Prompting patterns
**Prompting patterns** are reusable ways to structure instructions: roles, examples, checklists, step-by-step reasoning requests, and output schemas.
What it is
Prompting patterns are reusable ways to structure instructions: roles, examples, checklists, step-by-step reasoning requests, and output schemas.
Why it matters
Patterns improve average quality. They are not spells. Pair them with verification (hallucinations chapter).
How it works (plain)
Useful patterns:
- Role: “You are a careful editor…”
- Few-shot: show 1–3 examples of the format you want
- Constraints: length, audience, banned claims
- Plan then answer: ask for steps, then the final
- Self-check: “list assumptions; flag uncertainties”
Everyday example
Instead of “fix my email,” use: goal, tone, 3 bullets max, call-to-action, no fake deadlines—plus a draft you will rewrite.
Try it
Rewrite one work prompt using role + constraints + output format. Compare results.
Myths
- ⚠️ Myth: Chain-of-thought always unlocks secret intelligence.
- ✓ Reality: It can help some tasks and hurt others; it also changes what gets logged/shared.
- ⚠️ Myth: More patterns stacked always help.
- ✓ Reality: Contradictory instructions degrade results.
Sources
- Google AI Essentials: https://grow.google/ai-essentials/ ↗
- OpenAI Academy foundations: https://academy.openai.com/en ↗
- ANN live: https://www.ainerdnetwork.com/learn/prompting-patterns ↗