Prompt Engineering for Normal People

"Prompt engineering" sounds like something you'd need a degree for. It is not. It's just knowing how to ask AI tools for things in a way that gets useful results. That's it. That's the whole field.

I know because I've read the research papers and the Twitter threads and the online courses, and I can tell you that about 90% of what works comes down to being specific, giving context, and showing examples. The other 10% is tricks and techniques that help in specific situations.

You don't need a course. You need this article and about twenty minutes of practice.

The single most important rule

Be specific.

"Write me an email" produces generic slop. "Write a 3-paragraph email to my manager requesting a deadline extension on the Henderson project from March 15 to March 29. Tone should be professional but not formal. Mention that the client changed the scope last week as the reason." produces something you can actually use.

That's it. That's the rule. Specificity is the difference between useless output and useful output. Everything else is detail.

Let me break down what "specific" means.

The five things to include in every prompt

1. What you want (the task)

"Write an email" vs "Write a reply to this email declining the meeting but suggesting an alternative time."

"Summarise this document" vs "Summarise this document in five bullet points, each no longer than one sentence, focusing on financial implications."

"Help me with this data" vs "Analyse this CSV and tell me which product category had the highest growth rate between Q1 and Q4."

2. Who it's for (the audience)

"Explain machine learning" vs "Explain machine learning to a group of HR managers who have no technical background."

"Write a project update" vs "Write a project update for the board of directors. They care about budget, timeline, and risks. They don't care about technical details."

The same content, framed for different audiences, looks completely different. AI needs to know who's reading.

3. How it should sound (the tone)

"Professional." "Casual." "Formal." "Friendly but not over-familiar." "Direct and concise." "Empathetic."

AI defaults to a particular voice that sounds like... AI. Specifying tone pushes it towards something that sounds like you. Or at least like a version of you that's having a better writing day than usual.

4. How long it should be (the format)

"Keep it under 200 words." "Maximum three paragraphs." "One page." "A bullet-point list of no more than ten items."

Without length constraints, AI will ramble. It will give you a 500-word answer when 50 words would do. Tell it how much you want and it'll (usually) respect that.

5. What to avoid (the constraints)

"Don't use jargon." "Don't include an introduction or conclusion." "Don't use the word 'synergy' or I will close this tab." "British English spelling throughout."

Telling AI what NOT to do is sometimes more effective than telling it what to do. If you've noticed patterns in AI output that annoy you, name them and tell it to stop.

Techniques that actually help

Give it a role

"You are a financial analyst reviewing this data." "You are an experienced HR manager drafting a policy." "Act as a critical editor reviewing this document for weaknesses."

This changes the perspective and expertise the AI brings to the task. A "financial analyst" will focus on different things than a "marketing manager" when looking at the same data.

Show it an example

"Here's an example of the format I want: [paste example]. Now create something similar for [your topic]."

Examples are incredibly powerful. AI learns patterns from examples faster than from descriptions. If you have a report format you like, a writing style you want to match, or a template you want to follow, show it.

Ask it to think step by step

For complex problems, "think through this step by step" or "explain your reasoning" produces better results. This isn't magic. It forces the AI to work through the problem rather than jumping to an answer.

"What's the best marketing strategy for this product?" will give you a generic answer. "Think step by step: What are the key demographics for this product? What channels reach them? What messaging would resonate? Now recommend a strategy." will give you something much more useful.

Iterate

Your first prompt won't always produce the perfect result. That's fine. Treat it as a conversation.

"This is good but too formal. Make it more conversational." "I like the structure but point three is wrong. The actual situation is [X]. Rewrite with that correction." "This is too long. Cut it in half and keep only the most important points."

The people who get the most from AI aren't the ones who write perfect prompts. They're the ones who iterate. First draft, feedback, second draft, more feedback, done.

Common mistakes

Being too vague. "Help me with work stuff" tells the AI nothing. Be specific about the task, the context, and the desired output.

Not giving context. The AI doesn't know your company, your industry, your audience, or your situation unless you tell it. The more relevant context you provide, the better the output. This doesn't mean writing a novel. A sentence of context is often enough.

Accepting the first output. If it's not right, say so. Tell the AI what's wrong and ask it to fix it. Most people give up after one attempt and conclude "AI isn't useful." It's like asking a colleague to draft something, not giving feedback, and then complaining it's not what you wanted.

Using AI for things that don't need AI. If you can write the email faster than you can write the prompt... just write the email. AI is a tool, not a religion. Some tasks are faster without it.

For more on which AI tool to use and how to use them safely at work, we've got separate guides.

The one thing to do today: take your next AI prompt, whatever it is, and add two pieces of information you wouldn't normally include: who the audience is and what tone you want. Compare the result to what you'd normally get. That comparison will teach you more about prompt engineering than any course ever could.