How to Write AI Prompts That Actually Work

90% of AI users get mediocre responses because they write mediocre prompts. It’s not the model’s fault — it’s a matter of technique. In this guide, you’ll learn the core principles of prompt engineering that will transform how you interact with any AI.

Why prompt quality matters

Language models are, at their core, text prediction engines. They predict what text should follow yours. If your prompt is vague, the AI “predicts” a generic response. If it’s precise, it predicts a precise response.

The key is giving the model the contextual information it needs to “predict” exactly what you want.

The 5 principles of a good prompt

1. Specify the role

Tell the AI who it is before asking it anything.

❌ Bad: Explain what SEO is

✅ Good: You are a digital marketing expert with 10 years of experience. Explain what SEO is as if you were explaining it to a business owner who has never heard the term.

The role activates the model’s “expert mode” and leads it to use the right vocabulary and depth.

2. Define the output format

If you need a list, say so. If you need JSON, say so. If you need 3 paragraphs, say so.

❌ Bad: Give me ideas for my blog

✅ Good: Give me 10 article ideas for a personal finance blog. For each idea, include: title, subtitle, and one introductory paragraph. Format: numbered list.

3. Add relevant context

The AI knows nothing about your specific situation unless you tell it.

❌ Bad: Write an email to my client

✅ Good: Write a professional email to a client who has delayed a payment by 30 days. Tone: firm but friendly. The client is a mid-sized company with whom we have a good long-term relationship. The amount is $1,500.

4. Use examples (few-shot prompting)

If you want the AI to replicate a specific style or format, show it examples.

Transform these boring titles into curiosity-generating titles.
Example:
- Boring: "How to save money"
- Improved: "The Japanese saving method that changed the lives of 2 million people"

Now improve these:
1. "How to learn English"
2. "Benefits of exercise"

5. Ask for explicit iteration

The best results usually come on the second or third pass.

✅ Technique: end your prompt with: "Show me the response and explain what you could improve about it with more information from me."

This forces the model to be self-critical and gives you clues on how to refine further.

Advanced techniques

Chain of Thought (step-by-step reasoning)

For complex problems, ask the model to think out loud before responding:

“Before answering, write out your reasoning process step by step. Then give me the conclusion.”

This technique dramatically improves response quality for math, logic, and complex analysis.

System prompt vs. user prompt

If you use the API or advanced tools, separate:

• System prompt: Who the model is, its constraints, its permanent tone
• User prompt: The specific task for that conversation

This separation produces much more consistent responses in long conversations.

Common mistakes to avoid

1. Being too polite: “Could you please, if you have time…?” → Not needed. Just say what you want directly.
2. Not specifying the audience: Text for an expert and for a beginner are very different.
3. Single-line prompts for complex tasks: More context = better results.
4. Not iterating: The first result is rarely the best. Ask for specific improvements.

Universal template

Use this template for any task:

Role: [Define who the AI is]
Task: [Describe exactly what you need]
Context: [Add relevant information about your situation]
Format: [Specify how you want the response]
Constraints: [What it must avoid or must include]

Apply these principles and you’ll immediately see the quality of responses improve. The difference between an average AI user and an advanced one lies almost entirely in how they write their prompts.

---

Want more guides like this? Subscribe to our weekly newsletter.