What makes a high-scoring prompt

The quickest way to understand a good prompt is to watch a bad one become good. Below is a weak prompt and a strong one, each run through the Promptrace scorer, so you can see exactly what closes the gap.

Last reviewed

Before and after

Same goal, a blog post about coffee. The only thing that changes is how completely the prompt briefs the model. That alone moves the score by 87 points.

Weak prompt

write a blog post about coffee

13
out of 100
Role definition0
Task clarity60
Specificity12
Context0
Output format0
Constraints0
Examples0
Strong prompt

You are a senior content strategist with 10 years of experience writing for speciality coffee brands. Write a 1,200-word blog post about pour-over brewing for home baristas in the UK who already own a gooseneck kettle. The target audience is curious hobbyists, and the goal is to convert readers into subscribers to our monthly bean club. Structure the post with one H1, five H2 sections, and a short FAQ. You must keep each paragraph under 80 words. Use a warm, plain-spoken tone and British English. Only use jargon if you explain it: for example, describe the bloom as 'the first pour that lets trapped gas escape'. Do not include affiliate links.

100
out of 100
Role definition100
Task clarity100
Specificity100
Context100
Output format100
Constraints100
Examples100

What changed, line by line

The strong version is not longer for the sake of it. Every addition maps onto one of the seven dimensions the scorer measures.

Role

Names a senior content strategist with relevant experience, which loads the right vocabulary and standards.

Task

Uses concrete verbs (write, structure, describe) instead of leaving the model to guess the job.

Specificity

Adds a word count, a precise topic, a country, and a reader who already owns a gooseneck kettle.

Context

Spells out the audience and the goal: turning curious readers into bean-club subscribers.

Output format

Asks for one H1, five H2 sections, and a short FAQ, with a paragraph length cap.

Constraints

Sets a tone, British English, a per-paragraph limit, and bans affiliate links.

Examples

Shows what a plain-English jargon gloss looks like, so the model copies the pattern.

Mistakes that tank a score

Starting with the task, not the role

Jumping straight to "write a blog post" skips the single cheapest upgrade available. One line of role, "You are a senior content strategist", shifts the whole register of the answer.

Being vague to seem open-minded

Leaving things unspecified does not give the model freedom; it makes it guess, and it guesses towards the bland average. Specific inputs produce specific output.

Forgetting the format

If you do not name the shape of the answer, you get the model's default, which is usually a wall of prose. Say table, list, word count, or sections.

No constraints at all

Constraints cost nothing and tighten everything. A length cap, a banned-words list, and a tone instruction routinely lift an answer from usable to good.

Describing instead of showing

When the output is hard to describe, show one example. A single worked example often communicates more than a paragraph of instructions.

Score your own prompt

Paste it into the free scorer and see which of the seven dimensions you are missing. No signup.

Open the prompt scorer →

Frequently asked questions

What makes a prompt high-scoring?

A high-scoring prompt covers all seven dimensions the scorer measures: it assigns a role, states the task with a clear verb, includes specific details, supplies context, names the output format, sets constraints, and ideally shows an example. The weak example on this page scores low because it does only one of those; the strong version scores high because it does all of them.

How long should a good prompt be?

Long enough to cover the seven dimensions and no longer. In practice that is usually two to six sentences. The goal is not word count for its own sake; it is making sure the model has the role, task, context, format, and constraints it needs. A short prompt that covers them beats a long, rambling one that does not.

Does a higher score guarantee a better answer?

No. The score is a directional signal of how completely you have briefed the model, not a guarantee. A prompt can score well and still miss because the underlying request was wrong. Use the score to catch the dimensions you skipped, then judge the actual output on its own merits.