Anatomy of a Good Prompt

Every effective prompt has key structural components that work together like architectural pillars. Understanding how these parts interact at the model level helps you craft better instructions consistently.

The Five Core Components

A well-structured prompt contains five core blocks:

1. Role/Persona - Sets the AI's domain expertise, vocabulary, and perspective.
2. Context - Provides background facts, data, target audience, and environment.
3. Task - The specific, action-oriented request (the core verb).
4. Format - The exact output structure (JSON, Markdown table, bulleted points, word count).
5. Constraints - Negative and positive boundaries (what to avoid, style requirements, rules).

┌─────────────────────────────────────────────────────────┐
│ 1. ROLE: You are an executive speechwriter              │
├─────────────────────────────────────────────────────────┤
│ 2. CONTEXT: Company is announcing Q3 earnings (+14% YoY) │
├─────────────────────────────────────────────────────────┤
│ 3. TASK: Write opening remarks for the CEO              │
├─────────────────────────────────────────────────────────┤
│ 4. FORMAT: 3 short paragraphs with headline subheadings  │
├─────────────────────────────────────────────────────────┤
│ 5. CONSTRAINTS: Under 250 words, confident tone         │
└─────────────────────────────────────────────────────────┘

Why Delimiters and Tagging Matter

When prompts grow long, AI models can confuse user data with system instructions. Modern prompt engineering uses structural delimiters (such as XML tags or Markdown headers) to cleanly divide sections.

<role>You are a senior UX researcher.</role>

<context>
User research interviews show 65% of new users drop off during account creation.
</context>

<task>
Analyze the drop-off issue and propose 3 onboarding design fixes.
</task>

<format>
Numbered list with: (1) Proposed Fix, (2) Expected UX Impact, (3) Implementation Effort.
</format>

<constraints>
Focus on low-friction changes. Do not suggest third-party OAuth integrations.
</constraints>

Using XML tags like <context> or <constraints> prevents the AI's attention mechanism from leaking context into instructions.

Visualizing Prompt Structure in Action

Here is the 5-component framework applied to real-world scenarios across different domains:

Example 1: Travel Planning

Example 2: Business & Marketing

Example 3: Software Development

Click "Analyze" on any example to see how the prompt hits all key components.

Deep Component Breakdown

1. Role/Persona

Directs the LLM to prime specific latent knowledge weights in its parameters:

• "You are an expert data scientist..." → Primes technical statistical vocabulary.
• "Act as a patient elementary school teacher..." → Primes simple explanations and warm tone.

2. Context

Grounds the prompt and prevents hallucinations by anchoring the query to explicit facts:

• The situation, user profile, constraints, or raw source text.

3. Task

The direct imperative action verb:

• Use active verbs: "Analyze", "Extract", "Synthesize", "Audit", "Draft".
• Avoid vague directives like "Look over" or "Help me with".

4. Format

Dictates the physical token layout:

• Specify JSON schemas, Markdown tables, character limits, or section headers.

5. Constraints

Sets strict guardrails:

• Explicit rules on tone, banned vocabulary, length limits, or compliance requirements.

Exercise: Build a Complete Prompt

Common Prompt Structure Failures

Failure Pattern	What Goes Wrong	How to Fix
Missing Role	Generic, low-effort corporate response	Add "You are a [specialist] with expertise in [domain]"
Vague Task	Wall of unorganized text	Use explicit action verbs ("Identify 3 key risks")
No Format Spec	Paragraphs when you needed code or table	Define exact format: "Output as a Markdown table"
Conflicting Constraints	Model ignores rules or produces truncated text	Resolve contradictions (e.g. don't demand "detailed depth" in "30 words")

Quick Reference Template

Practice: Analyze This Prompt

What's missing from the prompt above? Think about:

• What role would improve the response?
• What context is the AI missing? (Dates, reason, coverage plan)
• What format would be most useful?
• What constraints would help? (Tone, word count limit)

Understanding prompt anatomy is the foundation for everything else in this course.