Prompt Engineering: A Practical Guide to Getting Better AI Results
Learn the techniques that separate great AI results from mediocre ones. A practical guide to writing prompts that actually work, with templates you can use today.
You've probably noticed that some people get incredible results from ChatGPT while others struggle to get anything useful. The difference isn't luck or a premium subscription—it's how they write their prompts. The good news? This is a learnable skill, and you don't need to be technical to master it.
Key Insight
Prompt engineering isn't about tricks or hacks—it's about clear communication. The better you can explain what you want, the better results you'll get. Think of it as learning to brief a very capable but very literal assistant.
Why Prompts Matter More Than You Think
When you type something into ChatGPT or Claude, you're not just asking a question—you're setting the entire context for how the AI will respond. A vague prompt gets a vague answer. A specific prompt gets a specific answer. It's that simple, and that powerful.
Consider the difference:
❌ Weak Prompt
"Write something about marketing"
Result: Generic, unfocused content that could apply to anything
✓ Strong Prompt
"Write 3 email subject lines for a SaaS company announcing a new feature to existing customers. Tone: excited but professional. The feature is automatic report scheduling."
Result: Specific, usable content you can actually use
The second prompt works because it specifies the format (3 subject lines), the context (SaaS company, existing customers), the tone (excited but professional), and the subject matter (automatic report scheduling). The AI has everything it needs to give you exactly what you want.
The Anatomy of a Great Prompt
Every effective prompt contains some combination of these elements. You don't always need all of them, but knowing what's available helps you construct better requests.
🎯 Task
What do you want the AI to do? Be specific about the action: write, analyse, summarise, compare, explain, create, etc.
🎭 Role
Who should the AI pretend to be? A marketing expert, a teacher, a critic, a supportive friend? This shapes the perspective and expertise level.
📋 Format
How should the output be structured? Bullet points, numbered list, table, essay, email, code, JSON? Specify length if it matters.
🌍 Context
What background information does the AI need? Who's the audience? What's the situation? What constraints exist?
💬 Tone
How should it sound? Formal, casual, humorous, empathetic, authoritative, friendly? This dramatically affects the output.
📝 Examples
Can you show what good output looks like? Examples are incredibly powerful for getting consistent results in a specific style.
Technique 1: Be Specific About What You Want
Vagueness is the enemy of good AI output. Every time you leave something unspecified, the AI has to guess—and it might guess wrong.
Example: Writing a Product Description
Too vague:
"Write a product description for headphones"
Specific and effective:
"Write a 100-word product description for wireless noise-cancelling headphones. Target audience: remote workers who take lots of video calls. Key features to highlight: 30-hour battery, comfortable for all-day wear, excellent microphone quality. Tone: professional but warm. Include a compelling opening line."
Notice how the second prompt eliminates ambiguity. The AI knows the word count, the product type, who it's for, what features matter, and how it should sound. There's no guessing involved.
Questions to ask yourself before submitting a prompt:
- Did I specify the format and length?
- Did I explain who the audience is?
- Did I mention what tone or style I want?
- Did I include all relevant context?
Technique 2: Assign a Role
One of the most powerful techniques is telling the AI who to be. This isn't just roleplay—it fundamentally changes how the AI approaches your request by activating different "knowledge patterns."
The Role Technique in Action
"You are a senior software engineer with 15 years of experience. Review this code and identify potential bugs, security issues, and performance problems. Explain issues in a way a junior developer would understand."
By assigning a role, you're telling the AI:
- What expertise level to operate at
- What perspective to take
- What kind of advice is relevant
Useful roles to try:
- "You are a [job title] with [X years] experience..."
- "Act as a devil's advocate and challenge this idea..."
- "You are a teacher explaining this to a complete beginner..."
- "Pretend you're a customer who is frustrated with..."
💡 Pro Tip
Combine roles with constraints for even better results: "You are a financial advisor who only gives conservative, low-risk recommendations" or "You are a copy editor who ruthlessly cuts unnecessary words."
Technique 3: Use Examples (Few-Shot Prompting)
If you want output in a specific style or format, show the AI what you mean. This technique—called few-shot prompting—is remarkably effective.
Example: Generating Product Names
I need creative names for a new project management app. Here are some examples of the style I like:
• Notion - simple, suggests ideas and concepts
• Slack - short, memorable, implies casual communication
• Asana - unique word, easy to pronounce, no obvious meaning
Generate 10 name ideas following this style. The app focuses on helping small teams stay organised without overwhelming them.
By showing examples, you've communicated more about what you want than paragraphs of description could. The AI understands the pattern: short, memorable, not too literal.
When to use examples:
- You want a specific writing style
- You need consistent formatting across multiple outputs
- The task is unusual or creative
- You've struggled to describe what you want in words
Technique 4: Chain of Thought (Think Step by Step)
For complex problems, asking the AI to think through its reasoning dramatically improves accuracy. This is especially useful for math, logic, analysis, and multi-step tasks.
Without Chain of Thought
"Should our startup raise funding or bootstrap?"
