Prompting 101: How to Make a Good Prompt

Most people write terrible prompts. They're vague, confusing, and then they wonder why the AI gives them garbage outputs.

Good news: Writing good prompts is a learnable skill. This guide will show you exactly how to do it.

The Problem with Most Prompts

Here's what most people write:

"Write a good email"

"Make this better"

"Help me with my recipe"

These prompts fail because they're missing critical information. The AI has to guess what you want, and it will guess wrong.

The Anatomy of a Good Prompt

A good prompt has five key parts:

1. Role & Context - Tell the AI who it is and what situation it's in. Instead of "Write an email", try "You are a customer success manager writing to a frustrated enterprise client."

2. Clear Objective - State exactly what you want to accomplish. Instead of "Help with recipe", try "Create a 30-minute weeknight dinner recipe using chicken and vegetables."

3. Specific Constraints - Set boundaries and requirements. Instead of "Keep it short", try "Maximum 5 sentences, professional tone, include next steps."

4. Output Format - Define how you want the response structured. Instead of no format specified, try "Format as bullet points with headers for each section." Note: Format can be anything verifiable - JSON, bullet points, numbered lists, etc.

5. Examples (When Needed) - Show what good looks like. Instead of "Make it sound friendly", try providing a tone example: 'Thanks for reaching out! I understand your concern and I'm here to help.'

Writing a Specification: The Do's and Don'ts Method

A good prompt needs a specification - a clear definition of what the AI should always do AND what it should never do. Both are equally important.

What is a Specification?

A specification is a contract between you and the AI. It defines:

• What to always do (positive constraints)
• What to never do (negative constraints)

These work together to create precise boundaries for the AI's behavior.

Pro tip: It's often easier to start by identifying what NOT to do. When you know what failures look like, it becomes clearer what success should be.

The Key Rule: Make It Verifiable

Every constraint in your specification must be easily verifiable. If you can't quickly check whether the AI followed a rule, that rule is too vague for effective evaluation.

Bad (Not Verifiable):

• "Write engagingly"
• "Be helpful"
• "Sound natural"

Good (Verifiable):

• "Include exactly 3 bullet points"
• "Never exceed 100 words"
• "Always end with a question"

Example Specification for a Recipe Generator:

Always Do:

• Include exact cooking temperatures and times
• List ingredients in order of use
• Specify serving sizes
• Use common supermarket ingredients

Never Do:

• Never include nuts or shellfish (allergy concerns)
• Never use vague terms like "cook until done"
• Never exceed 10 ingredients total
• Never require specialty equipment

Why Both Matter

Including both do's and don'ts in your prompt:

• Prevents common failures before they happen
• Reduces ambiguity by defining boundaries from both sides
• Improves consistency across multiple uses
• Makes debugging easier when outputs go wrong

A Real Example

Let's transform a bad prompt into a good one:

Before:

"Write a summary"

After:

You are a business analyst preparing an executive briefing.

Summarize the Q3 sales report for the CEO.

Always:

• Include revenue change vs Q2 (%)
• List exactly 3 top products
• Identify 1-3 Q4 risks
• Use exact numbers/percentages
• Start with 1-sentence overview
• Format as bullet points

Never:

• Exceed 5 bullet points total
• Use vague terms ("many", "some")
• Include operational details
• Exceed 150 words total

Quick Checklist

Common Mistakes to Avoid

The Golden Rule

If a human intern would need to ask clarifying questions, your prompt needs work.

Good prompts eliminate ambiguity. They're so clear that there's really only one way to interpret them.

Start Simple, Then Iterate

You don't need to write perfect prompts on the first try. Start with something basic, see what goes wrong, then add constraints to fix it.

Key Takeaways

1. Be specific - Replace vague words with concrete requirements
2. Tell, don't just show - Clear specifications beat vague examples
3. Think constraints - What should it NOT do?
4. Format matters - Define the structure you want
5. Test and refine - Your first prompt is a rough draft