Prompt Engineering 101: How to Get Better Results from AI
Your Prompts Are Probably Too Vague
The single biggest reason people get disappointing results from AI is that they write vague, underspecified prompts. "Write me a blog post about marketing" will give you a generic, forgettable post. But with proper prompt engineering, the same AI can produce genuinely excellent content.
The CRISP Framework
I teach my clients a framework I call CRISP. It stands for Context, Role, Instructions, Specifics, and Parameters.
Context
Give the AI background information it needs. The more context you provide, the more relevant the output.
Bad: "Write an email to a client." Good: "I run a 10-person accounting firm. We just finished the annual audit for a mid-size manufacturing client. They had three significant findings that need attention."
Role
Tell the AI who it should be. This shapes the tone, vocabulary, and perspective.
Bad: "Explain our refund policy." Good: "You are a friendly but professional customer service representative for a boutique clothing store. Explain our refund policy to a customer who received a damaged item."
Instructions
Be explicit about what you want the AI to do. Break complex tasks into steps.
Bad: "Help me with my business plan." Good: "Review the following business plan summary and provide feedback on three things: 1) Is the market size estimate realistic? 2) Are there gaps in the competitive analysis? 3) What risks are not addressed?"
Specifics
Include concrete details about format, length, tone, and audience.
- Specify word count or length
- Name the target audience
- Define the tone (formal, conversational, technical)
- Request specific formatting (bullet points, headers, tables)
- Provide examples of what good output looks like
Parameters
Set boundaries on what the AI should and should not do.
- "Do not use jargon. Write at an 8th-grade reading level."
- "If you are not sure about a fact, say so rather than guessing."
- "Focus only on strategies that cost under $500 to implement."
Advanced Techniques
Chain of Thought Prompting
Ask the AI to think through its reasoning step by step. This dramatically improves accuracy for analytical tasks.
Example: "Before giving your recommendation, first analyze the pros and cons of each option. Then explain your reasoning. Finally, give your recommendation."
Few-Shot Examples
Show the AI what you want by providing examples of good output.
Example: "Here are two examples of the email tone I want: [Example 1] [Example 2]. Now write a similar email for this situation: [situation]."
Iterative Refinement
Do not try to get perfect output in one prompt. Start with a rough version and refine:
- Generate the first draft
- Ask the AI to identify weaknesses in its own output
- Ask it to improve specific sections
- Request a final polished version
System Prompts and Custom Instructions
Most AI tools let you set persistent instructions. Use these to encode your brand voice, formatting preferences, and common constraints so you do not have to repeat them every time.
Real-World Examples
For a real estate agent writing listing descriptions: "You are an experienced real estate copywriter. Write a listing description for a 3-bedroom, 2-bathroom ranch home in suburban Denver. The home was built in 1998, recently renovated with a modern kitchen and new hardwood floors. The lot is 0.25 acres with a fenced backyard. Price is $485,000. Write in an engaging but honest tone. Highlight the renovation and outdoor space. Keep it under 200 words. Do not use cliches like 'move-in ready' or 'dream home.'"
For a restaurant creating menu descriptions: "Write menu descriptions for the following 5 dishes. Each description should be 15 to 25 words. Tone should be appetizing but not pretentious. Mention key ingredients and cooking methods. Do not use the word 'delicious' or 'amazing.' Here are the dishes: [list]."
The 80/20 Rule of Prompting
You do not need to be a prompt engineering expert to get great results. These five principles will get you 80% of the way there:
- Be specific about what you want
- Provide relevant context
- Give examples when possible
- Specify format and length
- Iterate rather than expecting perfection on the first try
Master these, and you will get dramatically better results from any AI tool you use.