Stop rewriting. Start diagnosing.
When people see a low score, the natural reaction is to scrap the prompt and start over. That wastes time. A score is not saying your prompt is wrong. It is saying the model would have to guess at a few things you never mentioned.
So before you touch a word, read the prompt back and ask four questions. Is the goal actually clear, or just clear to you? Did you say who this is for? Did you say what the answer should look like? Did you say anything to avoid? Whichever question you can't answer is almost always where your points went.
The two fixes that do the most work
Out of everything that goes into a score, two things move the needle more than the rest combined: telling the model who the answer is for, and telling it what shape you want back.
People skip both constantly, because in their own head the audience and the format are obvious. They are not obvious to a model reading a bare sentence. Add “for a first-time customer who has never used software like this” and you have handed over context worth a dozen points. Add “three short paragraphs, no bullet points” and you have removed a whole category of guesswork.
Before
“Explain how our return policy works.”
After
“Explain our return policy to a customer who wants to send an item back. Keep it to two short paragraphs, plain language, no legal terms.”
Nothing here is clever. It is just filling in what was missing.
More words are not the goal
A mistake people make once they know the score cares about detail: they add detail that has nothing to do with what the model actually needs. A longer prompt with the same gaps scores about the same as the short one did.
If you added three sentences and the score barely moved, check whether those sentences touched the goal, the audience, the format, or the limits. If they didn't, that is why nothing changed. Padding a prompt with backstory the model doesn't need is not the same as closing a real gap.
When a low score is actually fine
Not every prompt needs to hit 90. If you just want a quick idea, a rough draft, or a one-line answer, a shorter prompt with a lower score can still get you exactly what you wanted in less time than it takes to write a perfect version. Chase a high score when the answer actually matters and you plan to use it as-is. For a throwaway question, let it go.
Make it a habit, not a one-time fix
The real win isn't raising the score on one prompt. It's noticing your own pattern. Most people are weak on the same one or two parts every time, usually context or limits. Once you know your habit, you start writing it into the prompt automatically, and you stop needing the score to catch it for you.