SecondBrain · Guides · Transfer ChatGPT memory to Claude
How to transfer your ChatGPT memory to Claude
Does Claude have memory like ChatGPT?
Yes. Claude remembers you across conversations, takes standing instructions from your personal preferences, and keeps per topic context inside Projects. ChatGPT does the same thing with saved memories, custom instructions and its own project spaces. Feature for feature they are close enough that the comparison is not really the interesting question.
The interesting part is the boundary. Each memory is welded to the account it lives in. What Claude learns about you on Tuesday is invisible to ChatGPT on Wednesday, and the reverse, forever. That is not a gap either vendor is racing to close: your accumulated context is exactly what keeps you from switching.
Why there is no import button
Memory is stored server side inside each provider's account system. OpenAI does not expose your saved memories as a feed another assistant can subscribe to, and Anthropic does not offer an inbound importer. The two products are not interoperable by design, so every method below is either a manual copy, or a memory that lives somewhere neither of them owns.
The easy way: let ChatGPT do the moving
The part people get stuck on is assuming they have to be the one carrying the data across. You do not. Once ChatGPT can reach a memory that lives outside it, you can simply tell it to write what it knows into that memory, and any other assistant connected to the same memory reads the result. Three steps, about five minutes.
1. Connect SecondBrain to ChatGPT
Download SecondBrain free on iPhone or Android, or open the web app, and turn on Pro: the shared memory is a Pro feature. Then run npx secondbrain-connect once in a terminal and sign in with Apple or Google. ChatGPT can now read and write your SecondBrain memory. There are no API keys anywhere in this.
2. Ask ChatGPT to write its memory into SecondBrain
Open a new conversation and ask:
Save everything you remember about me into SecondBrain, grouped by topic.
ChatGPT reads its own saved memories and writes them into your memory itself. Nothing is copied by hand and nothing passes through a clipboard. Open the app afterwards to see what landed, and delete anything that was not worth keeping: it is easier to prune once here than to carry stale facts into a second assistant.
3. Connect Claude to the same memory
In Claude, open Settings, then Plugins, press "+" and choose Add marketplace. Paste fed3c3sa/secondbrain-shared-memory, install the SecondBrain plugin, press Connect and sign in with the same account. In Claude Code, run claude plugin marketplace add fed3c3sa/secondbrain-shared-memory, then claude plugin install secondbrain@secondbrain --scope user, then npx secondbrain-connect.
That is the transfer. Ask Claude something that only ChatGPT knew and it answers. The direction works both ways, so you can just as easily ask Claude to save what it knows and let ChatGPT pick it up.
The part worth noticing: this does not finish. Both assistants keep reading from and writing to the same memory, so anything either one learns from here is already available to the other. There is no second transfer later.
What actually moves, and what does not
| What you want to move | Can it be transferred | How |
|---|---|---|
| Saved memories | Yes | Ask ChatGPT to write them into SecondBrain, then connect Claude to the same memory |
| Custom instructions | Yes, by hand | They are already written as instructions, so they paste across almost word for word |
| A single conversation | Yes, as a summary | Ask for a handover note rather than pasting the transcript: the summary drops the dead ends |
| Full chat history | Not usefully | An export gives you the archive, but no assistant will read a year of transcripts as memory |
| Uploaded files | Re upload | Put them in a Claude Project so they stay attached to that work |
| Everything, kept current | Yes | One shared memory over MCP that both assistants read and write, so it never needs redoing |
The manual way, if you would rather not connect anything
You can move the memory by hand instead. Ask ChatGPT:
List everything you remember about me, grouped by topic, as plain bullet points.
That gives you a paste-ready block in seconds. To check nothing was missed, Settings, then Personalization, then Manage memories lists the raw saved entries. Cut the throwaway lines, keep the durable ones, then paste the result into Claude's personal preferences, which apply to every conversation. Context belonging to one piece of work is better placed in a Project, in the project knowledge.
It works, and it takes about ten minutes. The defect shows up a week later: it is a photograph. Everything you tell ChatGPT after the copy stays in ChatGPT, Claude drifts further out of date, and eventually you paste again. Do that three or four times and you have spent more effort than connecting a shared memory once would have cost.
Questions people ask
Can Claude import memory from ChatGPT directly?
Not directly. There is no import button and no official migration path, because ChatGPT's memory sits in your OpenAI account and Claude's sits in your Anthropic account, and neither company gives the other read access. You can still move it in about five minutes without copying anything by hand: connect SecondBrain to ChatGPT, ask ChatGPT to save what it remembers about you into SecondBrain, then connect Claude to that same memory and it is already there.
Does Claude have memory like ChatGPT?
Yes, Claude remembers across conversations, and like ChatGPT it also takes standing instructions from your personal preferences and keeps per-topic context inside Projects. The two behave similarly in practice, and the important difference is not the feature but the boundary: each memory is locked to its own account, so what Claude learns stays unknown to ChatGPT. The boundary is the part you can actually do something about. With SecondBrain connected to both, they draw on the same memory, so whichever one you happen to open already knows you.
How do I share a Claude chat with ChatGPT?
For a single conversation, ask Claude to summarize the thread into a handover note and paste that into ChatGPT: the summary carries better than a raw transcript because it drops the dead ends. If you find yourself doing this every week, the copying itself is the problem. With SecondBrain connected to both, you ask Claude to remember the outcome once and ChatGPT picks the thread up on its own, with nothing pasted anywhere.
What is the fastest way to get everything ChatGPT knows about me?
Ask it directly: list everything you remember about me, grouped by topic, as plain bullet points. That gives you a paste-ready block in seconds, and Settings, Personalization, Manage memories shows the raw saved entries if you want to check nothing was left out. Simpler still, if SecondBrain is connected to ChatGPT: ask it to save everything it remembers about you into SecondBrain instead. It writes the memories across itself, and every assistant you connect can read them from then on.
Will the copy stay in sync afterwards?
No, and this is why people end up doing the transfer twice. A pasted list is a snapshot of one day. Everything you tell ChatGPT afterwards stays in ChatGPT, and Claude keeps drifting out of date until you paste again. Only a memory that lives outside both assistants stays current for both, which is exactly what SecondBrain holds: both read from it and write to it, so neither one falls behind.
How does an MCP shared memory keep both assistants current?
The Model Context Protocol is an open standard that lets an assistant call external tools. Point Claude and ChatGPT at the same memory server and both can save to it and read from it during a chat, so a fact learned in one is available in the other on the next question. SecondBrain runs that memory as a managed service you sign into with Apple or Google, with no API keys, and adds voice capture from your phone so the memory fills up without you typing into a chat window at all. It is free to download, with a free plan you can stay on and a Pro subscription that raises the limits and switches on the shared memory. Setup runs about two minutes, and you can start with whichever assistants you already use.
Transfer it once, or never again
One memory for ChatGPT, Claude, Gemini, Cursor and Claude Code. Free to start.