A prompt is just the instructions you give an AI, the message you type in. There's no trick to it. Good prompting is plain, clear delegation.
The whole idea in one analogy: talking to an AI is like handing a task to a brilliant new contractor on their first day. They're sharp, fast, and capable, but they can't read your mind, they've never seen your business, and they only know what you put in the message. Tell them exactly what you want and they nail it. Be vague and they guess.
There's no secret incantation and no prize for length. Just a handful of moves that reliably turn a guessy answer into a good one. We'll walk through them, then you'll run a live upgrader on your own prompt.
Two beliefs send people down the wrong path. Let's kill both, plainly.
What people think prompting is.
What actually moves the needle.
One quick distinction. A system prompt is the standing instructions that apply to every message: "You are our support agent. Always answer in Spanish. Never quote a price." Your message (the user prompt) is the specific ask of the moment: "Where's order #4182?" Set the role and rules once in the system prompt; make the request in the message.
Not twenty tricks. Six. Reach for them like a checklist. You won't need all six every time; you'll feel which ones a prompt is missing.
Who's it for, what you're trying to do, any constraints. The model only knows what you tell it. Vague in, vague out.
"Few-shot" just means giving one or two examples of the output you want. One good example beats a paragraph describing the output.
"You are a no-nonsense bookkeeper" sets the vocabulary, tone, and judgment it answers with. Cheap to add, surprisingly effective.
For reasoning or analysis, "think it through step by step" lets it work in stages, which measurably improves accuracy on those tasks.
Bullets, a table, JSON, a max length, a tone. Tell it the shape you want instead of hoping it guesses right.
Treat answer one as a draft. "More concise." "Less salesy." "Add the price column." Refining is the move pros use most.
Type a rough prompt. A simple rule-based checker flags which of the six moves are missing, then scaffolds an upgraded version for you to copy and edit.
🔒 This runs entirely in your browser. It's a teaching linter built from plain rules and templates. Nothing is sent anywhere, no AI is called, no API, no key.
write a follow-up email
You're a friendly plumber. Write a 4-sentence follow-up to Dana, who got a quote Tuesday but hasn't booked. Warm, no pressure, one clear next step.
Sample answers are illustrative, written for this lesson, not generated live.
💡 The same six moves are exactly how you brief an AI agent to run a real task. See them doing live work in What is an AI agent?
You can spot the myth, you've got the six moves, and you know the first answer is a draft. One honest caveat before you go.
Writing one good prompt is easy. The hard part is the system around it: the prompts, data, and guardrails that produce the right answer reliably, day after day, without you babysitting it. That's what we build.
Next: what an AI agent actually is →
Day 7 of 30: free guides & working AI kits for small business.