Lesson 002Field guide · Prompting 101
Role, Task, Context, Format
The four building blocks behind almost every great prompt.
You'll learn
- What each of the four parts actually does to the output
- Strong vs. weak versions of role, task, context, and format
- When each part matters most — and when you can skip it
- How to keep a prompt short while all four parts pull weight
- How to rebuild a weak prompt part by part
01Section
Four parts, four different jobs
In the last lesson you saw that role, task, context, and format close the gaps a vague prompt leaves open. This lesson goes one level deeper: each part closes a different gap, and knowing which is which tells you exactly what to strengthen when a result falls flat.
Here are the four parts working on one request — notice how each steers a different dimension of the answer:
Role
Controls the vantage point — whose eyes and standards judge the work.
Task
Controls the action — what actually gets done, and to what.
Context
Controls the relevance — whether the answer fits your situation or an average one.
Format
Controls the usability — whether the answer works where you're going to put it.
Checkpoint
02Section
Role: the vantage point
A role doesn't give the AI knowledge it lacks — it can't actually become a lawyer or a CFO. What a role does is select a vantage point: the perspective, the voice, and the standard of "good" the answer gets held to. "Act as an editor" produces cuts and questions; "act as a fan" produces praise. Same text, different vantage point, different output.
That's why generic roles do so little. "You are a helpful assistant" is already the default — you've told it nothing.
Weak role
You are a helpful assistant. Look at my resume.
Strong role
Act as a hiring manager who reads a hundred resumes a week and spends thirty seconds on each. Look at my resume.
The strong version names a perspective and a standard. Now the AI knows what "good" means: survives a thirty-second skim.
Role matters most when the task involves judgment, taste, or tone — reviewing, critiquing, writing for an audience. It matters least for mechanical jobs like reformatting a list or fixing grammar, where any vantage point does the same work. Skip it there and lose nothing.
Checkpoint
03Section
Task: the verb and the deliverable
The task is the one part every prompt has — but weak tasks hide behind soft verbs. "Help me with," "look at," and "improve" all leave the real decision to the AI: help how? A strong task pairs a precise verb with a concrete deliverable.
Precise verbs: summarize, rewrite, list, compare, draft, critique, shorten. Each one names an action with a recognizable result. If you can't picture what the output looks like, the AI can't either.
Soft verb, no deliverable
Help me with my product page.
Precise verb, concrete deliverable
Rewrite the first paragraph of my product page so a first-time visitor understands what we sell within five seconds.
One task per prompt
"Rewrite this, then suggest headlines, then check the tone" splits the AI's attention three ways. Big requests work better as a sequence — the last lesson in this path shows you how.
Checkpoint
04Section
Context: what the AI can't know
Context is the highest-leverage part, because it's the only one the AI genuinely cannot supply itself. It can invent a role, guess a task, and default to a format — but it knows nothing about your situation unless you say it.
Useful context almost always falls into three buckets. If your prompt covers all three in a sentence each, you've done the job:
Audience
Who will read or use this — and what they already know.
Situation
What's true right now: what exists, what happened, what's changed.
Goal
What should be different if this works — the result you're actually after.
No context
Write a thank-you note to a client.
Audience, situation, goal
Write a thank-you note to a long-term client who just referred two new customers to us. She's informal and busy. Goal: make her feel genuinely appreciated without it reading like a form letter.
Three sentences of context — audience, situation, goal — and the AI is now writing your note instead of the average note.
Privacy note: Context means describing the situation, not exposing it. "A long-term client who referred two customers" works fine — real names, account details, and other people's private information stay out of the prompt.
Checkpoint
05Section
Format: the shape of the answer
Left alone, most AI tools default to the same shape: a medium-length essay with a cheerful intro and a tidy conclusion. That's rarely the shape you need. Format is you describing the container the answer has to fit — length, structure, layout.
Format matters most when the output is going somewhere: pasted into a slide, dropped into an email, read aloud in a meeting, compared side by side. Anything with a destination has a required shape — say it, or spend your time reshaping.
Format that matches the destination
Give me a two-column table: the customer objection in column one, a one-sentence response in column two. Five rows, ready to paste into a training doc.
Checkpoint
06Section
All four parts, still short
A common worry: won't covering all four parts make every prompt an essay? No — each part is usually one sentence. Four deliberate sentences beat four paragraphs of rambling.
Four parts in four sentences
Act as an experienced editor for a general audience. Cut this article introduction to half its length. It's for a local newsletter read mostly on phones, and the goal is that people keep reading past the first paragraph. Return only the edited version, no commentary.
Short on time — which part do you add first?
Context, almost always. Role and format have workable defaults, and the task is already in your head. But context is the part the AI cannot guess — one sentence about audience, situation, or goal moves the answer further than anything else you could type in the same ten seconds.
Key takeaway
When an output falls flat, don't rewrite the whole prompt — identify which part failed. Off-target voice: role. Wrong action: task. Generic: context. Wrong shape: format.
Pause and think: Think of the last AI answer that disappointed you. Using the four jobs — vantage point, action, relevance, usability — which part of your prompt was the weak one?
Checkpoint
Prompt rewrite exercise
Rebuild a weak prompt part by part
You're about to ask an AI for feedback on a presentation — a task where every one of the four parts earns its place. First spot what the weak prompt is missing, then build the strong version part by part: pick a preset or write your own. When it's assembled, copy it into your own AI tool and compare what you get against the weak version.
The weak prompt
Give me feedback on my presentation.
Quick check
3 quick questions — no pressure
There's no pass or fail here. Answer them all, and we'll show you the answers either way.