Guide · Comparison
ChatGPT vs Claude vs Gemini exports
Three companies, three JSON philosophies. Here's how OpenAI, Anthropic and Google structure the chat history you download — and why it matters when you analyze it.
ChatGPT — the mapping graph
OpenAI ships a conversations.json array. Each conversation is a mapping graph: every message is a node keyed by ID, with create_time, author.role, content parts, and — on assistant nodes — a model_slug. The graph structure captures branches and edits that linear formats lose. It's the richest export of the three, and the heaviest.
Claude — plain chat_messages
Anthropic's export is a per-conversation JSON with a flat chat_messages array: sender is human or assistant, timestamps are ISO strings, and content is text or structured blocks. There's no graph — just a clean, chronological list. That makes it the easiest to read by eye and the most portable for scripting.
Gemini — the Takeout list
Google's export arrives through Google Takeout as Gemini_History.json: a list of conversations, each with a title, created/updated timestamps and a messages array of parts with a role of user or model. Compact and uniform — but it's inside a Takeout archive you have to request and wait for.
How the analyzer unifies them
AI Data Inspector detects each shape — ChatGPT mapping graphs, Claude chat_messages, Gemini Takeout lists, DeepSeek maps, and generic role-based JSON/JSONL — and normalizes them into a single model. That means one search box, one scanner and one set of exports across every platform, all processed in your browser.
Questions
Which export format is easiest to analyze?
Claude’s per-conversation JSON is the tidiest for humans to read; ChatGPT’s mapping graph is the most complete; Gemini’s Takeout JSON is compact and simple. All three analyze cleanly with AI Data Inspector.
Do all three include timestamps and models?
Yes. Each export carries per-message timestamps; ChatGPT and Claude include model metadata on assistant messages more consistently than Gemini Takeout.
Can I analyze all three exports at once?
Yes. Drop multiple files — even from different platforms — into the analyzer and they are unified into one searchable, scannable archive.
Are the formats stable?
Platforms evolve their export shapes over time. The analyzer uses tolerant detection, and we update parsers as formats drift.