Export a ChatGPT conversation to JSON

Why this format

• A documented, versioned schema

  Every payload carries a schemaVersion so your parser can fail loudly instead of silently when the shape changes.

• Typed content blocks

  Paragraphs, headings, lists, code, tables, and quotes are distinct block types — no string-scraping to find the code.

• Roles and timestamps intact

  Each turn keeps its role and ISO timestamp, which is what you need for evaluation sets and audit trails.

• Drops straight into a pipeline

  Feed it to an embedding job, a dataset builder, or your own renderer without a normalization step.

Example output

{
  "schemaVersion": "1.0",
  "title": "Designing a typed conversion service",
  "messages": [
    {
      "id": "m2",
      "role": "assistant",
      "createdAt": "2026-03-14T14:02:29.000Z",
      "blocks": [
        { "type": "paragraph", "text": "Define the contract first…" },
        { "type": "code", "language": "typescript", "code": "interface…" }
      ]
    }
  ]
}

Step by step

1. 01Paste the public share link and load the conversation.
2. 02Keep 'Include metadata' on if you need the model, retrieval time, and message count.
3. 03Turn off pretty printing for a compact file when storing many exports.
4. 04Download the .json file, or copy it straight into your editor.

Tradeoffs and limits

• The schema describes normalized content, not ChatGPT's internal message tree.
• Alternate branches and regenerated replies are not represented.
• Field names may gain additions in future schema versions; existing fields are kept stable.

Files are generated in your browser. Nothing is stored.