Moving From ChatGPT to Claude: Memory, Projects and GPTs

Moving from ChatGPT to Claude is mostly a rebuild, with one shortcut. The shortcut is memory: Claude has a memory import that takes a list you ask ChatGPT to write out, and turns it into Claude memories in a minute or two. Nothing else transfers on its own. Your chat history stays in ChatGPT, although you can download it as an archive. Custom instructions, projects and their files, custom GPTs, and connected apps each have a Claude equivalent that you set up by hand. The work is less than it sounds if you move what you actually use, in the order below, and leave the rest behind.

This follows Anthropic’s and OpenAI’s documentation as it reads at the end of September 2026. How the two products’ projects compare is in Claude Projects vs ChatGPT Projects, and how Claude’s project memory behaves once you are there is in Claude project memory.

What carries over, and how

• Memory: yes, through Claude’s memory import. You paste in a list; Claude decides what to keep.
• Chat history: no. Anthropic documents importing memory, not past chats. Keep ChatGPT’s data export as your archive.
• Custom instructions: copy them into Claude’s Instructions for Claude by hand.
• Projects: re-create each one in Claude, re-upload the files that matter and paste the project instructions.
• Custom GPTs: rebuild as a Claude project or a skill, depending on what the GPT did.
• Apps and connectors: reconnect each service through Claude’s connectors, signing in again.
• Custom actions and your own MCP servers: add the MCP server to Claude as a custom connector.

Step 1: download your ChatGPT data

Before you change anything, take a copy. OpenAI’s help centre describes the export under Settings, Data controls, Export data: you confirm, and ChatGPT emails a download link when the archive is ready, which can take a while. The link expires, so download it promptly while signed in to the same account. The archive holds your chat history and other account data.

Claude cannot read that archive as history, but it is still useful. Open it, find the few conversations that hold decisions or reference material you still need, and save those as documents. They become knowledge files in the matching Claude project in step 4.

Step 2: move your memory

Anthropic’s article on importing and exporting memory↗ (opens in a new tab) lists memory imports for Free, Pro, Max and Team plans, on the web and in Claude Desktop. The flow has three parts.

1. In ChatGPT, ask it to list every memory it has stored about you and the context it has learned from past conversations, in a single code block, one entry per line with the date where it has one. Claude’s import screen shows a ready-made prompt for this; use it, since it also asks for your instructions, projects, tools and corrections.
2. Read the list before you copy it. Anthropic notes you can change the prompt to leave out sensitive information, and this is the moment to delete anything you do not want moved.
3. In Claude, open Settings, Memory, select Start import, paste the list and choose Add to memory. Then review the new entries in the memory panel.

Set expectations. Anthropic calls the import experimental and says Claude may not always incorporate imported memories. Claude’s memory is designed around work, so personal details unrelated to work may not be kept. Anything you need word for word, such as a client’s name or a house style, should also go into instructions or project knowledge, where you control the wording.

Step 3: rewrite your instructions

ChatGPT’s custom instructions become what Anthropic’s personalisation guide↗ (opens in a new tab) calls Instructions for Claude: account-wide instructions, set in Settings, that apply to all your conversations. Paste them across, then reread them as rules for a different assistant. Lines that existed only to correct a ChatGPT habit may no longer be needed, and a shorter list is easier to check. How to write them is in Claude prompting best practices.

Step 4: re-create your projects

Anthropic’s page on projects↗ (opens in a new tab) describes them as self-contained workspaces with their own chat histories and knowledge bases, available on every plan, with free accounts limited to five. For each ChatGPT project you still use, create a Claude project, paste its project instructions, and upload its files plus the conversations you saved in step 1. Each Claude project keeps its own memory, separate from your other projects, so the import in step 2 is not a substitute for putting project facts in the project.

Step 5: rebuild your custom GPTs

A custom GPT has no direct Claude equivalent; its parts map to different features, as Claude Projects vs Custom GPTs sets out. Decide by what the GPT did:

• Standing knowledge and instructions for ongoing work: a Claude project.
• A repeatable procedure, such as “turn meeting notes into our status format”: a skill.
• Calls to another system: a connector, covered in the next step.

Anthropic’s guide to creating custom skills↗ (opens in a new tab) describes a skill as a folder with a skill file whose YAML frontmatter holds a name of up to 64 characters and a description of up to 200, which Claude uses to decide when the skill applies. You upload it as a ZIP under Customize, Skills, with code execution turned on. If you already ran OpenAI’s migration, your GPT’s instructions are now a ChatGPT skill, and ChatGPT skills use the same shape: a folder with a SKILL.md carrying a name and description. That makes the migrated skill a sensible starting point. Trim the description to fit, check the instructions read correctly for Claude, and test it on familiar requests.

---
name: Status update
description: Turn meeting notes into our weekly status format with decisions, owners and risks. Use when asked for a status update.
---

1. Lead with decisions made, then next steps with an owner for each.
2. List risks last, one line each.
3. Under 250 words. British spelling.
4. If an owner is missing, write "owner needed" rather than guessing.

Step 6: reconnect apps and servers

Every ChatGPT app becomes a Claude connector you connect again, signing in to each service yourself. Claude’s directory is under Customize, Connectors, and on Team and Enterprise plans an owner has to enable a connector before members can connect to it. Where a GPT used custom actions against your own API, or you had a ChatGPT developer-mode app for your own MCP server, add the server as a custom connector.

Anthropic’s guide to custom connectors↗ (opens in a new tab) lists them on every plan, with free users limited to one. Claude connects to the server from Anthropic’s cloud, not from your machine, so the server must be reachable over the public internet. A GPT action that called a REST API needs an MCP server in front of it first, which is the same work OpenAI’s own migration asks for.

Checklist

1. Request the ChatGPT data export and download it when the email arrives.
2. Save the conversations you still need as documents.
3. Run the memory prompt in ChatGPT, edit the list, and import it in Claude under Settings, Memory.
4. Copy custom instructions into Instructions for Claude, and shorten them.
5. Re-create each project you use, with its instructions and files.
6. Rebuild each GPT you use as a project or a skill; test with prompts you saved from the GPT.
7. Reconnect apps as connectors, and add your own MCP servers as custom connectors.
8. Update team guides and bookmarks that point at ChatGPT projects or GPTs.
9. Keep your ChatGPT account until you have used Claude for real work for a couple of weeks.

The work list does not need to move

Much of what people move between assistants is really a list of work: tasks in a project file, reminders in memory, a GPT that filed tickets. Held in an assistant, that list has to be rebuilt every time you switch. On fenbs, a small task board that both assistants reach over MCP, it stays where it is. ChatGPT connects as a developer-mode app where your plan allows it, and Claude as a custom connector; each signs in through your browser, holds your role on the board, and has every change recorded under its own name. Shared rules live once as AI context, which both read with fenbs_get_context. fenbs does not import anything from ChatGPT, so the first move of your list onto the board is by hand; the point is that it is the last.