ChatGPT Data Export Not Working: Check the Manifest Before You Retry
By Matthew Firth, founder of ModelBrain · Nov 3, 2026 · 4 min readOur own proposal: check export_manifest.json before you retry a ChatGPT export, because it lists every file and size and so tells you whether your download is complete. Our real export was a 916 MB zip of 1,463 conversations in fifteen numbered files with no single conversations.json, which looks broken and is not.
Start with what you received
| What you have | What it usually means | What to do |
|---|---|---|
| No Export data button | Business and Enterprise workspaces cannot export through ChatGPT settings, per OpenAI | Export from a personal account, or ask your workspace admin |
| No email | OpenAI says an export can take up to 7 days to arrive; check spam and promotions for noreply@openai.com | Wait, and do not start a second request while one is running |
| A dead link | The link expires 24 hours after the email arrives, and it only opens while you are signed in to the account that asked | Sign in to the right account, or request a new export |
| A zip that opens but looks empty | Large exports do not hold one conversations.json | Look for numbered files, then check the manifest |
The first three rows come from OpenAI's export page and the Business note from its transfer page. The fourth row is where our own measurements apply.
A zip with no conversations.json is normal
Our real export was 916 MB and held 1,463 conversations. There was no single conversations.json. The conversations were split across fifteen numbered files, conversations-000.json to conversations-014.json, each a JSON array. chat.html sat beside them, with about 834 image files. If you went looking for one file and found none, nothing is missing. Our walkthrough of the layout shows how the files fit together.
Our own proposal: use export_manifest.json as the receipt
The export includes export_manifest.json, which lists every file in the archive with its size. A download that stopped early can still produce a zip that opens, so open the manifest and compare it with what you extracted: every listed file present, every size matching. It takes a minute, and it tells you whether to retry the download before you build anything on the archive.
A second check: the conversation count
Count the conversations across all the numbered files and compare the total with what you expect from your account. Ours came to 1,463. When we indexed the export with ModelBrain 0.6.6 on October 27, 2026, 1,462 conversations were indexed and one was skipped because it carried the do-not-remember flag, which accounts for all 1,463. If your indexed count and your skipped count do not add up to the files, something is missing.
Once the archive checks out, moving it into Claude and searching it by who wrote what are the next steps.