Four parts, one great prompt
Great prompts aren't long. They're complete. Most strong prompts contain up to four ingredients:
| Part | What it does | Ask yourself |
|---|---|---|
| Task | The verb. What you actually want done. | "What action am I requesting?" |
| Context | Background the model can't guess. | "What does it need to know?" |
| Format | The shape of the answer. | "What should the output look like?" |
| Examples | A sample of what "good" looks like. | "Can I show one?" |
Not every prompt needs all four. But when a result disappoints you, the fix is almost always one of these four gone missing.
A real-world test: the Airbnb message
You're hosting an Airbnb, and a guest just asked about early check-in. Watch what happens with and without the four parts.
โ Weak prompt
Prompt
Write a reply to my Airbnb guest
Output
Hi there! Thanks so much for reaching out. We're so glad you chose our place. Please let us know if you have any questions at all. We look forward to hosting you!
Cheerful, polite... and completely useless. The model didn't know what the guest asked, what your policy is, or what tone you use. So it produced professional-sounding fog.
โ Good prompt
Prompt
Task: Write a reply to my Airbnb guest. Context: They asked for a 1pm early check-in, but cleaning finishes at 2pm. I can offer free luggage storage from noon. My style is warm but brief. Format: 3-4 sentences, no emojis, sign off as Sam.
Output
Hi! Thanks for asking. Our cleaners wrap up at 2pm, so I can't quite swing a 1pm check-in. But you're welcome to drop your bags with us any time from noon and go explore. I'll message you the moment the place is ready. - Sam
Task, context, format. All present. The model had a real situation to respond to, so it produced a real answer: policy explained, alternative offered, tone matched.
Notice: the good prompt isn't fancy. It's just complete. The model went from improvising a soap opera to answering an actual question.
Breaking down each part
Task: say the verb out loud
"Write a reply," "summarize this," "list five options," "rewrite this in plain English." Lead with it. Buried tasks get buried results.
Context: the stuff only you know
The model has read half the internet, but it has read zero of your life. Your cleaning schedule, your audience, your deadline, your grumpy manager. If it matters, it goes in the prompt.
A quick context test: could a smart stranger do this task with only what you've written? If they'd have to ask you a question first, your prompt is missing context.
Format: describe the container
Bullets or paragraphs? Three sentences or three pages? Table? Casual or formal? The model will pick a format; the only question is whether you or the dice pick it.
Examples: show one, save a thousand words
If you have a past reply you loved, paste it in: "Match the tone of this example." We'll go deep on this trick in the Few-Shot lesson; it's one of the most powerful moves in prompting.
Your turn: assemble a prompt
Here's a fill-in-the-blanks template you can reuse for almost any message-writing task. Copy it, swap the bracketed parts, and you've got all four organs accounted for:
Task: Write a reply to my Airbnb guest.
Context: They asked about [their question]. My situation is [your constraints]. I can offer [your alternative or answer].
Format: [length] sentences, tone is [warm / formal / brief], sign off as [name].
Example (optional): Here is a past reply I liked, match its style: [paste it]
This template feels almost too simple: four labeled lines, that's it. But labeling the parts does two jobs: it forces you to notice which part you're missing, and it hands the model a tidy structure it can follow line by line. Everybody wins, especially your guest.
Checkpoint
A complete prompt has up to four parts (Task, Context, Format, Examples), and most bad outputs trace back to one of them being missing.
๐ Quiz
Question 1 of 3Which four ingredients make up the anatomy of a strong prompt?