Lesson 6 of 8
Measure, then refine
7 min read
You tweak a prompt, the new answer feels better, so you keep it. But feelings are a bad ruler — the next input might break it. The way pros know a prompt improved isn't a vibe. It's a score.
Stop guessing whether it's better
The fix is to test, not eyeball. Keep a small set of real cases — a handful of inputs where you know what a good answer looks like — and run every version of your prompt against all of them. Now "better" isn't a feeling; it's passed four of five instead of two. A prompt that nails your one example but fails the other four was never actually good.
Evaluate a prompt by running it against a small set of known cases, not one lucky example. A score across real inputs turns "feels better" into something you can actually check.
Change one thing at a time
When the score shows a gap, resist the urge to rewrite everything. Change one thing — add an example, name the audience, tighten a constraint — and re-test. If you change three things and it improves, you've learned nothing about which one worked. Start simple, add detail one revision at a time, and let each change earn its place against the cases.
Refine one variable at a time and re-test. Change several at once and you can't tell which helped — surgical edits are how you learn what your prompt actually needs.
A quick self-check on any answer: was it accurate, complete, clear, the right shape, and aimed at the right reader? Each "no" points at the one part of the prompt to fix next.
The gist
- —Test a prompt against a small set of known cases, not one lucky example
- —"Better" is a higher score across real inputs — not a gut feeling
- —Refine one thing at a time, so you know which change actually worked
You improved a prompt and the new answer looks great on the one example you tried. What should you do before trusting it?
Continue in the app
Take the whole Prompt Engineering course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.