Chapter 4: Prompt Architecture Patterns

Structural Frameworks for Effective Prompt Design

---

Learning Objectives

After completing this chapter, you will be able to:

1. Describe the anatomy of a well-structured prompt
2. Apply common prompt architecture patterns (CRAFT, RISEN, CO-STAR)
3. Select the appropriate pattern for different use cases
4. Compose prompts using building blocks
5. Create reusable prompt templates

---

The Anatomy of Effective Prompts

Structural Overview

Every effective prompt shares common structural elements, though not all elements are required for every prompt:

Figure 4.1: The five structural components of effective prompt architecture, showing the recommended sequence from preamble to output specification.

Component Details

Component	Purpose	Required?
Preamble	Set global behavior and persona	Optional
Context	Provide necessary background	Usually
Instruction	Define the task	Always
Input Data	Supply content to process	When applicable
Output Spec	Control response format	Recommended

---

The CRAFT Framework

Overview

CRAFT is a versatile prompt architecture pattern suitable for most everyday prompting needs.

C - Context
R - Role
A - Action
F - Format
T - Tone

CRAFT Components

Letter	Component	Description
C	Context	Background information and situation
R	Role	Who the AI should act as
A	Action	The specific task to perform
F	Format	How to structure the output
T	Tone	The voice and style to use

CRAFT Example

**Context:**
Our company is launching a new project management tool designed for
small creative agencies (5-20 employees). We're targeting design
studios, ad agencies, and content production houses.

**Role:**
You are an experienced product marketing specialist with expertise
in B2B SaaS marketing for creative industries.

**Action:**
Write a landing page headline and three supporting value propositions
that will resonate with our target audience.

**Format:**
- One headline (maximum 10 words)
- Three value propositions (each 15-20 words)
- Each value proposition should address a specific pain point

**Tone:**
Professional but creative. Use language that appeals to designers
and creatives—avoid corporate jargon.

When to Use CRAFT

Use Case	Suitability
Content creation	Excellent
Marketing copy	Excellent
General writing	Good
Technical documentation	Good
Code generation	Moderate
Analysis tasks	Moderate

---

The RISEN Framework

Overview

RISEN provides more structure for complex tasks requiring clear process steps.

R - Role
I - Instructions
S - Steps
E - End goal
N - Narrowing (constraints)

RISEN Components

Letter	Component	Description
R	Role	Persona or expertise to embody
I	Instructions	High-level task description
S	Steps	Specific process to follow
E	End goal	Desired outcome definition
N	Narrowing	Constraints and limitations

RISEN Example

**Role:**
You are a senior data analyst specializing in e-commerce analytics.

**Instructions:**
Analyze the provided customer purchase data to identify high-value
customer segments for a targeted marketing campaign.

**Steps:**
1. Calculate RFM scores (Recency, Frequency, Monetary) for each customer
2. Cluster customers into 4-5 segments based on RFM scores
3. Profile each segment with descriptive statistics
4. Rank segments by potential marketing ROI
5. Recommend targeting priorities

**End Goal:**
A clear recommendation of which 2-3 customer segments to prioritize
for our upcoming marketing campaign, with justification.

**Narrowing:**
- Focus on customers active in the last 12 months
- Budget constraint: can only target 30% of customer base
- Campaign goal is increasing repeat purchases, not acquisition
- Output should be actionable by non-technical marketing team

When to Use RISEN

Use Case	Suitability
Multi-step analysis	Excellent
Process-oriented tasks	Excellent
Research projects	Good
Technical procedures	Good
Simple questions	Overkill

---

The CO-STAR Framework

Overview

CO-STAR emphasizes the communication aspects of prompting.

C - Context
O - Objective
S - Style
T - Tone
A - Audience
R - Response format

CO-STAR Components

Letter	Component	Description
C	Context	Relevant background information
O	Objective	What you want to achieve
S	Style	Writing style (formal, casual, etc.)
T	Tone	Emotional register (friendly, serious)
A	Audience	Who will read/use the output
R	Response	Desired format and structure

CO-STAR Example

**Context:**
Our healthcare startup just received FDA clearance for our AI-powered
diagnostic device. This is our first major regulatory milestone.

**Objective:**
Write a press release announcing this achievement that will generate
media coverage and build credibility with potential investors.

**Style:**
Professional business journalism style. Factual and authoritative
with strategic quotes.

**Tone:**
Confident and optimistic, but not overly promotional. Emphasize
scientific rigor and patient safety.

**Audience:**
- Primary: Healthcare industry journalists
- Secondary: Potential investors and partners
- Tertiary: Healthcare providers who might adopt the technology

**Response:**
Standard press release format:
- Headline
- Dateline and lead paragraph
- 2-3 body paragraphs with key details
- Quote from CEO
- Boilerplate about company
- Media contact information

When to Use CO-STAR

Use Case	Suitability
PR and communications	Excellent
Audience-specific content	Excellent
Marketing materials	Good
Educational content	Good
Technical writing	Moderate

---

The RTF Framework

Overview

RTF is a minimalist framework for quick, focused prompts.

