Strict 0–100 rubric
Six weighted dimensions: specificity, context, structure, iteration, scope, meta-prompting. Calibrated so a competent median lands around 50, not 80.
A strict rubric for how you prompt AI
Find out what kind of prompter you actually are.
Drop in a ChatGPT or Claude.ai session, or score your Claude Code and Codex CLI sessions straight from the terminal. Rubric grades on six dimensions, names your prompting archetype, and gives you a shareable score. From your turns only.
Free to start · No account needed to see your score
User turns only · PII auto-redacted
Results in seconds
Drag-and-drop ingest
Web · CLI
Why Rubric
A red pen for your prompts.
Rubric is opinionated by design. It scores how you actually communicate with the model, not whether you got a good answer. The goal is to get better at prompting, not to feel good about every session.
Eight prompting archetypes
Get pinned to The Sculptor, Architect, Wanderer, Explorer: eight patterns with growth tips tailored to how you actually use AI.
Your messages only
Assistant replies are summarized locally and dropped before anything leaves your browser. The scorer never sees AI output.
Shareable score card
Unlock a public score URL and a 1200×630 PNG with one click. Perfect for a tweet or a LinkedIn post.
How it works
Three steps, under a minute.
- 1
Drop your export
Drag a ChatGPT or Claude.ai data export onto the analyzer. We detect the format automatically and strip everything that isn’t your turns.
- 2
See your score
A strict rubric grades the whole session in seconds. You get a number, an archetype, and a one-paragraph verdict.
- 3
Unlock the breakdown
Sign in via magic link to see per-dimension feedback with evidence turns, growth tips, and a shareable card.
No. The scoring model only ever sees your turns. Assistant text is summarized locally in your browser into three primitives (length, whether it contained a code block, whether it contained an error), and only those summaries leave your device. The full assistant text never goes to the analyzer. There's a methodology page with the wire-level details and the tests that verify it.
Ready to see your number?
Score a session right now. The email gate is only required for the full breakdown; the score and archetype come back first.
Prefer the terminal?npm i -g rubric-chat