Understanding Loops CHAPTER 01 · PROMPTS
Prompt Engineering

From wishing to guaranteed results

Ever thought you could get better results from your AI if you had better prompts? You can. Ask it to "build me a website for my AI automation business" and you'll get one — fast, clean, and looking like every other AI-generated site out there. There's a name for that: AI slop. It's not the AI falling short; it's the AI with nothing to aim at. There's no such thing as a perfect prompt, but there are proven techniques that deliver better results from the same AI. Keep reading for the four decisions behind every strong prompt, plus one alignment move most people skip. Get them down and you won't just write better prompts — you'll change what the AI is able to give you.

Same AI, two very different results

🫵
One-liner
"build me a website
for my AI business"
🧠
The model
predicts the most
likely continuation
📄
AI slop
the site everyone
else gets
🫵
Structured prompt
audience · traffic ·
goal · design
🧠
The model
doesn't have
to guess
🎯
Aligned build
quality up, from
prompt one
SAME MODEL · SAME DAY · ONLY THE PROMPT CHANGED
Two requests, two outcomes

A language model is a prediction machine: given your words, it produces the most statistically likely continuation of them. "Build me a website for my AI automation business" has millions of plausible answers, so the model fills every blank you left — sensibly, every time. But sensible, with nothing to aim at is the average, and the average is the template every AI automation business already has. Here are both prompts, side by side:

✗ "build me a website for my AI automation business"
# what comes back: a clean, competent template —
# the same site it builds for everyone (AI slop).
 "build me a website for my AI automation business.
Mainly targeting small business owners. Most of my
traffic comes from LinkedIn and Xprioritize SEO
so Google search brings organic traffic too. The
goal: sell my services and book calls. Design:
clean and modern, one page to start."
# what comes back: your site — built for your
# audience, your channels, your goal.
The wish vs. the structured prompt — same model received both

Read the second prompt again. No tricks, no magic phrasing, no secret words — just decisions made up front, before the model had to make them for you. You just guaranteed — or nearly guaranteed — alignment from your very first prompt, and the quality of the build went up with it. The slop was never the model being weak; it's a mirror — a request that didn't decide anything, reflected straight back at you.

🔓
The unlockSame model, wildly different result — so what changed? Exactly four things. Name them and you can repeat this on demand, for anything you ask. That's the next section.

Four decisions collapse the fog

🌫 millions of possible "websites"EVERY SITE THE MODEL HAS EVER SEEN
ROLEDELIVERABLECONSTRAINTSFORMAT
thousands that fit your goal & audience
🎯 the one you actually need
EACH DECISION REMOVES A GUESS THE MODEL WAS ABOUT TO MAKE FOR YOU
The four decisions as a filter

The second prompt made decisions — before the model could guess them. There are exactly four, and it had already nailed three of them without ever naming them:

"build me a website for my AI automation business…"

✓ DELIVERABLE  a website that sells services & books calls
✓ CONSTRAINTS  small business owners · LinkedIn + X · SEO
✓ FORMAT       design: clean and modern, one page
○ ROLE         the one it skipped — try "a conversion-
               focused web designer"
The prompt, dissected — three decisions made, one free upgrade left
ROLEWho should it be? Give it a point of view to answer from — "a conversion-focused web designer," "an SEO-minded copywriter," or both. Even a couple of words aims its expertise and tone; you don't need a bio.
DELIVERABLEWhat do you actually want? Say the outcome you're after — "a site that books calls," "a pricing page," "a landing page for one offer." You don't have to spell out every detail; naming the goal is enough to point it in the right direction.
CONSTRAINTSWhat should it keep in mind? The things only you know — who it's for, your real limits, what to avoid. "Small business owners," "on a tight budget," "nothing salesy." This is usually where the biggest gains hide.
FORMATWhat shape — if it matters? Models are good at picking a sensible layout on their own, so save this one for when you have a specific shape in mind: "one page, no menus," "clean and modern," "just the copy, no code."
The four decisions, defined
🔓
The unlockFour decisions, not forty tricks. Before you hit enter, check: did I say who, what, and within what limits — and in what shape, if I care? Ten seconds of deciding replaces three rounds of fixing.

Watch the same request, upgraded

You've seen it on a website build — now watch the same four decisions on three completely different asks. The vague wish, then the structured prompt that fixes it:

✗ "write me a marketing plan for my café"

✓ ROLE: social media strategist for local coffee shops
✓ DELIVERABLE: 30-day Instagram plan for the launch
✓ CONSTRAINTS: regulars, not tourists · 5 hrs/week to run it · tiny budget
✓ FORMAT: week-by-week table
✗ "build me a mobile app for tracking my habits"

✓ ROLE: mobile developer who keeps things simple
✓ DELIVERABLE: add habits, tick them off daily, see streaks
✓ CONSTRAINTS: iPhone only · works offline · no sign-up
✓ FORMAT: one screen, no menus
✗ "help me run a marathon"

✓ ROLE: running coach for first-time marathoners
✓ DELIVERABLE: 16-week plan from 5k to race day
✓ CONSTRAINTS: 4 training days/week · starts at a 5k · knee-friendly
✓ FORMAT: weekly table — mileage, workouts, rest
Same skill, any domain — the details change, the four decisions don't

Run any of these and the difference is immediate: what comes back is usable, addressed to your situation, in the shape you needed. Not because the model got smarter — because the request finally said something.

🔓
The unlockOne structure runs everything. A website, a workout, a launch plan — the domain changes, the four decisions don't. Fill them in and the model stops guessing, every single time.

The alignment move: make it interview you

🫵
You brief it
the goal, roughly
🧠
It asks you
one question
at a time
🫵
You answer
the parts only
you know
🎯
Aligned build
no wasted attempt
FLIP THE INTERVIEW — THE QUESTIONS SURFACE THE DECISIONS YOU DIDN'T KNOW YOU HADN'T MADE
Confirming alignment before the work starts

Here's the move almost nobody uses, and it separates the pros from everyone else on big tasks: before the model builds anything, make it interview you. You know things the model can't guess — your audience, your constraints, what "good" looks like to you. Instead of discovering the misalignment three drafts in, surface it before draft one:

Before you start: ask me up to 5 questions, one at a
time, until you're confident you understand exactly
what I need. Then restate the plan in 3 bullets and
wait for my go.
Paste this ahead of any big request

Two honest caveats. This is for the big ones — a project, a plan, anything you'd be upset to redo — not every quick ask. And sometimes you should skip it deliberately: when you're still developing the idea yourself — a wide-open brainstorm, a direction you haven't landed on — locking in answers too early narrows the field before you've seen what's in it. Constrain when you know what you want; stay loose when you're still figuring it out.

🔓
The unlockAlignment is cheaper before the build than after it. Five questions up front replace three rounds of "that's not what I meant." Skip it only when you want the AI unconstrained.

Start building

AI slop or premium builds — the difference is a structured prompt, and you can write one by hand right now. You know the four decisions. The more detail you add, the more the build bends to fit you; the real skill is knowing which details matter and which the model can safely figure out on its own.

And if you don't know exactly which details belong in a structured prompt, that's okay — some tools specialize in this. Prompt Optimizer takes one-liner prompts and transforms them into comprehensive structured prompts.

🔓
Start buildingIt writes the whole structured prompt for you — so you can guarantee high-quality builds from the first attempt. Paste your rough idea and go.Try Prompt Optimizer