Result: A generic answer that might miss your specific situation
With Chain of Thought
"Should our startup raise funding or bootstrap? Think through this step by step, considering: our burn rate, market timing, competitive landscape, founder goals, and current traction. Then give a recommendation with your reasoning."
Result: Structured analysis that addresses each factor
Phrases that trigger better reasoning:
- "Think step by step"
- "Walk me through your reasoning"
- "Consider the following factors: [list them]"
- "Before answering, analyse the pros and cons"
- "Explain your thought process"
Technique 5: Iterate and Refine
Great prompts rarely come out perfect the first time. The best results come from treating your conversation with AI as a collaboration, not a one-shot query.
The Iteration Process
Round 1: "Write a tagline for my coffee subscription service"
→ AI produces something generic
Round 2: "Make it more playful and mention the freshness aspect"
→ Better, but not quite right
Round 3: "I like option 2 but make it shorter—maximum 5 words"
→ Now you have something usable
Useful refinement phrases:
- "Make it more [adjective]" or "Make it less [adjective]"
- "Keep the [specific part] but change [other part]"
- "Give me 5 variations of option 2"
- "That's good but it's too [long/formal/vague]. Try again with..."
- "Combine the best elements of options 1 and 3"
Common Mistakes to Avoid
Being Too Vague
"Help me with my presentation" → What kind of help? What's the presentation about? Who's the audience? How long is it?
Asking Multiple Unrelated Things
"Write my bio, also what do you think about AI regulation, and can you help with this code?" → Split these into separate prompts for better results.
Not Providing Context
"Is this a good email?" → Good for what purpose? Who's receiving it? What's the relationship? What outcome do you want?
Accepting the First Output
The first response is rarely the best. Push back, ask for variations, request changes. The AI can do better if you guide it.
Overcomplicating Simple Tasks
Sometimes a simple prompt is best. "Summarise this in 3 bullet points" doesn't need elaborate role-playing or examples.
Practical Prompt Templates
Here are ready-to-use templates for common tasks. Copy, customize, and use them as starting points.
📧 Email Writing
Write an email to [recipient/role] about [topic]. Context: [relevant background] Goal: [what you want to achieve] Tone: [formal/casual/friendly/urgent] Length: [short/medium/detailed] Include: [specific points to cover]
📝 Content Creation
Write a [content type] about [topic]. Target audience: [who will read this] Purpose: [inform/persuade/entertain/educate] Key points to cover: [list main points] Tone: [describe the voice] Length: [word count or format] Avoid: [anything to not include]
🔍 Analysis & Feedback
Analyse [thing to analyse] and provide feedback. You are a [relevant expert role]. Focus on: [specific aspects to evaluate] For each issue found: - Explain the problem - Why it matters - How to fix it Be [honest/constructive/detailed].
💡 Brainstorming
Generate [number] ideas for [topic/problem]. Context: [relevant background] Constraints: [any limitations] The ideas should be: [criteria - creative/practical/bold] For each idea, briefly explain why it could work. Include at least one unconventional option.
Model-Specific Tips
Different AI models have different strengths. Here's how to get the best from each:
| Model | Best For | Prompting Tips |
|---|---|---|
| GPT-4 / ChatGPT | General tasks, creative writing, coding | Responds well to system prompts; use custom instructions for persistent context |
| Claude | Long documents, analysis, nuanced discussion | Handles very long context well; be direct about what you want |
| Gemini | Multimodal tasks, Google integration | Excellent with images and documents; leverage its web access |
| Smaller/Fast Models | Simple tasks, high volume | Keep prompts simpler; these struggle with complex multi-step instructions |
Putting It All Together
Let's see how these techniques combine in a real-world example:
Complete Prompt Example
You are an experienced startup mentor who has helped dozens of companies with their pitch decks. I'm preparing to pitch to seed-stage VCs next month.
Review my pitch deck summary below and provide feedback. Focus on: clarity of problem/solution, market size credibility, team slide effectiveness, and overall narrative flow.
For each issue you find, explain: (1) what's wrong, (2) why it matters to investors, and (3) a specific suggestion to fix it.
Be direct and critical—I'd rather hear hard truths now than get rejected later.
[Pitch deck summary would go here]
This prompt uses:
- Role: Experienced startup mentor
- Context: Seed-stage VC pitch, timing
- Task: Review and provide feedback
- Format: Structured feedback with three parts
- Tone: Direct and critical
- Specifics: Exactly what to focus on
🚀 Start Practising Today
Step 1: Take a task you'd normally ask AI to help with
Step 2: Before typing, write down: task, format, audience, tone, and any examples
Step 3: Construct your prompt including those elements
Step 4: Compare the result to what you'd normally get—you'll see the difference immediately
Prompt engineering isn't about memorising magic phrases or gaming the system. It's about communicating clearly with a tool that takes your instructions literally. The more precisely you can articulate what you want, the better your results will be. Start with one technique, practise until it's natural, then add another. Within a few weeks, you'll be getting dramatically better results from every AI interaction.
Stay Updated on AI
Get the latest news and tutorials