I'm Rick. I've been working with AI in learning and development for a while now, and I notice that most people prompt fine, they just don't know why it sometimes works and sometimes doesn't. That's what we're going to fix here.
No textbook, no lists of 47 tips, no AI-guru story. Just the building blocks I use myself every day, in the form I'd also use to explain them to my kids.
By the end of this mini you'll understand why a good prompt is almost always made of the same pieces, how to recognise those pieces, and above all: what to do next time AI gives you back something boring or wrong.
Lesson 1: the building blocks
A good prompt isn't a sentence. It's a stack.
When I explain a prompt, I always compare it to a sandwich. You've got a bottom, you've got a filling, you've got a top. Take one layer out and the thing falls apart. Prompts actually work exactly like that: there are five building blocks you can stack, and the more you consciously lay down, the better the answer gets.
Click the building blocks below to add them one by one. You'll see the prompt build up live on the right, as if you were really typing it into ChatGPT.
Tip: you don't always have to use all five. But if an answer disappoints, there's almost guaranteed to be one you forgot.
Lesson 2: role & examples
Say who the AI is. And show what you mean.
Two things that really make a difference, and that most people skip: a role and an example. Both give the AI something to hold on to.
Role: one sentence, huge effect
"You are an experienced L&D advisor writing for a non-technical HR manager" isn't a trick, it's just practical. It makes the AI make the same choices that L&D advisor would: which words to use, which not, which detail to cut.
I even use this for things that have nothing to do with work. "You are a dietician who understands that I do keto and that my kids eat along." Instantly different recipes. No chia seeds in every dish.
Examples: show, don't tell
If I explain that a text should be "sober and direct", AI half gets it. If I paste in two sentences that already are that, it suddenly gets it completely. That's called few-shot prompting. It's really just: rather one example than ten adjectives.
What I do myself: I sometimes paste in a bit of an earlier text I liked, with one line above it: "write in this style". It really saves me three rounds of steering every time.
Scenario
Your prompt is boring. Which building block is missing?
Situation: you asked ChatGPT for a LinkedIn post about your new e-learning project. You typed:
"Write a LinkedIn post about my new e-learning for prompt engineering."
You get back something that starts with "In the ever-faster-changing world of AI..." and ends with three emojis you'd never use. You think: this isn't me.
Which building block is missing here first?
Finally
What I'd do if I were you
Don't memorise everything. That's nonsense, it'll just wear you out. Do one thing: next time an answer from AI disappoints, stop for a moment. Read your own prompt back. And ask yourself which of those five building blocks you forgot.
Nine times out of ten it's role or example. Just because we're in a hurry and think "it'll get me". But it doesn't get you, it guesses. The more you give it, the less it has to guess.
Three things I'll leave you with
1. Write your prompt as an instruction, not a question. "Write...", "Make...", "Give me...". That works better than "Could you maybe...". Sounds blunt, isn't.
2. Paste in an example if you want a tone. One paragraph in the style you want, with "write in this style" above it. This is the biggest difference between someone who's used AI for two years and someone who's two days in.
3. If the answer is bad, change the prompt, not the AI. Sounds obvious, but most people start a conversation ("no, shorter", "no, different"). Much faster: close the chat, rewrite your prompt, start again.
And if this was useful to you: this is a taster. There's more coming. Prompt engineering is the skill I got the most out of in 2026, and I'm happy to share it, because I notice people get a lot better once they see these five blocks laid out.