The #1 Mistake Everyone Makes
You type: "Write me a marketing email."
The AI gives you a generic, forgettable email that sounds like every other AI-generated email on the planet.
So you think: "AI can't write good marketing emails."
Wrong. AI can write excellent marketing emails. You asked the wrong question.
The difference between a useless AI response and a brilliant one is almost never the AI. It's the prompt. And the difference between a bad prompt and a great one comes down to 7 learnable principles.
This guide doesn't give you prompts to copy. It teaches you the principles behind every good prompt — so you can write your own, for any situation, for the rest of your life.
Why This Matters
A person who memorizes 50 prompts can solve 50 problems. A person who understands 7 principles can solve any problem. This guide teaches the principles.
Specificity Beats Length
The most common advice is "be more specific." But what does that actually mean? It means answering three questions the AI can't answer itself:
The Three Questions
WHO is this for? (audience, expertise level, context)
WHAT exactly do you want? (format, length, style, tone)
WHY does it matter? (goal, use case, what success looks like)
Notice the specific version isn't just longer — it's more constrained. "150 words" is a constraint. "Casual and direct, not salesy" is a constraint. "One specific example" is a constraint. Each constraint eliminates thousands of possible bad outputs.
The Rule of Thumb
If your prompt is shorter than the response you want, it's probably too vague. Invest 30 seconds in the prompt to save 10 minutes of editing.
Structure Controls Output
The format of your prompt directly controls the format of the response. If you want structured output, use a structured prompt.
The Pattern
Numbered lists in your prompt → numbered lists in the response.
Tables in your prompt → tables in the response.
Headers in your prompt → organized sections in the response.
The AI mirrors your structure.
1. TOP 3 BENEFITS (one sentence each, with supporting data)
2. TOP 3 RISKS (one sentence each, with mitigation strategy)
3. DECISION FRAMEWORK: 5 yes/no questions the CEO should ask
4. RECOMMENDATION: one paragraph, clear stance
Advanced move: If you want the AI to output JSON, CSV, or any specific data format, include a single example of the format in your prompt. The AI will match it exactly.
Constraints Create Quality
This is counterintuitive: giving the AI fewer options produces better output. Constraints force creativity and eliminate generic responses.
Useful Constraints
Length: "in 3 sentences" / "under 100 words" / "one paragraph"
Exclusion: "don't use jargon" / "no clichés" / "don't say 'leverage' or 'synergy'"
Difficulty: "explain so a 12-year-old would understand"
Format: "as bullet points" / "as a table" / "as a conversation"
Perspective: "argue against your own recommendation" / "from the customer's POV"
The most powerful constraint is exclusion. Telling the AI what NOT to do is often more effective than telling it what to do, because it eliminates the AI's default patterns.
Examples Are Worth 1000 Words
If you can show the AI one example of what you want, it will produce dramatically better output than any amount of description.
The Few-Shot Pattern
Give 1-3 examples of the input/output pair you want. The AI will extrapolate the pattern and apply it to new inputs. This works for any task: writing style, data formatting, classification, analysis structure.
Feature: 5G connectivity
Benefit: Download a full movie in 30 seconds. Stream without buffering. Video calls that don't freeze.
Now do these:
- Fast processor
- 12-hour battery
- Lightweight design
Pro Tip: Use Your Own Best Work
If you've written something you liked before (an email, a report, a post), paste it as an example and say "write another one in this style about [new topic]." The AI clones your voice better than any "tone" instruction.
Audience Changes Everything
The same information explained to different audiences produces completely different outputs. Specifying your audience is the highest-leverage single change you can make to any prompt.
Audience Dimensions
Expertise: beginner / intermediate / expert
Role: CEO / developer / student / customer
Emotional state: skeptical / excited / frustrated / busy
Goal: learn / decide / buy / fix a problem
Chain Steps, Don't Stack Asks
One complex prompt produces worse results than a sequence of simple prompts. Break big tasks into steps.
