Anatomy of a good prompt

A good prompt isn't a long sentence — it's a structure. The same structure works for any model — Claude, ChatGPT, Gemini.

The 5 elements

1. Role — who is the model in this interaction? ("You are a BA with 10 years experience")
2. Context — what relevant data does it need? (specs, examples, constraints)
3. Task — what do you want it to do, exactly? One clear thing.
4. Format — how do you want the output? (bullet list, JSON, table, paragraphs)
5. Constraints — limits, tone, length, what to avoid.

Before (weak)

scrie-mi user stories pentru un login

After (structured)

# Rol
Ești un Senior BA care livrează pentru echipe Agile.

# Context
Construim un app B2B SaaS de gestionare flotă auto. Userii sunt
manageri de flotă (non-tech). Avem deja signup, dar nu și login.
Stack: web, mobile-responsive, MFA opțional.

# Task
Scrie user stories pentru fluxul de login.

# Format
Format Gherkin (Given/When/Then), max 5 stories,
fiecare cu câte 2-3 acceptance criteria.

# Constrângeri
- Limba engleză.
- Nu include "forgot password" (separat în alt epic).
- Acoperă: happy path, credențiale greșite, MFA, sesiune expirată.

The 4 common pitfalls

• Vague — "make it better". Better how? Shorter? More formal? Be specific.
• Multi-task — 5 requirements in one sentence. The model will prioritize randomly. Split into 2 prompts.
• Negation — "don't use technical words". Models understand positives better. Say "use everyday language".
• No examples — if you want a specific format, give 1-2 examples. The quality leap is huge.