AN
Teach·Anatomy of a Power Prompt·Lesson 1

Anatomy of a Power Prompt

Five building blocks that turn a vague idea into a result the AI can nail on the first try.

14 min read·7 parts

Before you touch any AI builder, learn the one skill that separates "meh, generic result" from "wow, that's exactly what I wanted": how to write a prompt with real structure. In this lesson you'll learn a five-block anatomy that works on every AI tool — Lovable, ChatGPT, Claude, Cursor — and you'll practice it against fresh, everyday product ideas.

Part 1 · Why prompts matter

The same AI can give you a forgettable website or a portfolio-worthy one. The model didn't change between those two runs — the instruction did. Think of your prompt as the brief you'd hand to a freelancer: a vague brief buys you a vague result, no matter how talented they are.

Figure 1.1
Prompt: "build a website for my bakery"
sweet-bakery.lovable.app
SWEET BAKERY
HomeMenuAboutContact
Welcome to Our Bakery
Fresh baked goods daily
Learn More
Breads
Lorem ipsum
Cakes
Lorem ipsum
Pastries
Lorem ipsum
Generic stock layout · placeholder copy · bootstrap blue.
Prompt: 5-block spec
marlow-sourdough.co
MARLOW & RYE
LoavesBakesVisit
Est. 2019 · Preorder Fri
Slow-fermented sourdough, baked Saturdays.
Twelve loaves. One oven. Reserve yours by Thursday night.
Reserve a loaf →
This week's bake
№04 · Country Rye
Editorial palette · confident headline · a clear next action.
Same AI, same afternoon. The only thing that changed is the shape of the request.

A structured prompt isn't "prompt engineering wizardry" — it's just saying five simple things in a specific order. That's the whole trick.

Part 2 · The 5-block anatomy

Every solid prompt names five blocks. Miss one and the AI has to guess — and it will guess in the most generic direction possible. Stack them like LEGO and your intent stops leaking.

Figure 2.1
1GOAL
What are we building?
2INPUT
What does the user give?
3LAYOUT
How should it look?
4FEATURES
How does it behave?
5OUTPUT
What does it return?
Five blocks, stacked. Small, ordered, hard to skip once you know it.
  • Goal — the thing you're trying to build and who it's for.
  • Input — what data or actions the user provides.
  • Layout — how it should look and feel.
  • Features — the behavior and logic inside it.
  • Output — what the system returns to the user.

Part 3 · Block 1 — Goal

The Goal answers two questions in one sentence: what is this and who is it for. A clear goal gives every later decision a target to aim at.

Avoid
Vague
"Make a habit tracker app." Habits for what? Who is using it? On phone or desktop? The AI defaults to a generic dashboard.
Do
Power
"A habit tracker for busy parents to log three daily habits (hydration, walk, sleep) in under 10 seconds on their phone before bed."
Rule of thumb
If someone read only your Goal line, could they draw the first screen on a napkin? If yes, you're set. If not, you're missing an audience, an outcome, or a constraint.

Part 4 · Block 2 — Input

Input is what the user gives the system. Naming each field beats saying "add a form". If a field could be a dropdown, a slider, or a toggle, decide — don't leave it to chance.

Avoid
Vague
"Users can search for recipes." Search by what? Ingredients? Cuisine? A single messy text box?
Do
Power
For a recipe finder, collect: ingredients on hand (tag input), diet (dropdown: any / vegetarian / halal / gluten-free), and max cook time (slider, 10–60 min).
Figure 4.1
pantry-chef.app
Find recipes →
A specified form leaves no room for the AI to invent random fields.

Part 5 · Block 3 — Layout

Layout is the visual direction. Palette, typography, density, and device priority. You don't need to be a designer — you just need to commit to a lane so the AI stops drifting into generic bootstrap-blue.

"For a book-club community page: soft editorial feel, warm cream background with forest-green accents, serif headline / clean sans body, mobile-first, card grid of current reads."
Figure 5.1
Palette
Type
Serif headline
Sans body — 16px, generous line-height for reading long reviews.
Wireframe
Palette + type + wireframe together give the AI a lane, not a canvas.

