← Back to Claude Mastery
Claude Mastery · Chapter 7 of 30
07

Advanced Prompting Patterns

Professional-Grade Techniques

⏱️ 14 min read 📊 Advanced 🎯 Expert Techniques

Introduction

You've mastered the fundamentals. You know how to be clear, use examples, encourage thinking, refine iteratively, and provide context.

Now it's time to level up.

This chapter introduces advanced prompting patterns that professional AI engineers use for sophisticated tasks. These aren't just "tips and tricks" — they're systematic techniques that unlock Claude's full potential for complex work.

Think of the difference between cooking from a recipe and understanding the culinary techniques that let you improvise. The fundamentals got you following recipes. These advanced patterns let you create your own.

What you'll learn:

Ready? Let's dive into the advanced playbook.

Pattern 1: Chain-of-Thought (CoT) Prompting

What it is: Explicitly asking Claude to show its reasoning process step-by-step before giving an answer.

Why it works: Forces systematic thinking, reduces errors, makes logic transparent, and improves accuracy on complex problems.

Basic Format

[Your question] Think through this step-by-step before answering: 1. First, [step 1] 2. Then, [step 2] 3. Finally, [step 3] After thinking through each step, provide your answer.

Real Example

Without Chain-of-Thought:

A store sells apples for $2 each. If you buy 5 or more, you get 20% off. You buy 6 apples. How much do you pay?

Claude might jump to an answer, potentially making an error.

With Chain-of-Thought:

A store sells apples for $2 each. If you buy 5 or more, you get 20% off. You buy 6 apples. How much do you pay? Think through this step-by-step: 1. Calculate the base price (price × quantity) 2. Determine if the discount applies 3. Calculate the discount amount 4. Calculate the final price Show your work for each step, then give the final answer.

Claude's response now shows:

Final answer: $9.60

The reasoning is transparent, verifiable, and more likely to be correct.

When to Use Chain-of-Thought

Pattern 2: Constrained Outputs

What it is: Specifying the exact format, structure, or constraints for the response.

Why it works: Ensures Claude's output fits directly into your workflow without manual reformatting.

Constrain by Format

Example:

List 5 productivity tools for remote teams. Requirements: - JSON format - Each tool must include: name, category, price_range, best_for - Sort by price (lowest to highest) - Price ranges must be: "free", "budget", "premium"

Result: Clean JSON you can paste directly into code.

Constrain by Length

Example:

Explain quantum computing. Constraints: - Exactly 3 paragraphs - Paragraph 1: What it is (50 words max) - Paragraph 2: How it works (75 words max) - Paragraph 3: Real applications (50 words max)

Result: Precise length for a specific context (website copy, presentation slide, etc.)

Constrain by Structure

Example:

Analyze this product launch strategy. Provide your analysis in this exact format: STRENGTHS (3 bullet points max) WEAKNESSES (3 bullet points max) OPPORTUNITIES (2 bullet points max) THREATS (2 bullet points max) RECOMMENDATION (1 paragraph, 100 words max)

Result: Structured SWOT analysis ready to present.

Industry Example: Marketing Brief

Create a campaign brief for our Q3 product launch. OUTPUT FORMAT: Campaign Name: [single line] Target Audience: [2-3 sentences] Key Message: [1 sentence, under 20 words] Channels: [bullet list, 3-5 items] Budget Allocation: [table format with percentages] Timeline: [week-by-week breakdown, 8 weeks] Success Metrics: [numbered list, 4-5 metrics] Keep total output under 400 words.

Pattern 3: XML Tags for Structure

What it is: Using XML-like tags to organize complex prompts with multiple sections.

Why it works: Clear separation of different information types, easy for Claude to parse, handles complexity elegantly.

Basic Structure

<context> [Background information] </context> <task> [What you want Claude to do] </task> <data> [Input data or examples] </data> <constraints> [Rules and limitations] </constraints> <output_format> [How to structure the response] </output_format>

Real Example: Investment Analysis

<context> I'm a startup founder raising a Series A. We have $2M ARR, growing 15% MoM. Enterprise SaaS in the HR tech space. </context> <task> Analyze this investment opportunity using systematic reasoning. </task> <data> - Current ARR: $2M - Growth rate: 15% month-over-month - Burn rate: $400K/month - Team size: 25 people - Market size: $10B TAM - Competitors: 3 major players, market leader at $50M ARR </data> <analysis_framework> STEP 1: Market Analysis - Market size and growth trajectory - Competitive landscape assessment - Defensibility and moat STEP 2: Financial Health - Unit economics evaluation - Growth sustainability analysis - Path to profitability timeline STEP 3: Risk Assessment - Market risks - Execution risks - Financial risks STEP 4: Valuation - Comparable companies analysis - Growth-adjusted valuation - Recommended valuation range </analysis_framework> <output_format> For each step: - Analysis (show your reasoning) - Key findings (3-5 bullet points) - Risk score (1-10 scale) Final recommendation: - Decision: INVEST / PASS / WATCH - Rationale (2-3 sentences) - Conditions (if applicable) </output_format>

This combines:

Pattern 4: Few-Shot Prompting

What it is: Providing 2-4 examples of the pattern you want Claude to follow.

Why it works: Shows Claude exactly what you mean, better than explaining. Especially powerful for style, format, or tone.

