Module 1 · Lesson 2
Anatomy of a Great Prompt
Breaking down the components that work.
10 min read
Most people prompt like they search — short, keyword-heavy, impatient. Then they get mediocre output and blame the model.
The model did not fail. You handed it a vague task and expected telepathy.
Why this lesson matters
By the end of this page you will be able to look at any prompt — yours or someone else's — and diagnose what it is missing in under thirty seconds. That diagnostic speed is the difference between someone who "uses AI" and someone who gets AI to work for them.
The four pressure points
A great prompt is not magic. It is information transfer. Your job is to give the model enough information that a capable stranger, reading your prompt cold, would know exactly what you want, why, and what success looks like.
That breaks down into four pressure points. Not rules to follow mechanically. Checks to run.
| Pressure point | The question it answers |
|---|---|
| Role | Who should the model be when it answers? What expertise is it drawing on? |
| Task | What specifically is it doing — and what is the deliverable? |
| Context | What situation is this happening in? What does the model need to know? |
| Constraints | What must the output avoid, include, or stay within? |
Miss any one and ambiguity rushes back in.
Role
The role frames the voice and the expertise. "Act as a senior M&A attorney" gets you a very different output than "Explain this to me." Not because the model becomes an attorney — it does not — but because it narrows the space of plausible completions toward the kind of output an attorney would produce.
Heuristic
If the task requires a particular flavor of expertise, bolt the role on first. The first six words of a prompt do more to shape the output than the next sixty.
Task
The task is the verb plus the deliverable. "Help me" is a verb with no deliverable. "Draft a 250-word product announcement" is a verb and a deliverable. The second one gives the model something to stop at. The first gives it nothing.
Vague task vs. bounded task
| Vague | Bounded |
|---|---|
| "Summarize this." | "Summarize this in five bullet points, under 15 words each, aimed at a non-technical reader." |
| "Help me improve this copy." | "Rewrite this paragraph to cut 30 words while keeping the two statistics and the call-to-action." |
| "Give me ideas." | "Give me ten distinct angles for a B2B LinkedIn post about employee retention, each under 20 words." |
Context
Context is what a capable stranger would need to know before they could start the task. Often this is the hardest part to write because it is the part you already know in your head and forget to say out loud.
- Who is the audience, specifically?
- What is the situation or moment this artifact will land in?
- What has already been tried or decided?
- What is the relationship between the people involved?
The context trap
Most vague prompts are not vague about the task — they are vague about the context. "Write a follow-up email" is not missing the verb. It is missing the relationship, the history, and the goal of the follow-up. Fix that first.
Constraints
Constraints are the guardrails. Length, tone, format, what to avoid, what to include. They are where most people stop short — and stopping short on constraints is why outputs drift.
Those are approximate rates of constraint usage across a thousand prompts we reviewed from new learners. The cheapest upgrade you can make to your prompting is to add the constraints you were already thinking but not writing down.
The finished article
Here is the same prompt run through the four pressure points.
Four pressure points in one prompt
Role: Act as a senior B2B copywriter who has shipped landing pages for 30+ SaaS companies.
Task: Draft three distinct hero-section headlines and one supporting subhead for each.
Context: The product is an AI meeting notes tool for sales teams. The current headline is "AI-powered meeting intelligence." It is not converting. The audience is VP-of-Sales buyers who are skeptical of AI hype and tired of notes tools.
Constraints: Each headline under 9 words. Each subhead under 16. Avoid the words "revolutionize," "unlock," "empower." Avoid em-dashes. Tone: confident, specific, not cute.
You could hand that prompt to any capable freelancer and get something useful back. You can hand that prompt to any modern LLM and get three actually-good options. That is the bar.
What this looks like when it goes wrong
A common failure mode: you write a good role, a good task, strong constraints — and forget the context. The model produces crisp prose that misses the point. Always sanity-check that you have named the audience and the situation before you hit send.
Another common failure: all four pressure points exist but the constraints conflict. "Write it warm but formal, detailed but short, specific but hedged." The model will pick one and drop the others. If your constraints contradict, ship fewer, stronger ones.
Key takeaway
Every prompt has four pressure points — role, task, context, constraints. Before you hit send, check all four. Missing one is where outputs drift. Missing two is why people think LLMs are unreliable.
Check your understanding
According to the lesson, what is the primary reason to include format instructions in a prompt?
The lesson's one-rule version of prompting compares the model to which of the following?
Great prompts have four pressure points: role, task, context, and constraints. Miss one and the model averages. Hit all four and it performs.
Open the Prompt Workshop, paste any prompt you wrote this week, and let it grade you on the four pressure points before you send it next time.
Refine a prompt. Get the diff, rationale, and tips.