Visual clues that work

  • Style words — editorial, brutalist, minimalist, playful.
  • Colors — 2–4 concrete hexes or a named palette.
  • Hierarchy — headline / subhead / body sizes, or a reference site.
  • Responsiveness — mobile-first, tablet, or desktop-primary.

Part 6 · Block 4 — Features

Features are behaviors, not screens. Each one should describe a rule the system enforces or a moment it reacts. Aim for a small, sharp list — three to six items — that all serve the Goal.

Figure 6.1
  • ✓
    Form validation
    Reject empty duration; require at least one activity.
  • 🔥
    Streak counter
    Increment when today's session is logged.
  • 🎉
    Milestone burst
    Confetti at 7-day and 30-day streaks.
  • 📶
    Offline-first
    Queue logs locally and sync when back online.
Features for a focus-timer + habit-log hybrid. Every item is a behavior, not just a screen name.
Watch out
Features should help the Goal, not bury it. If you can't say which user story a feature serves, cut it or postpone it to a later iteration.

Part 7 · Block 5 — Output

Output is what the system gives back after the user acts — a confirmation, a computed result, a chart, an email, a redirect. Weak prompts stop at "show a report". Strong prompts describe the shape.

Avoid
Vague
"Show a summary at the end of the week."
Do
Power
A weekly dashboard showing: completion percentage as a large number, a 7-bar chart of daily minutes, current streak as a circular progress, and a "next milestone" nudge.
Figure 7.1
focus-tracker.app/summary
This week
82%
completion
MTWTFSS
Streak
12d
An output described this concretely almost draws itself.

Part 8 · Iterate to power

Version 1 is a draft, not a verdict. Great prompt engineers keep the good bits from v1 and send small, surgical follow-ups. Each iteration should change one thing at a time, so you always know what caused the improvement — or the regression.

Figure 8.1
  1. 1v1 — Baseline

    Ship the 5-block prompt as-is and take stock.

  2. 2v2 — Add depth

    Introduce a streak counter and empty-state copy.

  3. 3v3 — Polish

    Tighten the palette, add micro-animations, rewrite the CTA.

  4. 4Save

    Bookmark the version you like before iterating again.

Three focused iterations beat one massive rewrite. Each version is a checkpoint you can roll back to.

Follow-up phrases that work

  • Keep everything — only change the primary color to forest green.
  • Revert the last change and try again with a smaller step.
  • Explain the root cause before fixing anything.

Part 9 · Pro tip: the PRD shortcut

Writing all five blocks from a blank page is hard. The shortcut: use the AI itself to draft a short product requirements doc first, then ask it to reshape that doc into a 5-block prompt. Paste the result into a fresh chat and build.

Figure 9.1
  1. 1Discuss

    Talk your idea through with the AI — target user, problem, must-haves.

  2. 2Ask for a PRD

    Have the AI write a short product requirements doc.

  3. 35-block prompt

    Ask it to rewrite the PRD as Goal · Input · Layout · Features · Output.

  4. 4Paste & build

    Open a fresh chat, paste the 5-block prompt, hit send.

A four-step loop that turns fuzzy ideas into a ready-to-run 5-block prompt.
"Act as a product manager. I want to build a book-club community site. Ask me five clarifying questions, then write a one-page PRD. After I approve it, rewrite the PRD as a Goal / Input / Layout / Features / Output prompt I can paste into a builder."

Cheat sheet

Print this. Pin it above your keyboard. Read it before every new prompt.

  • Goal — one sentence: what and for whom.
  • Input — name every field and its control type.
  • Layout — palette, type, hierarchy, device.
  • Features — 3–6 behaviors that serve the goal.
  • Output — describe the shape, not just "a report".
  • Iterate — one change per version, save what works.
  • PRD shortcut — let the AI draft the prompt for you.
Recap
You now have a reusable anatomy for turning any idea into a buildable prompt. In Lesson 1 we'll open Lovable and put this into practice — starting your first project the right way.
CourseBack to Anatomy of a Power Prompt