The Art of Prompting

By the end of this lesson you’ll be able to turn a vague request into a prompt that gets you something you can actually use. You’ll learn a simple four-part structure you can reuse forever, and you’ll practice it on two real tasks. No memorizing magic words — just a way of asking that works.

The idea in plain English

A prompt is just what you type into an AI tool like ChatGPT, Claude, or Gemini. That’s it. There’s no secret language, and anyone who tells you there’s a list of “power words” that unlock better results is selling you something.

Here’s the thing that trips people up: these tools are very good at doing what you asked and very bad at guessing what you meant. If you type “write a bio,” the tool has no idea whether you want three sentences for LinkedIn or two paragraphs for a wedding program. So it splits the difference and gives you something bland and generic. Then you conclude the AI isn’t very good.

It wasn’t the AI. There genuinely wasn’t enough information in the question to answer it well.

Think of it like handing off a task to a brand new, extremely fast assistant who has read most of the internet but has never met you. They don’t know your job, your customers, your writing style, or what you’re going to do with the thing they make. Anything they don’t know, they’ll fill in with the most average possible answer. Prompting well is really just the habit of saying the things out loud that you were keeping in your head.

How it actually works

Almost every good prompt does four things. You don’t need all four every time, but when a response comes back disappointing, one of these is usually missing.

1. Role — who should the AI be? “You are a high school biology teacher” and “You are a research scientist” will explain photosynthesis very differently. This one line pulls the whole answer toward the vocabulary and depth you want.

2. Task — what exactly do you want made? Be specific about the verb and the thing. Not “help me with my email” but “rewrite this email so it’s shorter and less apologetic.”

3. Context — what does it need to know? Who’s the audience, what’s the situation, what’s already been tried. This is where beginners leave the most on the table, because context feels obvious to you. It isn’t obvious to the tool.

4. Format — what should it look like? Length, structure, tone. “Three bullet points, under 15 words each.” “A table with two columns.” “Plain language, no business jargon, around 200 words.”

Two more ideas make a real difference:

Show an example. If you have a version you liked — an old email, a competitor’s product description, a paragraph in your voice — paste it in and say “match this style.” One good example beats five adjectives. You’ll see this called a few-shot prompt, which just means “here are some samples, do more like these.”

It’s a conversation, not a vending machine. The biggest beginner upgrade isn’t a better first prompt — it’s replying. The tool remembers what you both just said, so “too formal, and cut it in half” works perfectly. Most people type one prompt, feel let down, and close the tab. Your second and third messages are where the good version shows up.

One caution before you start: AI tools sometimes state wrong things with total confidence. This is often called hallucination. Prompting well reduces it but doesn’t eliminate it, so anything factual — a statistic, a date, a legal or medical detail, a quote — you check yourself before you use it.

Try this now

Open ChatGPT, Claude, or Gemini. Any of them, the free version is fine. We’re going to do the same task twice so you can see the difference with your own eyes.

1. First, type a deliberately vague prompt so you have something to compare against:

   Write a post about my business.

   Read what comes back. It’s probably grammatically fine and completely forgettable — full of phrases like “we’re passionate about serving our community.” Notice it invented details about a business it knows nothing about.

2. Now start a new chat and use all four parts. Swap in your own details wherever you see brackets — and delete the brackets:

   You are a copywriter who writes plain, friendly social posts for small local businesses.
   
   Write one Facebook post announcing that [Miller's Bakery] is now open on Sundays from 8am to 1pm.
   
   Context: we're a small family bakery in [Springfield]. Most of our customers are regulars who walk over from the neighborhood. Our Sunday special is cinnamon rolls, and we usually sell out by 11. We're warm and a little bit funny, never salesy.
   
   Format: 80 words or less, no hashtags, no emojis, one clear line at the end telling people to come early.

   Compare the two. Same tool, same amount of typing effort, completely different result.

3. Now practice the part most people skip: replying instead of starting over. Type a correction like Shorter, and lose the exclamation points or Make the last line sound less like an ad. Do this two or three times. Notice you never had to re-explain the bakery.

4. Ask for options instead of one answer. Try: Give me 3 different versions — one funny, one warm, one very short. Picking from three is faster and easier than judging one in a vacuum, and it shows you the range available.

5. Now try it on something from your actual week. Copy this template and fill it in:

   You are an experienced [job title, e.g. office manager].
   
   I need you to [the specific thing you want, e.g. write a polite email declining a vendor's meeting request].
   
   Here's what you need to know: [who it's for, what's happened so far, anything that must or must not be included].
   
   Here's the raw material: [paste your notes, the email you're replying to, or your rough draft].
   
   Format: [length, tone, structure — e.g. under 120 words, friendly but final, no bullet points].

6. One more trick worth building into your habits — turn the questioning around before a big task:

   Before you write anything, ask me up to 5 questions that would help you do this well.

   Add that line to the end of any prompt for something important. The questions it asks will often be things you hadn’t thought about, which is useful information all by itself.

Common mistakes

• Being polite instead of being specific. “Could you please help me with a little something for work?” wastes the whole message. The tool has no feelings to spare, so spend your words on details. Fix: lead with the concrete task, then the context.
• Asking for everything in one prompt. “Write my whole website” gets you five mediocre pages. Fix: break it into pieces and do them one at a time — homepage headline first, then the About page, then the service descriptions. Smaller asks get better answers.
• Starting a new chat when the answer isn’t right. This throws away everything the tool learned about your situation. Fix: stay in the same conversation and say what’s wrong. “Too long.” “Wrong audience — this is for parents, not teachers.” “I liked the second paragraph, build on that one.”
• Not saying who it’s for. Audience changes vocabulary, length, and examples more than almost anything else. Fix: add one line — “this is for retirees with no tech background” or “this is for my accountant.”
• Accepting the first draft because it sounds confident. Fluent writing feels correct even when it isn’t, and made-up facts read exactly like real ones. Fix: treat every draft as a first draft, and verify anything you’d be embarrassed to get wrong.
• Describing your voice instead of showing it. “Make it sound like me” means nothing to a tool that’s never read your writing. Fix: paste two or three things you’ve written and say “match this style.”

Key takeaways

• A prompt is just a request. Clear requests get useful answers; vague requests get generic ones.
• Cover four things when it matters: role, task, context, format.
• Your follow-up messages are where the good version appears — reply and refine, don’t restart.
• Show an example of what you want instead of describing it with adjectives.
• Ask for three options, and add “ask me questions first” before anything important.
• Confident-sounding output can still be wrong. Check anything factual before you use it.