The text box that runs the show
A prompt is simply the text you send to an AI model. That's it. No secret handshake, no incantation. You type words, the model reads them, and it generates a response by predicting what should come next, one small chunk of text (a token) at a time.
But here's the twist that makes prompting a real skill: the model responds to exactly what you wrote, not what you meant. It can't read your mind, peek at your screen, or remember that thing you were thinking about in the shower.
The genie with a law degree
Imagine a genie who grants wishes, but this genie went to law school. It fulfills your wish precisely as worded, no more, no less.
Wish for "a summary of this article" and you'll get one. Maybe it's three paragraphs when you needed three bullets. Maybe it's written for experts when you needed it for your grandmother. The genie didn't fail you. Your wish just had loopholes.
Prompt engineering is the art of writing wishes with no loopholes.
The genie isn't malicious. It's literal. "Make me a sandwich" gets you a sandwich. Which sandwich? The genie picks the statistically most likely sandwich. Hope you like turkey.
The barista test
Think about ordering coffee. When you tell a barista "the usual," they nod: they know you, they remember yesterday, they can see you shivering and guess you want it hot.
An AI model starts every conversation as a brand-new barista on their first shift. Brilliant, fast, well-read, but with zero knowledge of you. Say "make it good" and they'll make something good. Say "a double oat-milk latte, extra hot, light foam" and you'll get exactly that.
The difference between a vague prompt and a specific one is the difference between "surprise me" and an actual order.
See it in action
Same task (a birthday message) asked two ways. Watch what one sentence of detail does.
โ Weak prompt
Prompt
Write a birthday message
Output
Happy birthday! Wishing you a wonderful day filled with joy and celebration.
Perfectly fine. Perfectly forgettable. The model had nothing to work with, so it picked the most statistically average birthday message on Earth.
โ Good prompt
Prompt
Write a short, funny birthday message for my younger sister who just adopted a very dramatic cat. Two sentences max.
Output
Happy birthday to my favorite sister and her new furry landlord. May your year be full of joy and your couch survive the cat. Here's to another year of you being the best (and now most scratched-up) person I know.
One extra sentence of detail (who, tone, topic, length) and suddenly the message sounds like it came from an actual human who knows an actual sister.
Same task, wildly different results. The only thing that changed was the wording. When you're in a real AI chat, try this yourself with a prompt like this one:
Write a short, funny birthday message for [person] who recently [detail about their life]. Keep it under 3 sentences and make it sound like me: casual, warm, a little sarcastic.
Swap in the bracketed parts and notice how each detail closes a loophole the genie would otherwise exploit.
What a prompt is (and isn't)
| A prompt is... | A prompt is not... |
|---|---|
| Instructions the model follows literally | A mind-reading request |
| The model's entire view of your world | A conversation with someone who knows you |
| Something you can revise and improve | A one-shot magic spell |
| Plain language, no code required | Programming (though it rewards precision like programming does) |
Checkpoint
A prompt is the literal text you give an AI, and since the model answers what you wrote (not what you meant), clear wording is your superpower.
The big idea of this whole course
Every lesson ahead teaches one thing: how to close the gap between what you mean and what you say. Roles, examples, formats, and iteration are all just tools for writing better wishes for a very literal genie.
๐ Quiz
Question 1 of 3What is a prompt, at its core?