R - Role
T - Task
F - Format

RTF Example

**Role:** Technical writer

**Task:** Explain how HTTPS encryption works

**Format:** 5 bullet points suitable for non-technical readers

When to Use RTF

Best for:

• Quick tasks
• Simple requests
• When context is minimal
• Rapid prototyping prompts

---

Pattern Selection Guide

Decision Matrix

Need	Recommended Pattern
Quick, simple request	RTF
Content with specific voice	CRAFT
Multi-step process	RISEN
Audience-focused communication	CO-STAR

Selection Flowchart

Figure 4.2: Decision flowchart for selecting the appropriate prompting framework based on task characteristics.

---

Building Blocks Approach

Modular Prompt Components

Rather than always using complete frameworks, you can build prompts from modular components:

Core Blocks

Block	Purpose	Example
Role Block	Set expertise	“You are a [role]…”
Context Block	Provide background	“Background: [context]”
Task Block	Define action	“Your task is to [action]”
Input Block	Supply data	“Given the following: [data]”
Format Block	Specify output	“Format as: [format]”
Constraint Block	Set boundaries	“Constraints: [limits]”
Example Block	Provide samples	“Example: [example]”

Block Composition

Figure 4.3: Modular building blocks combine to create prompts of varying complexity, from simple single-block prompts to comprehensive seven-block structures.

---

Template Development

Creating Reusable Templates

Templates allow you to standardize prompts for recurring tasks:

Template Structure

# [Template Name]

## Purpose
[What this template is for]

## Variables
- {variable_1}: [Description]
- {variable_2}: [Description]

## Template

---
[Your role is {role}.]

Context: {context}

Task: {task}

Requirements:
- {requirement_1}
- {requirement_2}

Format: {output_format}
---

## Usage Example
[Filled-in example]

Example Template: Code Review

Purpose: Systematic code review with consistent criteria

Variables:

• {language}: Programming language
• {code}: Code to review
• {focus_areas}: Specific concerns

Template:

You are a senior {language} developer conducting a code review.

Review the following code for:
1. Correctness and potential bugs
2. Performance issues
3. Security vulnerabilities
4. Code style and readability
5. {focus_areas}

Code to review:
[Insert code here]

Provide your review as:
- Issues Found: Numbered list with severity (Critical/Major/Minor)
- Positive Observations: What's done well
- Recommendations: Specific improvements

Usage Example (with variables filled in):

Variable	Value
{language}	Python
{code}	def process_user_data(data): result = eval(data['formula'])...
{focus_areas}	Error handling patterns

---

Anti-Patterns to Avoid

Common Architectural Mistakes

Anti-Pattern	Problem	Solution
Wall of Text	No structure, hard to parse	Use clear sections
Missing Context	AI guesses incorrectly	Provide relevant background
Vague Instructions	Unpredictable results	Be specific and direct
No Format Spec	Unusable output	Specify format explicitly
Kitchen Sink	Overwhelming, confused focus	Focus on essentials

Before and After

Before (Anti-pattern):

I need you to help me write something for my boss about the project
we've been working on that shows the results and makes us look good
but is also honest and professional and not too long.

After (Structured):

**Role:** Business communications specialist

**Context:**
We completed a 3-month software modernization project that:
- Reduced page load times by 40%
- Decreased server costs by 25%
- Some features were delayed to Phase 2

**Task:** Write a project summary for executive leadership

**Format:**
- 200-250 words
- Executive summary style
- Include: achievements, metrics, lessons learned
- Tone: Professional, balanced, factual

**Constraint:** Acknowledge the delayed features without dwelling on them

---

Key Takeaways

• Prompt architecture provides structure for effective communication
• CRAFT, RISEN, CO-STAR, and RTF offer different approaches for different needs
• Select patterns based on task complexity and requirements
• Building blocks allow flexible prompt composition
• Templates enable consistency and reusability
• Avoid anti-patterns like walls of text and vague instructions

---

Summary

Prompt architecture patterns transform ad-hoc prompting into a systematic discipline. By using frameworks like CRAFT, RISEN, and CO-STAR, you ensure your prompts include all necessary elements in a logical structure. Choose patterns based on your specific needs, and don’t hesitate to adapt them or build from modular components. Well-architected prompts consistently produce better results than unstructured requests.

---

Review Questions

1. What are the five components of the CRAFT framework?
2. When would you choose RISEN over CRAFT?
3. What distinguishes CO-STAR from other frameworks?
4. What is the purpose of the “Narrowing” component in RISEN?
5. Name three prompt anti-patterns and their solutions.

---

Practical Exercise

Exercise 4.1: Pattern Application

Choose a task you commonly perform with AI and write three versions of the prompt using:

1. CRAFT framework
2. RISEN framework
3. RTF framework

Compare the results and note which produced the best output for your specific task.

Exercise 4.2: Template Creation

Create a reusable template for a task you perform frequently. Include:

• Clear purpose statement
• Variable definitions
• Complete template structure
• One filled-in example

---

Chapter Navigation

Previous	Up	Next
← Chapter 3: Prompting Pillars	Part II: Core Frameworks	Chapter 5: Context Engineering →