How to Move From ChatGPT to Claude Without Losing Your Context

You already made the call. Somewhere between the third rewrite of a prompt that used to work and the model update that changed the answers, you decided to leave. So the interesting question is not whether to switch. It is what actually survives when you transfer ChatGPT to Claude, and what quietly does not.

Short version: your transcripts move fine. Your memory moves partially. The invisible layer, the thing that made your old assistant feel like it knew you after two years of daily use, moves worst of all, because most of it was never a file you could copy in the first place.

Here is the migration I would run, in order, and the part nobody puts in the tutorials: how to stop paying this tax again the next time you change tools.

Three different things get called "context"

Before you touch a settings page, separate them:

• Chat history. The transcripts. Fully portable, and the least useful of the three.
• Memory. What the assistant inferred about you over time. Semi-portable, and lossy in ways the interface does not advertise.
• Working setup. Custom instructions, project files, connected apps, custom GPTs. Not portable at all. You rebuild it by hand, and that is fine, because it takes an afternoon.

Most migration guides stop at the first bullet. That is why people arrive in a new assistant with a giant archive and the distinct feeling of having lost something.

Step 1: export before you cancel anything

In ChatGPT: profile menu, then Settings, then Data controls, then Export data, then Confirm export. Two details from OpenAI's own help page decide your timeline: exports "can take up to 7 days to arrive", and the download link "expires 24 hours after you receive it". The same page says an export cannot restore chats you have already deleted, and that self-service export is not available inside Business, Enterprise or Healthcare workspaces, where you have to go through the workspace owner.

So request it now, while you are still a paying customer, and do not schedule the cancellation for the same week. A seven-day fetch against a 24-hour link is the one part of this migration that can genuinely fail while you are asleep.

What lands is a ZIP with your conversations as JSON. OpenAI's guide to reusing an export tells you to extract the ZIP and find conversations.json, noting that "larger exports may contain numbered conversation JSON files instead". Treat it as an archive, not as an input. Pasting two years of transcripts into a fresh assistant produces noise, not context.

Step 2: transfer ChatGPT to Claude memory, and know what gets dropped

Anthropic actually built a path for this. Claude's memory import page hands you a prompt to paste into your current assistant, which asks it to dump everything it has stored about you into one code block. You then paste that output into Claude at Settings, Memory, Start import. Per Anthropic's help article, the import flow is available on "Free, Pro, Max, and Team plans on the web and Claude Desktop".

The same article carries two caveats, and they are the whole ballgame. First, memory imports are "experimental and still in active development, and at this stage, Claude may not always successfully incorporate imported memories". Second, Claude's memory is deliberately biased toward work, so it "may not retain imported personal details unrelated to work".

The source side leaks too. OpenAI's Memory FAQ says the memory summary "will not include everything that ChatGPT remembers based on your chats", and when it asks itself whether the summary includes all your memories, the answer is "Not necessarily". You are exporting a summary of a synthesis. Expect gaps.

Which is why I would not stop at the automated prompt. After you run it, spend twenty minutes writing your own context brief by hand: what you do, who you serve, the tools you actually use, your writing rules, and the three decisions you never want to argue about again. That document beats any automated dump, it is yours forever, and it takes less time than waiting for the export email.

Step 3: rebuild the working setup as projects, not chats

This is the step that determines whether the new assistant feels smart in week two.

Claude Projects are self-contained workspaces with their own knowledge base and their own instructions. They are available on free accounts too, capped at five projects, and on paid plans project knowledge scales through retrieval, which Anthropic describes as expanding capacity "by up to 10x" when your uploads approach the context limit. Claude also keeps memory scoped per workspace: each project has its own separate memory space and its own summary.

So the mapping is simple. Custom instructions become project instructions. The reference files you used to re-upload every session become project knowledge. Each recurring job gets its own project rather than one endless thread.

What moves, what does not

What the overlap month costs

Run both for one billing cycle. ChatGPT Plus is listed at $20 per month, and Claude Pro at $20 if billed monthly, or $17 per month on the annual plan billed as $200 up front. Paying both monthly rates for a single overlap month is $40, and it buys you the ability to go back and check something rather than discovering the gap after the archive link expired.

If you would rather script the move than paste anything, summarise your export with the API instead of feeding raw transcripts to a chat window. Anthropic's platform pricing page lists Claude Sonnet 5 at $2 per million input tokens and $10 per million output tokens, and Claude Opus 5 at $5 and $25. Count the tokens in your own export before you decide which one to point at it; the size of a two-year archive varies wildly depending on whether you used it for code.

The migration you should not have to run twice

Here is the uncomfortable part. This migration hurts because your context lived inside one vendor's product. Every assistant is incentivised to keep it that way. Do the whole exercise again in eighteen months and you will lose the same afternoon.

The durable fix is to keep the memory and the connections outside the model, so switching models becomes a settings change rather than a moving day.

That is the job I built Notis to do. Your long-term memory, your skills and your connected apps sit with Notis, not with a model provider, and they travel with every request no matter which surface you send it from: WhatsApp, iMessage, Telegram, Slack or plain email. Claude connects either as an MCP connector or through the Notis CLI, and when Notis runs work on Claude, the model cost lands on the Claude subscription you already pay for rather than being billed twice.

Two honest caveats. Notis will not import your ChatGPT archive for you, so steps one to three above still apply. And the results land in Gmail, Notion or your CRM rather than in a chat panel, which is either the entire point or not what you wanted, depending on how you work. Pro is $13 a month billed annually, and usage is included in your plan up to the monthly allowance rather than metered on top of it, with a typical task drawing a few cents from it.

What I would do, by situation

• You use ChatGPT casually. Run the import prompt, write five lines of context by hand, skip the export. Twenty minutes.
• You run a business on it. Export first, wait for the ZIP, write the context brief yourself, then build three to five projects that mirror how you actually work. Half a day, and the projects are what make it stick.
• You expect to switch again. Do all of the above once, then move the durable context out of the chat app entirely, so the next migration is an afternoon you never spend.

The tools will keep changing. The brief you wrote about your own work will not.