The anatomy of a great prompt: 8 layers AI needs
Updated:
When an AI gives you a bland, generic answer, the model usually isn’t the problem. The prompt left it guessing. A great prompt isn’t a clever one-liner. It’s a complete brief, the kind you’d give a smart new hire on their first day.
A useful way to make sure nothing is missing is to think of a prompt as eight layers. Not every task needs all eight, but every layer you skip is a decision the AI makes for you.
1. Role: who the AI should be
A role shifts the vocabulary, the depth and, most of all, the judgment behind the answer. “You are a copywriter with 10 years of experience writing for independent coffee roasters” beats “You are an expert” every time. We cover this in how to give AI a role.
2. Context: your situation
This is the layer most prompts are missing. What you do, who the result is for, how it’ll be used, what you’ve already tried. Three or four real facts are usually enough. If a constraint matters, say why. Models apply rules better when they understand the reason behind them.
3. Goal: the task and the deliverable
A clear verb and a concrete output. “Help me with my website” is a topic. “Write the copy for the pricing section, covering three plans” is a task. Naming the deliverable (three options, a final draft, a checklist) stops the AI from choosing the format for you.
4. Steps or specific questions
If you were the expert, what would you check first? Those questions become instructions: “Lead with the customer’s problem,” “Compare Q1 to Q2,” “Explain each part of the code.” Number them if order matters.
5. Examples
Show, don’t describe. Paste something you like and say what to take from it: the tone, the structure, the length. No examples? Ask for two contrasting drafts and pick one. More in examples in prompts.
6. Constraints
Word limits, tone, budget, things that must not change. Phrase them positively when you can: “Use short sentences” is clearer than “Don’t ramble.”
7. Format
Table, bullet list, paste-ready text, one code block per file. If you don’t ask, the AI picks, and you end up reformatting.
8. Success criteria
How you’ll judge the result. “Makes sense on a phone screen in five seconds.” “Every claim is backed by a number from the data.” Criteria give the AI a checklist to review its own work before handing it over.
A worked example
The original request:
Write a LinkedIn post about our new feature.With all eight layers:
You are a B2B content strategist who writes for small SaaS companies.
We make scheduling software for dental clinics. Next week we launch
automatic appointment reminders by text. Our followers are clinic
managers, not dentists, and they're tired of no-shows.
Write 3 versions of a LinkedIn post announcing the feature.
- Open with the cost of a no-show, not with the product.
- Mention that setup takes under 10 minutes.
- End with a question that invites comments.
Here's a post of ours that performed well. Match the tone, not the topic:
"Your front desk answered 40 calls yesterday. How many were
'just checking my appointment time'?"
Under 120 words each. No hashtags beyond #DentalPractice.
Return them as a numbered list.
A good version makes a clinic manager think "that's us."Longer, yes. But every line answers a question the AI would otherwise have guessed at, and that’s why it lands on the first try.
Practice without memorizing
You don’t need to remember all eight layers. The prompt coach asks one question per layer, with options to tap, and builds the prompt alongside you. After a few rounds, you’ll start writing them on your own.