What the number actually means
A prompt score rates how clear and complete your prompt is. Think of it as a readiness check, not a report card. A low score does not mean your request is silly. It means you left gaps the model will have to fill by guessing.
This matters because an AI builds its answer from the words you give it. A prompt with gaps forces the model to guess the parts you left out, and a guess is often not what you wanted. A high score means there is little left to guess, so the answer is far more likely to match what you had in mind. The score is really a measure of how much guessing you have removed.
The parts behind the score
The number is not one vague judgment. It is built from a few clear parts, each one a question about whether the model has what it needs. When you understand the parts, the score stops being a mystery and becomes a checklist.
Clear goal
Is it obvious what you want done? "Write something about our app" is fuzzy. "Write a 200-word intro for our app for busy small-business owners" is clear.
Context
Did you give the background the model cannot see? Who is the reader, what is the purpose, what is the situation.
Format
Did you say how the answer should look? A list, a table, three bullets, under 100 words, a certain tone.
Limits
Did you say what to avoid? "No jargon", "do not oversell", "assume no technical background".
Specifics
Did you include the concrete details that anchor the answer? Real names, numbers, or examples instead of vague terms.
A prompt that covers all of these scores high. A prompt missing several of them scores low. Most prompts sit in the middle, strong on one or two parts and thin on the rest.
A low score and a high score, side by side
The clearest way to feel what a score measures is to see the same request written two ways.
Low score
“Write a welcome email.”
No goal detail, no context, no format, no limits. The model has to guess almost everything.
High score
“Write a welcome email for people who just signed up for our budgeting app. They are nervous about money and new to budgeting. Keep it under 120 words, warm and plain. No finance jargon, and do not oversell.”
Goal, context, format, and limits are all there. Little is left to guess.
How to read your own score
When you see a score, do not just look at the number. Look at which parts are weak, because that is where the useful information is. The number tells you there is room to improve; the breakdown tells you exactly where.
A prompt scoring in the 80s or 90s is usually ready to send. Something in the middle is worth one quick fix to the weakest part. A low score means the model would be guessing a lot, and a minute of editing will change the answer completely. You do not need a perfect score, you need enough that the model is not guessing about the things that matter to you.
One thing a score cannot do
A prompt score measures clarity, not correctness. It can tell you your prompt is well-built. It cannot tell you the AI's answer will be true, because the model can still make things up. So use the score to write a clear request, and still check the important facts in the answer. A strong prompt gets you a better draft, not a guaranteed one.