The Chaining Pattern
Step 1: Generate raw material (brainstorm, research, outline)
Step 2: Evaluate and select (pick the best ideas, critique the outline)
Step 3: Refine and polish (write the final version from the selected material)
Each step uses the output of the previous step as input.
Prompt 2: "Of those 10, pick the 5 most surprising. For each, give me a real-world example and one sentence of evidence."
Prompt 3: "Write a 600-word blog post using those 5 tips. Hook: start with a counterintuitive claim. Structure: tip + example + why it works. End with a single actionable takeaway."
Meta-Prompting: Make AI Write Prompts
The most advanced technique: ask the AI to write the prompt for you. This works because the AI knows what information it needs to produce good output.
The Meta-Prompt
"I want to [goal]. Before you do anything, ask me the 5 most important questions you need answered to do this well. Then use my answers to produce the output."
This technique is especially powerful when you don't know enough about a topic to write a good prompt. Instead of guessing what details to include, you let the AI tell you what it needs.
Level 2: Make AI Improve Your Prompts
Paste a prompt you've written and ask: "Improve this prompt using best practices for prompt engineering. Explain each change you made and why." You'll learn the principles by seeing them applied to your own work.
The Complete Cheat Sheet
7 Principles at a Glance
| Principle | What to Do | Key Phrase |
|---|---|---|
| 1. Specificity | Answer WHO, WHAT, WHY | "For [audience], write [format] that [goal]" |
| 2. Structure | Dictate the output format | "Format: 1. [section] 2. [section] 3. [section]" |
| 3. Constraints | Limit length, exclude defaults | "In 3 sentences. Don't use [cliché]." |
| 4. Examples | Show one input/output pair | "Follow this format exactly: [example]" |
| 5. Audience | Specify expertise + goal + state | "Explain to a [role] who is [state] and wants to [goal]" |
| 6. Chaining | Break into generate → evaluate → refine | "First brainstorm 10, then pick 3, then write from those" |
| 7. Meta-prompt | Ask AI what it needs to know | "Before starting, ask me the 5 things you need to know" |
Power Phrases (copy-paste these into any prompt)
| Phrase | What It Does |
|---|---|
| "Don't be nice. I need the truth." | Disables the AI's politeness filter |
| "Rate each item FATAL / SERIOUS / MINOR" | Forces severity-based prioritization |
| "What would you ask if you were the expert?" | Surfaces questions you didn't think of |
| "Argue against your own recommendation" | Finds weaknesses in the AI's own logic |
| "Explain why you chose this over alternatives" | Makes reasoning explicit and auditable |
| "If you're unsure about anything, flag it" | Reduces hallucination by making uncertainty safe |
| "Don't rewrite it — mark problems and suggest fixes" | Preserves your voice while getting feedback |
Building Your Own Prompts
Now that you know the 7 principles, here's a framework for building any prompt from scratch:
The 5-Line Prompt Template
Line 1 — Role + Context: "You are a [role] helping [audience] with [situation]."
Line 2 — Task: "I need you to [specific action] about [topic]."
Line 3 — Format: "Format: [structure]. Length: [constraint]."
Line 4 — Quality bar: "Don't [common AI default]. Do [what I actually want]."
Line 5 — Evaluation: "After you finish, rate your confidence 1-10 and explain any assumptions."
You don't need all 5 lines for every prompt. But if you're not getting good results, this template tells you exactly what's missing. Most bad prompts are missing lines 3 and 4 — they don't specify format or quality constraints.
The goal isn't to memorize prompts. It's to internalize these principles until they're automatic. Once you can look at any AI output and think "this is vague because I didn't constrain the format" or "this is generic because I didn't specify the audience" — you've mastered prompt engineering.
The Final Test
Take your worst recent AI interaction. Rewrite the prompt using these 7 principles. Compare the outputs. That's the difference this guide makes.