The Pattern

Convert [INPUT] into [OUTPUT]. Examples: INPUT: [Example 1 input] OUTPUT: [Example 1 output] INPUT: [Example 2 input] OUTPUT: [Example 2 output] INPUT: [Example 3 input] OUTPUT: [Example 3 output] Now convert these: [Your actual inputs]

Real Example: Feature → Benefit Copy

Convert product features into benefit-focused marketing copy. Examples: Feature: "256-bit encryption" Benefit: "Your data stays private and secure, protected by bank-level encryption" Feature: "Real-time sync across devices" Benefit: "Start on your phone, finish on your laptop—your work is always up to date" Feature: "Automated backups every hour" Benefit: "Never lose your work. We save your progress automatically, so you can focus on creating" Now convert these features: - Unlimited cloud storage - Advanced analytics dashboard - API access with 99.9% uptime - Custom branding options

Claude learns the pattern: technical feature → user-focused benefit.

When Few-Shot Shines

Pattern 5: Role Assignment

What it is: Asking Claude to adopt a specific role or perspective before responding.

Why it works: Frames the response from an expert viewpoint with appropriate depth and terminology.

The Pattern

<role> You are a [specific role] with [X years] experience in [domain]. Your expertise includes [key areas]. </role> <task> [Your question or request] </task>

Real Example

<role> You are a senior DevOps engineer with 10 years of experience scaling cloud infrastructure. Your expertise includes Kubernetes, AWS, CI/CD pipelines, and cost optimization. </role> <task> Our API response times increased from 200ms to 2000ms after deploying a new microservice. Walk me through your debugging approach. </task>

Claude now responds with the depth, terminology, and systematic approach of an experienced DevOps engineer.

Pattern 6: Constrain-Then-Expand

What it is: First get a constrained/simplified response, then progressively expand with detail.

Why it works: Ensures clarity before complexity, creates scaffolding for understanding.

The Pattern

LEVEL 1 (Simplest): Explain [topic] in [constraint: 2 sentences for a 10-year-old] LEVEL 2 (Moderate): Now expand that explanation for [target: college student] in [constraint: 1 paragraph] LEVEL 3 (Advanced): Finally, provide a technical explanation for [target: professional] in [constraint: 2-3 paragraphs with terminology]

Real Example: Machine Learning Explanation

Explain machine learning. LEVEL 1: Give me a 2-sentence explanation suitable for a 10-year-old. LEVEL 2: Expand that explanation for a college student (1 paragraph). LEVEL 3: Provide a technical explanation for a software engineer (2-3 paragraphs with proper terminology). Build each level on the previous one.

Result: Three explanations, each adding appropriate complexity.

Pattern 7: Conditional Logic

What it is: Asking Claude to handle different scenarios differently based on conditions.

Why it works: One prompt adapts to various situations, reduces back-and-forth.

The Pattern

Analyze [input]. IF [condition A]: - [Action A1] - [Action A2] IF [condition B]: - [Action B1] - [Action B2] IF [condition C]: - [Action C1] - [Action C2] [Input here]

Real Example: Code Review

Review this code for bugs and optimization opportunities. IF the code has bugs: - Identify each bug - Explain why it's a bug - Provide corrected code IF the code is correct but inefficient: - Explain the inefficiency - Suggest optimization approach - Show optimized version with comments IF the code is correct and efficient: - Confirm it's well-written - Suggest minor improvements (if any) - Provide edge cases to test [Your code here]

This ensures Claude provides the RIGHT type of feedback for the situation.

Combining Patterns: Expert-Level Prompting

The real power comes from combining multiple patterns strategically.

Example: Comprehensive Business Analysis

<role> You are a senior strategy consultant with 15 years experience in SaaS. </role> <context> Client: B2B marketing automation platform, $5M ARR, 200 customers. Challenge: Considering expansion into SMB market (currently enterprise-focused). </context> <task> Analyze this market expansion opportunity using chain-of-thought reasoning. </task> <analysis_framework> STEP 1: Market Assessment Think through: - SMB market size vs current TAM - Competitive landscape in SMB - Buyer behavior differences STEP 2: Business Model Impact Think through: - Pricing implications (enterprise vs SMB) - Sales motion changes (high-touch vs self-serve) - Support requirements STEP 3: Operational Feasibility Think through: - Product adaptations needed - Team/hiring implications - Timeline to market STEP 4: Financial Modeling Think through: - Revenue projections (3-year) - Cost structure changes - Break-even timeline For each step, show your reasoning process before conclusions. </analysis_framework> <output_format> For each step above: ANALYSIS (your thinking) - [Bullet points showing reasoning] KEY FINDINGS - [3-5 critical insights] RISK LEVEL: [LOW / MEDIUM / HIGH] After all 4 steps, provide: FINAL RECOMMENDATION Decision: [EXPAND / DON'T EXPAND / PHASE APPROACH] Rationale: [2-3 sentences] Conditions: [If conditional, what must be true] Next Steps: [3-5 immediate actions] </output_format>

This combines:

When to Use Advanced Patterns

Use Advanced Patterns For:

Stick to Simple Prompts For:

Rule: Match prompt complexity to task complexity.

Common Mistakes with Advanced Patterns

Mistake 1: Over-Engineering Simple Tasks

❌ Using XML tags and chain-of-thought for "What's the capital of France?" ✅ Just ask the question

Mistake 2: Unclear Constraints

❌ "Make it concise" ✅ "Maximum 100 words, 2 paragraphs"

Mistake 3: Contradictory Requirements

❌ "Be comprehensive but brief, detailed but concise" ✅ "Cover these 3 points in 150 words total"

Mistake 4: Too Many Patterns at Once

❌ Trying to use 5 different patterns in one prompt ✅ Pick 2-3 that complement each other
Key Takeaways
Your Turn: Assignment

Challenge: Pick a complex task you need to do. Design an advanced prompt using at least 2 patterns from this chapter.

Requirements:

  1. Choose appropriate patterns for your task
  2. Structure the prompt clearly
  3. Test it with Claude
  4. Refine based on results
  5. Compare to how you would have asked before

Reflection Questions:

Share: What patterns worked best for your use case?