How to Write Prompts That Actually Work
Have you ever asked ChatGPT something and been completely disappointed with the answer? Too generic, off-topic, or just plain useless?
Good news: the problem is rarely the tool. It's how you communicate with it.
Writing good prompts is a skill — and one that can completely transform what you get out of AI, whether you're using it for work, studying, or running a business. This guide shows you exactly how to do it.
The 5 Pillars of a Prompt That Works
Before diving into techniques, you need the foundations. We call them P-C-R-F-I: Precision, Context, Role, Format, and Iteration. Sounds complicated — it isn't.
1. Precision — Be Specific, Always
AI doesn't guess intentions. The vaguer you are, the more generic the response.
❌ Weak prompt: "Tell me about digital marketing."
✅ Strong prompt: "Write a 300-word summary of the three main on-page SEO strategies to increase organic traffic in 2025."
How to be more specific:
- Define the topic precisely
- Specify the platform or channel (e.g., "LinkedIn post")
- Mention the target audience
- State the goal: inform, sell, summarise, persuade?
2. Context — Give It a Map Before Asking for Directions
Without context, AI is guessing. With rich context, the result is a different story.
❌ No context: "Write a sales email."
✅ With context: "Write a sales email for 'ProductXPTO', a natural energy supplement for busy professionals aged 30–55. Highlight that the ingredients are natural and the dose is once daily. The goal is to schedule a demo."
The more relevant information you provide, the more precise and useful the response.
3. Role — Tell It Who to Be
Asking AI to take on a specific role is one of the most powerful techniques available.
❌ No role: "Give me marketing tips."
✅ With role: "Act as a digital marketing consultant with 15 years of experience in startups. What are your top three recommendations for a B2B software company looking to generate more qualified leads on a limited budget?"
The difference in output is enormous. The AI adapts its tone, depth, and language to the role you assigned.
4. Format — Specify How You Want the Information
The same content can arrive as an unreadable wall of text or as something ready to use. You choose.
❌ No format: "What are the benefits of meditation?"
✅ With format: "List the 5 main benefits of meditation for stress reduction in bullet points. For each benefit, add a one-sentence description."
You can request:
- Numbered or bulleted lists
- Comparison tables
- Code (specifying the language)
- Summary in X paragraphs
- Text with a word limit
5. Iteration — The First Prompt Is Rarely the Best
Nobody nails it on the first try. That's completely fine. Prompt engineering is a process of continuous refinement.
Draft → Test → Evaluate → Refine → Repeat
Every AI response is feedback. If you weren't satisfied, adjust the prompt — add more context, change the format, or reword the instruction.
Advanced Techniques (Without the Complexity)
With the five pillars in place, these techniques will take your results to the next level.
Zero-Shot — The Direct Request
The most basic approach: you ask, the AI responds using what it knows.
"Translate 'Hello, world!' into French."
"Classify this text as positive, negative, or neutral: [text]"
When to use: Simple, fast tasks that don't require much context.
Few-Shot — Show Examples
Instead of explaining what you want, show it. Give the AI one or two examples of the expected output and it follows the pattern.
When to use: When you need a specific style, tone, or format that's hard to describe in words.
Chain-of-Thought — Ask It to Think Out Loud
Instructing the AI to solve a problem step by step significantly improves accuracy, especially for reasoning and calculations.
"Calculate the total spent, showing your working: the customer bought 5 apples at €2 each and 3 oranges at €3 each."
When to use: Logic problems, maths, or any time you need transparency in the reasoning.
Delimiters — Clearly Separate Parts of Your Prompt
Use markers like """ or ### to distinguish your instructions from the content you want the AI to analyse.
Summarise the text below in 2 sentences.
"""
[paste your long text here]
"""
Without delimiters, AI can confuse your instructions with the content. With them, everything is clear.
Negative Prompts — Say What You Don't Want
Sometimes it's easier to exclude than to describe.
"Suggest names for a new coffee shop. Names should be short and modern. Do not include the word 'coffee' or any coffee-related puns."
Use sparingly — positive instructions tend to work better, but negatives are useful for fine-tuning.
Technique Comparison Table
| Technique | What It Is | Best For |
|---|---|---|
| Zero-shot | Direct request, no examples | Translation, factual questions |
| Few-shot | Provide 1–3 output examples | Classification, specific style |
| Chain-of-thought | Ask for step-by-step reasoning | Logic and calculation problems |
| Delimiters | Separate instructions from content | Summarising text, analysing documents |
| Negative prompts | Say what you don't want | Restricting unwanted outputs |
Ready-to-Use Examples
Copy, adapt, and use these prompts in your daily work.
✉️ Professional Email
Write a professional email to [client name] thanking them for yesterday's meeting about [topic].
The goal is to reinforce the points discussed, highlight next steps, and close with an invitation to clarify any questions.
Tone: cordial and direct. Max 200 words.
📝 Text Review
Review the text below:
1. Fix grammatical and punctuation errors
2. Improve readability without changing the content
3. Maintain a professional tone
"""
[paste your text here]
"""
📊 Tool Comparison
Compare [Tool A] and [Tool B] across the following criteria:
- Ease of use
- Pricing
- Team collaboration
- Integrations
Present as a table. At the end, recommend which is better for [profile or use case].
🗓️ Study Plan
Create an 8-week study plan to learn [topic], focused on [goal].
The person can dedicate [X hours] per week.
For each week: topics, free resources, and one practical exercise.
Present as a table: Week | Topics | Resources | Exercise.
The Most Common Mistakes (and How to Avoid Them)
| ❌ Mistake | ✅ Fix |
|---|---|
| Vague prompt ("talk about X") | Be specific: define topic, format, and goal |
| No context | Provide relevant background information |
| No format specified | Ask for a list, table, paragraphs, etc. |
| Giving up after a bad response | Iterate — adjust and test again |
| Trying to solve everything in one prompt | Break it into smaller tasks |
The Secret Nobody Tells You
The biggest difference between people who get mediocre results from AI and those who get extraordinary ones isn't the model — it's the clarity of thought before writing the prompt.
Before writing anything, ask yourself:
- What exactly do I need?
- Who is this response for?
- In what format do I want to receive it?
- What context does the AI need to know?
When you have these answers, the prompt almost writes itself — and the results speak for themselves.
💡 At WizardingCode, we help businesses integrate AI into their processes in a practical and effective way. If you want to know how intelligent automation can transform your business, talk to us.