Why Most People Write Bad Prompts
The difference between a beginner prompt and an expert prompt is often just 20-30 extra words. But those words change the output from "meh" to "wow."
Most people write vague prompts: "Write me a function to handle users." Experts write specific prompts: "Write a TypeScript function that validates user registration input. Check email format, password strength (min 8 chars, 1 uppercase, 1 number), and return typed error messages. Use zod for validation."
Here are 21 tips that separate amateurs from experts.
Tips 1-7: Structure Your Prompts
1. Specify the output format — "Return as a JSON object with keys: name, type, description" vs "tell me about it."
2. Give a role — "You are a senior TypeScript developer with 10 years of experience" primes the AI for higher-quality output.
3. Show an example — Include a before/after example of what you want. AI learns patterns instantly.
4. Break complex tasks into steps — "First analyze the code, then identify bugs, then suggest fixes with explanations."
5. Set constraints — "Use no external dependencies. Keep the function under 50 lines. Follow the existing code style."
6. Specify what NOT to do — "Do not add comments. Do not change the function signature. Do not use any deprecated APIs."
7. Provide context — Paste relevant code, documentation, or requirements. More context = better output.
Prêt à Maîtriser l'IA ?
Rejoignez 2 500+ professionnels qui ont transformé leur carrière avec le Bootcamp IA CodeLeap.
Tips 8-14: Code-Specific Techniques
8. Reference the tech stack — "Using Next.js 14 App Router with TypeScript and Tailwind CSS" prevents framework version confusion.
9. Ask for error handling — "Include error handling for network failures, invalid input, and rate limiting."
10. Request tests alongside code — "Write the function AND unit tests using Jest. Cover edge cases."
11. Chain prompts — Start with architecture, then implementation, then tests, then optimization. Each prompt builds on the previous.
12. Use diff mode — In Cursor, ask for changes as diffs to understand what's changing.
13. Ask for explanations — "Implement this and explain each design decision" helps you learn while building.
14. Iterate, don't restart — If the output is 80% right, refine it: "Good, but change X to Y and make Z more efficient."
Tips 15-21: Advanced Patterns
15. Mega-prompt pattern — Combine role, context, task, format, and constraints in one structured prompt. This is the single most impactful technique.
16. Few-shot prompting — Provide 2-3 examples of input/output pairs before your actual request.
17. Chain-of-thought — Ask AI to "think step by step" for complex reasoning tasks.
18. Critique and revise — Ask AI to review its own output: "Now review this code for security vulnerabilities and fix them."
19. Persona stacking — "Review this as both a security expert and a performance engineer."
20. Temperature control — For creative tasks, ask for "multiple diverse approaches." For precise tasks, ask for "the most standard, production-ready solution."
21. Meta-prompting — Ask AI to write a better prompt: "I want to achieve X. Write me the optimal prompt to ask an AI to do this."
Master Prompt Engineering in Practice
These tips are a starting point. Real mastery comes from practice — hundreds of prompts across real projects, learning what works and what doesn't.
CodeLeap's AI Bootcamp dedicates an entire module to prompt engineering, with live exercises, peer reviews, and expert feedback. You'll build a personal prompt library that you'll use for years to come.