Part 4 - Getting It Right
15. Avoid the Common Mistakes
Goal: Recognise and fix the prompt errors that cause most bad answers.
Most disappointing answers trace back to a handful of prompt errors: being vague, asking for too much at once, giving no context, and giving contradictory instructions. Knowing the list lets you diagnose a bad answer instead of blaming the AI.
The fix is almost always to make the prompt clearer, smaller or better-supported - never to repeat the same weak prompt louder.
The Script (Template)
Diagnose a bad answer: Was I vague? Did I overload it? Did I give context? Did I contradict myself? -> Fix that one thing and resend.
Worked Example
You write
Bad prompt: 'Write something good about our business.' Improved: 'You are a copywriter. Write a 2-sentence tagline for a family-run bakery known for sourdough. Tone: warm. No exclamation marks.'
AI replies (sample)
Improved answer: Real bread, made the slow way, by a family who still counts every loaf as ours. Come taste the difference a little patience makes.
Why it works
The bad prompt failed on vagueness, no context and no format. Fixing exactly those three turned a hopeless request into a usable one.
Do & Avoid
Tips
- When an answer is bad, re-read your prompt before re-reading the answer.
- Change one weakness at a time so you learn what actually helped.
Common Mistakes
- Blaming the AI for a vague prompt.
- Making the prompt longer instead of clearer.
Practice
- Take a bad answer you got earlier and diagnose which error caused it.
- Rewrite that prompt fixing only the single biggest weakness.
Teacher Notes Suggested: 20 min
Learning objectives
- Students diagnose prompt errors systematically.
- Students fix the root cause, not the symptom.
Talking points
- Build a class 'prompt error checklist' poster.
- Frame bad answers as debugging practice, not failure.
Answer key / what good looks like
- Diagnoses should name a specific error (vague/overloaded/no context/contradictory).
- Fixes should target that error only.
Discussion questions
- Why is 'the AI is dumb' usually the wrong conclusion?