No AI platform can import another's history directly. The working method is to export from each platform, convert everything into Markdown or plain text, and keep that as your own archive. From there you can upload the relevant parts into whichever platform you are using now, and you are never locked into one vendor again.
Short answer
No AI platform can import another's history directly. The working method is to export from each platform, convert everything into Markdown or plain text, and keep that as your own archive. From there you can upload the relevant parts into whichever platform you are using now, and you are never locked into one vendor again.
Step by step
Export from every platform you use: ChatGPT via Settings, then Data Controls. Claude via Settings, then Account. Gemini via Google Takeout, selecting the Gemini product specifically. Each arrives in a different format, and each has its own delay before the file is ready.
Convert everything to one format: Markdown is the right target. It is readable as plain text, every AI ingests it cleanly, notes apps import it natively, and it will still open in fifty years. Convert all platforms to the same format so the archive is uniform.
Organise by topic, not by vendor: Folder your archive by project or subject rather than by which AI produced it. Which model you happened to use is the least interesting attribute a year later, and vendor folders make the archive useless the moment you switch again.
Upload the relevant slice into the new platform: Do not bulk-import. Pick the conversations carrying standing context and upload those as project knowledge, a Claude Project, a Gemini Gem, or a custom GPT's reference files.
Re-export on a schedule: An export is a snapshot. Put a quarterly reminder in your calendar and re-export from every active platform, so the archive stays close to current rather than freezing on the day you set it up.
Most people did not choose to spread their history across four AI vendors. It happened one product launch at a time, and now the useful context is scattered: the architecture discussion in Claude, the research in Perplexity, the first two years in ChatGPT, the odd thing in Gemini because it was already in the browser.
There is no interoperability standard coming to fix this. What there is, is a format that every platform reads and none of them owns.
Why every platform exports differently
| Platform | Where | Format | Delay |
|---|---|---|---|
| ChatGPT | Settings, Data Controls, Export Data | ZIP with conversations.json | Minutes to hours |
| Claude | Settings, Account | JSON by email | Usually under an hour |
| Gemini | Google Takeout, select Gemini | HTML or JSON in a ZIP | Hours to days |
| Grok, Poe, NotebookLM, AI Studio | Varies by product | Varies | Varies |
Four platforms, four schemas, four failure modes. This is not accidental. There is no commercial reason for any vendor to make your history portable, and the format differences are what keep switching costs high.
Markdown is the interchange format
Pick one target format and convert everything into it. Markdown wins for four reasons that no other candidate matches at once:
Every AI reads it natively. Upload Markdown to a Claude Project, a custom GPT, or a Gemini Gem and it is parsed without preamble.
Every notes app imports it. Obsidian, Notion, Logseq, and plain folders all treat it as first-class.
It is human-readable as raw text. No viewer required, now or in twenty years.
It diffs and greps. Your archive becomes searchable with tools you already have.
PDF is the wrong choice for an archive you intend to reuse, because it is designed to be final rather than machine-readable. Keep PDFs for records that need to be fixed, and Markdown for everything you might feed back into an AI.
Converting all of them in one place
ChatExports reads the export formats from nine platforms, including all of those above, and converts any of them into Markdown, Word, PDF, CSV, or plain text. Parsing runs locally in your browser, so a file containing your entire history across every AI you use is never uploaded anywhere.
The practical shape of the job is: export from each platform, drop each file into the converter, pick Markdown, and bulk-export into one folder. An afternoon, once, and then a quarterly top-up.
Structure the archive so it survives the next switch
Organise by subject, not by vendor:
Vendor-based folders encode a detail that stops mattering almost immediately, and guarantee a reorganisation the next time you move. Subject-based folders survive every switch you will make.
Then only move what earns its place
The migration mistake is treating the new platform as somewhere to dump the archive. Platforms have knowledge limits, and irrelevant history measurably degrades answer quality. Upload the primer plus the genuinely load-bearing conversations, and leave the rest in your own archive where you can search it.
Can any AI platform import another one's chat history?
None of them can, and none has announced plans to. Every migration goes through your own files: export from the source, convert to a common format, and upload the relevant parts into the destination as project knowledge.
What is the best format for archiving AI conversations?
Markdown. Every AI ingests it, every notes app imports it, it stays readable as raw text without any viewer, and it works with ordinary search tools. Use PDF only for records that need to be fixed and unchangeable.
How often should I export from each platform?
Quarterly for most people, monthly if you rely on AI heavily for work. Always export before changing plans, closing an account, or deleting conversations, because those actions are irreversible and the export only captures what exists at that moment.
Is it safe to convert exports from several platforms in one tool?
It depends entirely on where the conversion happens. ChatExports parses every file locally in your browser and uploads nothing, which is the only sensible model for a combined archive of everything you have asked every AI.
Should I organise the archive by platform?
No. Organise by subject or project. Which model produced a conversation is the least useful attribute a year later, and vendor folders force a reorganisation every time you switch platforms.