Course A · Module A2 Prompting · Lesson 4 of 17

Anatomy of a Reliable Prompt

Lucky one-off results are accidents. A prompt built from five parts works every time.

~12 min · Have a chat assistant open to test with · Prereq: Lessons A1–A3

You've probably had the experience: you ask an AI something, get a great answer, try the same type of question a week later, and get something mediocre. The output isn't random — you changed the input. A well-built prompt has five predictable parts. Once you can write all five, you stop getting lucky and start getting consistent. That single shift — from luck to system — is this lesson's win. And because the same five parts apply whether you're writing a chat prompt, an automation step, or a build brief, this skill underpins everything else in this course.

The five parts

1 · Role Tell the AI who it is in this context. "You are a friendly customer-service writer who works for a small e-commerce brand." Role sets the voice, expertise, and judgment the model leans on. Omit it and you get a generic, hedged answer.

2 · Context Give it the background it can't know on its own — your product, your audience, the previous email in the thread. Remember the forgetful intern from A1: it has never seen your business. Context fixes that.

3 · Task One clear instruction. "Write a 3-sentence follow-up email." Not five instructions at once. If you want multiple things, send multiple prompts or number them. Ambiguity in the task is the most common reason for a mediocre result.

4 · Examples Show the model what "good" looks like. Paste one previous email in your voice, one sample output you liked, or a before/after pair. Examples do more work than long instructions — they show the standard rather than describing it.

5 · Output format Tell it exactly what to produce: length, structure, tone, what to include or skip. "Under 100 words. Bullet points. No sign-off." Without this, the model guesses — and its default is verbose.

You don't need all five for every prompt. A quick "summarize this email" needs only a task. But any recurring prompt — one you'll use again and again — is worth building all five. The parts you include are the parts you control; the parts you leave out are the parts the model guesses.

Weak prompt vs. strong prompt

Here's the same job done both ways. The task: write a follow-up email after a sales call.

# Weak prompt — one line, no context

Write a follow-up email after a sales call.


# Strong prompt — all five parts

Role: You are a warm, concise business-development writer for a boutique
design studio that works with early-stage startups.

Context: I just had a 30-minute discovery call with Maya Chen, founder of
a seed-stage fintech. She mentioned two pain points: inconsistent brand across
decks and no design support for an upcoming fundraise.

Task: Write a follow-up email referencing those two pain points and
proposing a short next step (a 20-minute scoping call).

Example of our tone: "Great talking through your roadmap today —
the deck challenge you mentioned is exactly the kind of problem we fix fast."

Output format: Plain text, under 120 words, no formal sign-off, subject
line included.

Notice what changed. The weak prompt produces a generic template the model could write for any business. The strong prompt produces something that sounds like you, references this prospect, and tells the model exactly how long to go. Same tool. Same model. Completely different output — because the input was specific.

This skill travels everywhere

Every tool in this course takes a prompt:

Automations (Course A)

The AI step inside a Zapier or Make workflow is a prompt
Better prompt → better drafted email, summary, or tag every time the flow runs
You'll paste your five-part prompt into the AI action step

Building apps (Course A)

Your first message to Lovable or Bolt is a prompt
Better prompt → cleaner first build, fewer revision rounds
The build brief you write in A13 uses this exact structure

A prompt is a job posting. A vague job posting — "we need someone good with computers" — attracts generic applicants. A specific one — role, requirements, examples of the work, expected output — attracts exactly the right person. The AI responds the same way: specificity in, specificity out.

Check yourself

Answer before opening. Recalling it now is what makes it stick.

You get a great AI result but the output is always three paragraphs when you want two sentences. Which part of the prompt is missing?

Output format. Without it, the model defaults to verbose. Adding "two sentences, plain text, no sign-off" — or whatever you actually want — locks the length on every run.

Which of the five parts does the most work when you want the AI to match your brand voice?

Examples. Describing your voice in words ("warm but professional") is vague. Pasting one email in your actual voice shows the model the standard directly — and it calibrates to that.

Your automation runs a prompt 200 times a month. You improve the prompt once. What happens?

Every one of those 200 runs improves automatically. That's the leverage of a well-structured prompt inside an automation — fix it once, every future output benefits. Sloppy prompt in a high-volume flow multiplies the sloppiness; a tight one multiplies the quality.