A prompt that works
Lesson 10 of 23 in the course “AI basics”. Free, with animations and a quiz.
Role, task, format, example
A prompt is your message to the model, your request. A good prompt has four parts. Role: who to be (“You are a chemistry tutor”). Task: what to do (“Explain what a mole is”). Format: who it's for and what shape (“for 9th grade, 5 points”). Example: show the style you like. The model can't read your mind, so put everything into words.
Specifics instead of fluff
A vague prompt gets a vague answer. “Tell me about marketing” gets you a lecture about nothing. “Give me 5 TikTok post ideas for a coffee shop near a campus, one line each” gets you something you can actually use. The difference is in the details: how many, for whom, where, in what shape. Spoiler: the model doesn't mind long prompts, they actually help.
Building a full prompt
Example: you need to prep for a test. Weak: “explain the mole”. Strong: “You are a chemistry tutor. Explain what a mole is for a 9th grader. Give 5 short points and one solved problem. Explain it like in this example: …” Role, task, format and example are all there. The answer lands at your level, not at university-textbook level.
Refine instead of starting over
The first answer is rarely perfect, and that's fine. Don't open a new chat, reply in the same one: “shorter”, “simpler”, “add a real-life example”. Within one chat the model sees the whole conversation and takes it into account. Often 2–3 follow-ups are enough to get the answer where you need it. Start a new chat when you switch topics.
Check yourself
- Which prompt will work better?
Correct answer: “You're a coach. Make a 2-week push-up plan for a beginner, as a table”. It has a role, a task, who it's for and a format. The others don't say what exactly is needed or in what shape. - How do you get the answer in a handy shape right away?
Correct answer: State the format: list, table, under 100 words. With the format in the prompt, the model answers in that shape right away. Caps and repeats add nothing. - The answer came out long and complicated. What now?
Correct answer: Reply in the same chat: “shorter and simpler”. In the same chat the model sees the context and will adjust. A new chat starts from zero and may give you the same thing.
← How to tell if an AI is good Inside a chat: system, user, history →