Prompt Autopsy
— external requests · nothing leaves this page external requests ext

Client-side forensics · no upload, no account, no advice

What your AI chats
say about you.

Paste the prompts you have sent to ChatGPT, Claude, Gemini, or anything else that talks back. Get a clinical readout of how you treat a machine that cannot be hurt, cannot remember, and has never once deserved your apology.

Everything below runs on your device. There is no server to upload your chats to, no analytics, and no storage — the only copy of your conversation is the one in your clipboard.

Start the autopsy

01 Evidence

Paste a conversation, or as many of your own messages as you can be bothered to select. Speaker labels are detected automatically. A conversations.json from a ChatGPT data export works too — drop it in, it is read locally with the File API.

Try:

06 Method, and the honest caveats

How it works

Your paste is split into turns, the ones you wrote are isolated, and each axis counts phrase events in your own words: apologies, courtesies, accusations, near-duplicate re-asks (measured with token-set overlap), constraints, median length, compliments, chaos markers, and urgency words. Verdicts are a weighted fit over those nine axes, resolved by a fixed table in js/archetypes.js.

Why there is no leaderboard

A leaderboard would need a server, a database, and your messages. That is the trade this site refuses to make. Results are not stored, not counted, and not seen by anyone — including whoever built it.

Where each bar reaches 100

Saturation point for each of the nine axes

What this is not

Not a psychological assessment, not a diagnostic tool, and not evidence of anything. It is a phrase counter with opinions. Small samples produce confident nonsense — paste at least twenty of your messages before you argue with the result.

Cost, stated plainly

Static files on a CDN. No functions, no database, no third-party APIs, no per-request compute. If this page gets ten million visits, the bill is still zero dollars, which is the only reason it can afford to be honest about your prompts.