Improve an AI prompt
Paste a prompt you already use and this tool reads it back into its parts — role, context, task, constraints, examples — scores it against a published rubric, then rebuilds it in the right order for your model. Every change comes with the reason for it. Free, no sign-up, and it runs in your browser.
What makes a prompt weak?
Most weak prompts fail in the same five ways. They ask for something “good” without saying what good means. They leave out the background the model cannot infer. They do not say what shape the answer should take. They give no example of a correct result. And they never state how the output will be judged, so the model has nothing to check itself against.
Structure fixes some of that for free — putting context before instructions and the task last measurably changes what a model produces. The rest only you can supply, which is why the score does not move when a prompt is merely reorganised.
Common questions
- How do I know if my AI prompt is any good?
- Check it against six things: is the task specific, is the background supplied, is the output format defined, are there explicit constraints, is there an example, and is there a stated way to judge the result. A prompt missing three or more of those will produce inconsistent output no matter which model you use.
- Why does my prompt score stay the same after it is rewritten?
- Because the score rates the brief you supplied, not the prompt that came out. Reorganising your words adds no information, so the number does not move. It rises only when you add something real — context, a constraint, an example.
- Can it read a prompt written for a different model?
- Yes. It recognises XML tags, markdown headings, uppercase labels and plain "Label:" lines, and anything it cannot identify is preserved as the task rather than discarded. It then rebuilds the prompt in the syntax of whichever model you pick.
Building something from scratch instead? Use the prompt builder, or browse the prompt library by category.