Prompt Engineering
The skill of writing inputs that get great outputs. The most valuable AI skill you can develop.
Why prompts matter
The same AI model can produce brilliant output or useless noise — the difference is almost entirely your prompt.
Think of it this way: the model is fixed. You can't change it. What you can change is the instruction you give it. Learning to write better prompts is learning to get more out of the tool you already have.
The four elements of a strong prompt
Every effective prompt contains four things:
1. Role — Who should the AI be? "You are a senior DTC copywriter with 10 years of experience" performs better than a blank start.
2. Context — What does it need to know? Your product, your audience, your constraints, any relevant background.
3. Task — What exactly should it do? Be specific. "Write a product description" is vague. "Write a 100-word product description for a $47 AI scanner tool, targeting first-time dropshippers who want to find winning products fast" is specific.
4. Format — How should the output look? Length, structure, tone, whether you want bullet points or paragraphs.
You are a [specific role with relevant experience]. Context: [everything the AI needs to know — product, audience, constraints, examples] Task: [exactly what you want it to do, as specifically as possible] Format: [length, structure, tone, any specific requirements]
Role prompting
Setting a role is one of the highest-leverage things you can do in a prompt. It frames every response that follows.
Instead of: "Write a business plan"
Try: "You are a venture capitalist who has evaluated 500+ startup pitches. Write a business plan for [company] that addresses the questions VCs actually care about."
The role tells the model what expertise to draw on, what perspective to take, and what quality bar to aim for.
Giving examples (few-shot prompting)
If you want the AI to match a specific style or format, show it an example. This is called few-shot prompting.
"Write a tweet about [topic] in this style: [paste an example tweet you like]"
Examples are often more powerful than descriptions. Instead of trying to describe "a casual, direct tone with no buzzwords," paste two examples of that tone.
Iteration is the workflow
Don't expect the first output to be final. Prompting is a conversation.
If the output is off: tell it why. "That's too formal — make it more conversational." "The second paragraph is too long." "Great structure but the hook doesn't grab me — try 5 different opening lines."
The best prompting sessions look like a back-and-forth with a very capable collaborator — you're steering, it's producing.
Common mistakes
Too vague — "Write me something good" gives you something mediocre. Specificity is everything.
No format specified — If you don't say you want bullet points, you'll get paragraphs. If you don't say 100 words, you might get 500.
Not iterating — One prompt, take it or leave it. Most good outputs come after 2-3 rounds of refinement.
Treating it like a search engine — One-line questions work for Google. They rarely produce the best output from an LLM.
Role + context + task + format. This structure turns an average prompt into a great one. Iterate from there.