desktop, Claude Code, Cursor, Codex, and any ChatGPT-compatible agent runtime that supports MCP. Third, insist on an export format. If you cannot dump the whole store to files and reload it elsewhere, you have recreated the lock-in one level down.

Anthropic's own API design points the same way. The Claude API memory tool runs client-side: Claude requests reads and writes, and your application executes them against storage you choose. That makes the application, not the model, the owner of memory, which is the property you want when the model underneath changes.

How Cognee keeps memory portable across models

Cognee is an open-source memory engine (Apache 2.0, github.com/topoteretes/cognee) that treats portability as a design goal rather than an afterthought. It stores memory as a knowledge graph plus vector index in backends you run, including a single Postgres with pgvector, Neo4j, Kuzu, FalkorDB and LanceDB, and exposes four verbs: remember, recall, improve and forget.

Three features map directly onto the migration problem in this article:

• MCP server. cognee/cognee-mcp exposes remember, recall and forget over stdio, Streamable HTTP and SSE, so Claude Desktop, Claude Code, Cursor, Codex, Cline and Continue all read and write one shared store. Switching the model does not switch the memory.
• Import adapters. The Mem0Source adapter loads an existing Mem0 export with one call, and the same pattern covers Letta, Zep and Graphiti through the COGX exchange format. A plain-text export like the ChatGPT block above can be passed to remember directly.
• COGX export. Any dataset can be written out to the open COGX archive format or GraphML, which is the developer equivalent of keeping the text file from Step 1.

import cognee
from cognee.migration import Mem0Source

# Bring an existing memory store across, re-deriving the graph on the way in
await cognee.remember(Mem0Source("mem0_export.json", mode="re-derive"))

# Or ingest a plain-text export such as the ChatGPT memory block
await cognee.remember(open("chatgpt_memory_export.md").read())

Cognee Cloud is free for the first 1M tokens and one workspace; the Standard plan is $1.00 per 1M tokens processed plus $5 per additional workspace as of October 2026, and self-hosting is free. Details are on the pricing page and in the docs.

Next steps

For a personal account, the whole transfer is three actions: run the export prompt in ChatGPT, paste the result into Settings > Memory > Start import in Claude, then audit Topics and delete what should not be there. Keep the exported file so the next move is just as quick. If you build agents, read the portable AI memory guide and try the Cognee MCP server so the memory layer stays yours no matter which model sits on top.

FAQ

Can Claude import memory from ChatGPT?+

Yes. Claude has a built-in memory import under Settings > Memory > "Start import." You ask ChatGPT to list everything it remembers about you using the export prompt Anthropic provides, paste the result into Claude's import box and click "Add to memory." Claude extracts the text into individual memory entries that you can review in the memory panel. Anthropic labels the feature experimental, so check the results with the "See what Claude learned about you" button and add anything that was skipped.

Does importing memory copy my ChatGPT chat history?+

No. The import moves only the text you paste, which is ChatGPT's description of what it remembers about you. Your actual conversations stay in ChatGPT. If you want a copy of them, request a data export under Settings > Data controls in ChatGPT; OpenAI sends a ZIP by email within up to 7 days. That archive is for your records and cannot be loaded into Claude's memory. Claude's past-chat search only covers conversations that happened inside Claude.

Is Claude memory import available on the free plan?+

Yes, according to Anthropic's help center, memory imports are available on Free, Pro, Max, Team and Enterprise plans on the web and Claude Desktop. On Team and Enterprise plans an owner must have memory enabled for the organization first, and each member turns it on individually. If you do not see "Start import" under Settings > Memory, check whether memory itself is on, then consult the Anthropic help center article on importing and exporting memory, since availability can change during the experimental period.

How do I delete imported memories in Claude?+

Go to Settings > Memory, open the topic you want to remove and select "Delete." You can also tell Claude in any chat to forget a specific detail, and the change applies to your next conversation. To wipe everything, use "Reset memory," which permanently deletes all memories including project memories and cannot be undone. Deleting a conversation does not delete memory entries derived from it, so always remove the memory topic itself rather than the chat.

Can I transfer Claude memory back to ChatGPT?+

Yes. Ask Claude in a chat: "Write out your memories of me verbatim, exactly as they appear in your memory." Save the answer to a local file, then paste it into ChatGPT and ask it to remember the entries. ChatGPT applies its own rules about what it keeps, so confirm the result under Settings > Personalization > Memory. The same file works for Gemini, Copilot or any other assistant with a memory setting, which is why keeping a plain-text copy is the simplest protection against lock-in.

Is it safe to paste my ChatGPT memory into Claude?+

It is as safe as anything else you type into Claude, with one extra step worth taking: read the export before pasting and remove lines you do not want stored. The export prompt deliberately pulls in personal details alongside work context. Claude stores memory encrypted at rest, skips sensitive topics such as health or beliefs by default, and never saves government ID numbers, financial account numbers, criminal history or immigration status. On Team and Enterprise plans, memory follows your organization's retention policy and is included in admin data exports.