A perfect prompt isn’t long — it’s complete. It answers, in advance, every question the AI would otherwise have to guess: who am I supposed to be, what exactly do you want, for whom, in what shape, and within what limits. Miss a part, and the AI fills the gap with “average.” That’s the entire anatomy of a perfect prompt.
In this guide we’ll dissect one real prompt layer by layer — the same five parts you saw on the cheat sheet, now with the reasoning behind each, the optional “power parts” pros add, and the assembly order that works. This is guide #6 of our Prompt Engineering roadmap.
The Specimen — One Prompt, Fully Assembled
Sixty words. Nothing fancy. But every one of the five organs is present and doing a job. Let’s take it apart.
The 5 Parts, Dissected
Who should the AI be?
The role sets vocabulary, standards, and perspective before a single word of the task is read. “Act as a senior developer” produces stricter code review than no role; “act as a patient teacher” produces gentler explanations. Pick the person you’d hire for this exact job.
What exactly should it do?
One clear action with a strong verb. “Tell me about resumes” is a topic; “rewrite these bullets” is a task. Weak verbs (discuss, explain about, help with) invite essays — strong verbs (rewrite, compare, list, diagnose, draft) invite results. One task per prompt; chain the rest.
What does it need to know?
Context is the difference between generic and personal. Who is this for? What’s the situation? What’s the goal? Every relevant detail you leave out is a guess the AI makes for you — and it guesses “average.” Two sentences of context routinely doubles output quality.
What should the answer look like?
Table, numbered steps, word limit, tone, language. Format stated upfront shapes the entire generation — asked at the end, it costs you a second prompt. Be concrete: “a table with columns Before and After” beats “organized nicely.”
What are the rules and limits?
Constraints prevent the failure modes you can already predict: length, banned words, things not to invent. This is where you encode your experience of past bad answers. The most valuable constraint of all: “don’t invent numbers — put [X] where I should add real figures.”
The Power Parts (Optional, But Pros Use Them)
- Examples (few-shot): “Here’s one bullet done right: [example]” — the strongest signal you can send. Full guide: Few-Shot Prompting.
- Audience: “…for a recruiter who spends 8 seconds per CV” — sharpens tone and depth beyond what role alone does.
- Reasoning request: “think step by step before answering” — for anything with logic or trade-offs. See Chain-of-Thought.
- Escape hatch: “ask me clarifying questions first if anything is unclear” — converts your blind spots into questions instead of guesses.
- Self-check: “then review your answer against my constraints and fix any violations” — a built-in quality pass.
Does Order Matter?
Less than people think — but two placements are proven winners: rules go early (instructions buried mid-paragraph get skipped) and the most important requirement gets repeated at the end in long prompts. The natural reading order — Role → Task → Context → Format → Constraints — works because it mirrors how you’d brief a person: who you are, what I need, the background, the deliverable, the rules.
Anatomy Mistakes (Quick Autopsy)
- Two tasks in one body: “summarize this AND write a reply AND translate it” — split into a chain; each organ can only do one job well.
- Role without task: “Act as a marketing expert” …and then what? A role with no verb is a costume with no play.
- Constraints that contradict: “detailed but under 50 words” — the AI will sacrifice one silently. Decide which matters.
- Context dumping: pasting 3 pages of background with the task buried in the middle. Lead with the task; attach context after.
- Format as an afterthought: asking for a table AFTER the essay arrives. Front-load it and save a turn.
Your Turn — The 60-Second Drill
Five parts. Zero guesses. 🧩
Keep building the skill — the full roadmap has 60+ free guides.
Frequently Asked Questions
Role (who the AI should be), Task (the specific action), Context (background and audience), Format (the shape of the answer), and Constraints (rules and limits). Together they remove every guess the AI would otherwise make.
No — simple tasks work with just a clear Task. But each part you add sharpens the output, and for anything important, all five take under a minute to write.
Context, for most people — because it’s the part most often missing entirely. A role plus two lines of context typically transforms a generic answer into a personal one.
Match the role to the job: an expert for depth, a teacher for explanations, a critic for feedback, even ‘a skeptical customer’ for testing your ideas. The role is a lens, not a rank.
No — a complete prompt is better. Length without information (padding, politeness, repetition) does nothing. Sixty precise words beat three hundred vague ones.
After the task and context, before or alongside the format. When you include a good example, it often replaces most of the format section — the AI copies the pattern it sees
