solidcodersolidcoder
Explore Courses
solidcodersolidcoder
CoursesAboutPrivacy PolicyTerms
© 2026 solidcoder · Practical courses for software engineering interviews.
Home/AI Engineering/Prompt Engineering/Grading Your Own Prompts
Chapters — Prompt Engineering▾

Grading Your Own Prompts

Prompt Engineering·When Things Go Sideways·6 min read·Sep 5, 2026

Five angles for judging whether a prompt is actually good, plus a repeatable loop for improving one that isn't.

Five angles for judging a response

A prompt can be well-written and still miss: technically clear but wrong for your situation, or great once and unreliable the second time. Judge results, not just prompts, along five angles: is it technically well-formed, actually relevant, usable as-is, worth the effort it took, and reliable across similar requests.

Quick diagnostic checklist
Clarity
Could this prompt be read more than one way?
Context
Does the model have everything it needs about your specific situation?
Format
Did you specify structure, length, and tone, or leave them to chance?
Constraints
Are the boundaries realistic and non-contradictory?
Technique
Would a role, an example, or step-by-step reasoning have helped here?
✕ Under-specified
How can I grow my business?
Open in ChatGPT ↗

Here are some general ways to grow a business: know your audience, improve your product, invest in marketing, build partnerships, gather customer feedback, expand your offerings, monitor competitors... (a list that fits literally any business)

✓ Diagnosed and rewritten
I run a 4-person landscaping business relying entirely on word-of-mouth. I want 2–3 new residential clients per month with a marketing budget under $300/month. Recommend 3 specific low-cost tactics I can start this week, with expected time investment and how I'd measure if each is working.
Open in ChatGPT ↗

1) Google Business Profile optimization with before/after photos: 2 hours setup, measure by profile views and call clicks. 2) A referral incentive for existing clients ($25 credit per referral): 1 hour to set up, measure by tagged referrals. 3) Partner with a local realtor for move-in yard cleanups: a few outreach emails, measure by signed jobs from the partnership.

The improvement loop

  • Baseline: write the prompt, note what you actually wanted, run it once.
  • Gap analysis: name the specific shortfall (vague, missing context, wrong format, wrong scope).
  • Targeted revision: fix one category at a time, not everything at once.
  • Re-test: compare the new result to the old one; keep what changed for the better.
  • Save it: a prompt that worked well for a recurring task is worth keeping as a template.

Two techniques for pressure-testing a good answer

The devil's advocate test
Review the recommendation you just gave and identify: 1) what assumption might be wrong, 2) what could go wrong if I follow it, 3) what's missing.
Open in ChatGPT ↗

Useful right after getting advice you like: it surfaces blind spots the first pass glossed over.

The specificity challenge
Before answering, ask me 3 clarifying questions that would help you give a better response.
Open in ChatGPT ↗

If the questions it asks are ones you'd already answered, your original prompt was specific enough. If they're new, you just found the gap.

Up next4/4
Part 3 · When Things Go Sideways
← Prev Section
Core Skills
Why Instructions Matter More Than the Model
Next Section →
Leveling Up
Getting the Model to Show Its Work
Part of a free guide

Prompt Engineering

A simple guide to writing better prompts, so AI gives you the answer you actually want.

Browse All Guides →
On this page
  • Five angles for judging a response
  • The improvement loop
  • Two techniques for pressure-testing a good answer