Free tool · No account · Nothing leaves your browser
X-ray your prompt.
Most bad AI answers are underspecified prompts. Paste yours and five checks name exactly what it gives the model, and what it forces the model to guess.
X-RAY YOUR PROMPT
RUNS IN YOUR BROWSER · NOTHING IS SENT
- A clear ask: what exactly should it do?
- An audience: who is this for?
- A format: list, email, table, tone?
- Constraints: length, bounds, must-nots?
- Context: enough background to work with?
The five-ingredient method
A language model answers with whatever the prompt does not pin down. Across every chat tool (ChatGPT, Claude, Gemini, Copilot), weak answers trace back to the same five missing ingredients. The X-ray checks each one:
The most common failure is a prompt that never states the deliverable. "Thoughts on this?" makes the model guess whether you want a critique, a rewrite, or a summary. Strong prompts name the action and the artifact.
The same facts are explained differently to a customer, a colleague, or an expert. Without an audience the model explains to nobody in particular, which reads as generic. One word moves vocabulary, depth, and tone.
A model told "an email", "a table", "twelve bullet points" spends its effort on content instead of guessing shape. Format words are the cheapest quality upgrade in prompting.
Length bounds, must-includes, things to avoid. Constraints are how you say no in advance. Without them, output trends long and hedged, padded with caveats you will delete anyway.
Background the model cannot know: what the project is, what you tried, what good looked like last time. Context is the difference between advice in general and advice for you.
How the checker works
The X-ray is deliberately not AI. It is a set of deterministic checks (word patterns and length thresholds) that run as JavaScript on this page. Pressing the button makes no network request: your prompt never leaves the browser, is never logged, and is never used to train anything. You can verify this in your browser's network inspector.
Deterministic checks can name what is missing; they cannot write it in. That second step, turning a 2/5 draft into a complete prompt with your context and constraints filled in, is what the Pliny extension does with a language model, directly in whatever text field you are typing in.
Common questions
- Is my prompt sent to a server?
- No. Every check runs locally in your browser. This page makes no network request when you press the button, and nothing you paste is stored.
- Does the checker use AI?
- No, and that is the point. The checks are instant and free at any volume, and everything stays in your browser. The AI-powered rewriting happens in the Pliny extension, where a full model reads your draft and produces the stronger prompt.
- Which AI tools do the five ingredients apply to?
- All of them. The ingredients describe what any instruction-following model needs to stop guessing. They apply equally to ChatGPT, Claude, Gemini, Copilot, and whatever ships next.
- What is the difference between this page and Pliny?
- This page diagnoses. Pliny treats: it rewrites the prompt in place, in any text field on any site, and shows you the exact diff before anything is replaced. The free plan includes 5 rewrites a month.
The X-ray names what is missing.
Pliny writes it in.