Let's be honest: you're probably terrible at prompting AI. And before you get defensive, so was I.
You type "write me a resume" or "create an apology email," and what do you get? Generic garbage that sounds like it was written by a robot having a bad day.
But here's the thing: the problem isn't AI. It's you. (Don't worry, I'm here to fix that.)
What you'll learn in this article
- What AI prompting actually is (and why most people get it wrong)
- The 5 core techniques for writing better AI prompts
- How to reduce hallucinations and generic outputs
- Advanced prompting methods like Chain of Thought and Tree of Thoughts
- Real-world examples of bad vs good AI prompts
Why Most AI Prompts Fail (And Why You Get Generic Results)
Most people fundamentally misunderstand what prompting actually is. It feels like talking to a human, but you need to remember: you're talking to a computer.
According to Dr. Jules White from Vanderbilt University, a prompt isn't just a question—it's a call to action. You're not asking the AI anything. You're programming it with words.
Think of AI as a super-advanced autocomplete system. It's a prediction engine that completes patterns. If you give it a vague pattern, it guesses vaguely. Give it a focused pattern? You hack the probability and get gold.
The Foundation: 5 Core AI Prompting Techniques That Actually Work
1. Persona - Give AI a Role
Never let AI be "nobody." Always tell it WHO it is.
❌ Bad: "Write an apology email about a server outage"
✅ Good: "You're a senior site reliability engineer at Cloudflare with 10 years of experience. Write an apology email about a server outage."
By setting a specific persona, the output immediately becomes more professional, uses correct terminology, and adopts the appropriate tone. The AI narrows its focus and stops being generic.
2. Context - Always Be Contexting (ABC)
AI is a people-pleaser. If you don't give it facts, it will hallucinate and make stuff up just to complete the pattern.
The solution? Provide context. Give it the necessary details so it doesn't have to guess.
❌ Bad: "Explain what happened during the outage"
✅ Good: "Here are the FACTS about today's outage:
- We made a database permissions change at 02:47 UTC
- This caused duplicate metadata entries
- Services were down for 47 minutes
- 3,847 customers were affected"
More context = fewer hallucinations. Simple as that.
3. Permission to Fail - Let AI Say "I Don't Know"
This is crucial for reducing hallucinations. Explicitly tell the AI it can admit when it doesn't have the answer.
Add this to your prompts: "If you don't have enough information to answer accurately, say 'I don't know' instead of guessing."
4. Output Format - Specify the Structure
Never assume AI knows what format you want. Be specific.
Tell it:
- How long should it be? (300 words? 5 paragraphs?)
- What structure? (Bullet points? JSON? Markdown?)
- What tone? (Professional? Casual? Technical?)
- Any specific requirements? (Include a timeline? Add cost estimates?)
Example: "Write a 500-word explanation in 3 sections with headers. Use a conversational but professional tone. Include 2 concrete examples."
5. Few-Shot Prompting - Show Examples
Want AI to match a specific style? Show it what "good" looks like.
Provide 2-3 examples of the exact output you want. This is insanely effective.
Example:
Here are examples of the style I want:
Example 1: [paste your example]
Example 2: [paste another example]
Now create something similar for [your topic].
Advanced AI Prompting Techniques (Chain of Thought, Tree of Thoughts)
Once you've mastered the basics, these advanced methods will blow your mind.
Chain of Thought (CoT) - Make AI Think Step-by-Step
Don't let AI jump to conclusions. Force it to show its work.
Add this magical phrase: "Think step-by-step" or "Let's approach this systematically"
Example:*"Calculate the ROI of this marketing campaign. Think step-by-step:
- First, identify all costs
- Then, calculate total revenue
- Finally, compute the ROI percentage"*
This dramatically improves reasoning and reduces errors.
Tree of Thoughts (ToT) - Explore Multiple Paths
For complex problems, make AI explore different approaches before settling on an answer.
Example:*"I need to reduce our cloud costs by 30%. Generate 3 completely different strategies:
- Strategy A: Focus on resource optimization
- Strategy B: Focus on architectural changes
- Strategy C: Focus on provider negotiations
For each strategy, list pros, cons, and estimated savings. Then recommend which approach is best and why."*
This forces the AI to explore the solution space instead of taking the first path it thinks of.
The Playoff Method - Let AI Compete Against Itself
Want the absolute best output? Make AI generate multiple versions and then judge them.
Example:*"Generate 3 different email subject lines for our product launch.
Then, evaluate each one on:
- Click-through potential (1-10)
- Clarity (1-10)
- Emotional appeal (1-10)
Finally, pick the winner and explain why."*
Self-Correction Loop
Make AI review its own work:
*"After you write the email, review it and identify:
- Any unclear statements
- Any missing information
- Any tone issues
Then provide an improved version."*
The Meta-Skill That Changes Everything
Here's what the experts don't tell you enough: All these techniques only work if your thinking is clear.
If your thinking is messy, your prompts will be messy, and you'll get messy outputs. No amount of fancy prompting will fix unclear thinking.
Before you type into the chat box, sit down and describe exactly what you want to accomplish. If you can explain it clearly to a human, you can explain it to AI.
Think first. Prompt second. That's the true secret.
Real-World Example: From Garbage to Gold
Let's see these principles in action.
❌ Terrible Prompt: "Write an email about the outage"
Result: Generic, soulless corporate speak that helps nobody.
✅ Expert Prompt:
"You're a senior site reliability engineer at Cloudflare who values transparency and clear communication.
Context: We had a service outage today caused by a database configuration change. Here are the facts:
- Change made at 02:47 UTC
- Duplicate metadata entries caused cascading failures
- 47 minutes of downtime
- 3,847 customers affected
- Issue resolved by rolling back the change
Task: Write an apology email to affected customers.
Format:
- Keep it under 250 words
- Start with a clear apology
- Explain what happened in simple terms (no excessive jargon)
- State what we're doing to prevent this
- End with contact info for questions
Tone: Professional but human. Acknowledge the impact on their business. Don't be overly technical.
Permission to fail: If any facts are unclear, ask me before making assumptions."
Result: A clear, empathetic, professional email that actually helps customers understand what happened.
Your Action Plan: Start Prompting Like a Pro
Here's your checklist for every important prompt:
✅ Persona: Who is the AI?
✅ Context: What facts does it need?
✅ Format: What structure do you want?
✅ Examples: Can you show what "good" looks like?
✅ Permission to fail: Can it say "I don't know"?
✅ Reasoning method: Should it use CoT? ToT? Playoff?
The Bottom Line
Stop treating AI like a magic oracle that reads your mind. It can't.
Start treating it like an incredibly capable assistant who needs clear, detailed instructions.
You're not asking questions. You're starting patterns. You're not chatting. You're programming with words.
Master these techniques, and you'll go from getting garbage outputs to creating work that makes people say, "Wait... AI wrote this?"
The technology is here. The question is: are you ready to get good at using it?
Pro Tip: Start small. Pick ONE technique from this list and practice it for a week. Master personas, then add context, then work your way up to the advanced stuff. Don't try to use everything at once.
Now stop reading and go prompt something. You've got this.
If you want to read more on AI check out our articles here, to continue the conversation start discussion in our AI category.