Guide · Data inventory
What is actually inside your AI export
Before you trust a file with "your data", know what it really holds. Here is the honest inventory of a typical ChatGPT, Claude or Gemini download.
The core: full conversation transcripts
Every message — user, assistant and system — with its role, ordering and roughly the point in time it was
sent. For ChatGPT this is usually inside conversations.json or
per-chat files in a ZIP; Claude exports plain JSON per conversation; Gemini’s Takeout archive ships
Gemini_History.json. Prompts, outputs, and code are all in here
verbatim.
System instructions & custom prompts
ChatGPT custom instructions and Claude project/system prompts surface as system-role content. If you have ever pasted an API key, a company policy or your own PII into a system prompt, it is right there in the file.
Metadata: timestamps, models & shared links
Records typically add creation/update timestamps, an internal conversation ID, model version, and references to shared chats or plugins. None of it is a problem by itself — but combined with transcripts it tells a surprisingly complete story about how you work.
The part people forget: secrets travel with the file
If you ever pasted an API key, a token, a database URL or a password fragment into a chat, that credential is sitting in plain text inside the JSON. Before you share, upload or archive the file, run the analyzer's secret scanner — it flags recognisable key and token patterns entirely in your browser.
What is usually missing
Permanently deleted conversations, prompt text you typed into a chat that was purged, and any data tied to a differently-signed-in account are generally absent. The export is a snapshot of what existed at download time, not a full server mirror.
Want to see your own archive in plain language? Drop the file into the analyzer — no upload needed.
Questions
Does a ChatGPT export include everything I ever typed?
It includes the conversations in your account download — but file distribution varies (a ZIP or a few JSON files). Details like deleted items are generally not included once purge has run.
Are custom instructions inside the export?
Yes. System-role content — including ChatGPT custom instructions and Claude system prompts — appears in the conversation files and can be extracted for auditing.
Do shared links appear?
Conversation records frequently carry an ID or reference for shared chats, though the export is primarily the transcript plus metadata rather than a link directory.
Is the export readable without special software?
Not really — it is structured JSON meant for machines. A local analyzer turns it into searchable text and clean exports.