Advanced Prompt Patterns: Combining Techniques for Maximum Impact
You've learned the individual techniques: role prompting, chain-of-thought, few-shot examples, output formatting, the COSTAR framework. Now it's time to combine them. The real power of prompt engineering emerges when you stack these techniques thoughtfully for complex tasks.
This guide shows you how to combine techniques, when to use which combinations, and provides patterns you can apply immediately.
Why Combinations Matter
Individual techniques solve specific problems:
- Role prompting shapes perspective
- Chain-of-thought improves reasoning
- Few-shot examples demonstrate format
- Output formatting controls structure
- Specificity eliminates ambiguity
Complex tasks have multiple requirements. A single technique rarely addresses everything. But stack the right techniques together? You get outputs that are expert-level, well-reasoned, properly formatted, and precisely targeted.
The Art of Technique Stacking
Not every technique belongs in every prompt. The key is matching techniques to needs:
| If you need... | Use... |
|---|---|
| Expert perspective | Role prompting |
| Careful reasoning | Chain-of-thought |
| Specific format/style | Few-shot examples |
| Structured output | Format specifications |
| Relevant focus | Context/specificity |
| Complete coverage | COSTAR framework |
For complex tasks, identify what you need most, then layer techniques accordingly.
Pattern 1: Role + Chain-of-Thought
Use when: You need expert reasoning you can follow and verify.
Structure:
You are a [specific expert] with [relevant experience]. I need help with [your challenge]. Think through this step by step: 1. First, [initial analysis step] 2. Then, [deeper analysis] 3. Next, [consider alternatives] 4. Finally, [reach conclusion] Show your expert reasoning at each step.
Example:
You are a CFO with 20 years of experience in SaaS businesses, known for making sound financial decisions under uncertainty. I need to decide whether to raise our next funding round now (in a down market) or wait 6 months hoping conditions improve. Think through this step by step: 1. First, what factors should drive this timing decision? 2. Then, analyze the risks of each option 3. Next, consider what signals would indicate the right choice 4. Finally, what would you recommend and why? Show your reasoning at each step so I can follow your logic.
Why it works: The role ensures relevant expertise colors every step. The chain-of-thought prevents jumping to conclusions and lets you verify the reasoning.
Pattern 2: Few-Shot + Formatting
Use when: You need consistent, specific output format that's hard to describe.
Structure:
[Task description] Here are examples of exactly what I need: Example 1: Input: [example input 1] Output: [example output 1] Example 2: Input: [example input 2] Output: [example output 2] Now follow the same format exactly for: Input: [your actual input] Output:
Example:
Convert customer feedback into structured product insights. Here are examples of exactly what I need: Example 1: Input: "Your mobile app crashes every time I try to upload a photo. Very frustrating!" Output: - Category: Bug Report - Feature: Mobile App > Photo Upload - Severity: High (blocks functionality) - User Emotion: Frustrated - Action: Investigate mobile upload crash; prioritize fix Example 2: Input: "Would be nice if I could export reports to PDF" Output: - Category: Feature Request - Feature: Reporting > Export - Severity: Low (workaround exists) - User Emotion: Neutral - Action: Add to feature backlog; assess demand Now follow the same format exactly for: Input: "The new dashboard is confusing. I can't find where my saved filters went." Output:
Why it works: The examples demonstrate both format AND the level of analysis expected. Format specs alone couldn't communicate the nuances as clearly.
Pattern 3: COSTAR + Chain-of-Thought
Use when: Complex tasks requiring complete context AND careful reasoning.
Structure:
CONTEXT: [Background information] OBJECTIVE: [What you want to achieve] STYLE: [How it should be written/presented] TONE: [The emotional quality] AUDIENCE: [Who will consume this] RESPONSE: [Format specifications] --- Now, think through this systematically: Step 1: [First consideration] Step 2: [Second consideration] Step 3: [Analysis/synthesis] Step 4: [Conclusion/output] Work through each step before providing your final response.
Example:
CONTEXT: I'm the VP of Engineering at a 100-person company. We're evaluating whether to adopt AI coding assistants company-wide. Current state: 10 developers piloting GitHub Copilot with mixed reviews. Concerns about security, code quality, and dependency. OBJECTIVE: Make a recommendation to the executive team with clear justification. STYLE: Executive memo format—clear, evidence-based, actionable. TONE: Balanced and thoughtful—acknowledge both benefits and risks. AUDIENCE: CEO and executive team who are non-technical but data-driven. RESPONSE: - Executive summary (3-4 sentences) - Analysis section with headers - Recommendation with conditions - Next steps - 500-700 words total --- Now, think through this systematically: Step 1: What are the key factors to evaluate for this decision? Step 2: What does the pilot data suggest about each factor? Step 3: What risks need to be addressed before broader adoption? Step 4: What's the right recommendation given this analysis? Work through each step, then provide the executive memo.
Why it works: COSTAR ensures you don't miss context that shapes the answer. Chain-of-thought ensures the reasoning is sound before the conclusion is written.
Pattern 4: Role + Few-Shot + Formatting
Use when: You need expert-quality output in a very specific format.
Structure:
You are a [specific expert role]. I need you to [task] in a specific format. Here's an example of what I'm looking for: Example: [Example input/scenario] [Example output in exact format needed] Notice: - [What to observe about the format] - [What to observe about the approach] Now, apply your expertise to: [Your actual request]
Example:
You are a professional copywriter who specializes in B2B SaaS website copy. You're known for being concise, benefit-focused, and conversion-oriented. I need you to write feature descriptions. Here's an example of what I'm looking for: Example: Feature: Automated reporting Copy: **Stop Building Reports. Start Using Them.** Your weekly report takes 4 hours to build and 10 minutes to read. Our AI compiles your data, surfaces insights, and delivers polished reports to your inbox—every Monday at 8 AM. You review in minutes, not hours. ✓ Automatic data pulling from 50+ sources ✓ Customizable templates that match your brand ✓ Scheduled delivery to stakeholders Notice: - Punchy headline that hints at the transformation - Pain point acknowledged, then solution - Specific, concrete benefit - Feature checkmarks at the end Now, apply your expertise to: Feature: Real-time collaboration [Write the copy]
Why it works: The role brings expertise to the writing. The example demonstrates exact format AND quality bar. Together they produce polished, on-brand copy.
Pattern 5: Iterative Refinement
Use when: The task is too complex for a single prompt, or quality matters enough to iterate.
Structure:
Prompt 1 (Generate):
[Initial request with core requirements] Provide a first draft focusing on [primary criteria].
Prompt 2 (Critique):
Review this draft against these criteria: [List of criteria] Identify: 1. What's working well 2. What needs improvement 3. Specific suggestions for revision
Prompt 3 (Refine):
Revise the draft based on this feedback: [Critique from Prompt 2] Specifically: - [Improvement 1] - [Improvement 2] - [Improvement 3] Provide the revised version.
Example sequence:
Prompt 1:
Write a 200-word product announcement for our new AI-powered scheduling feature. Target: busy professionals. Focus on time savings.
Prompt 2:
Review this announcement: [Draft from Prompt 1] Evaluate against: - Clarity of value proposition - Emotional resonance - Call to action strength - Appropriate length What works? What needs improvement?
Prompt 3:
Revise based on this feedback: [Critique from Prompt 2] Make the value prop more concrete, strengthen the emotional hook, and make the CTA more urgent.
Why it works: Breaking complex tasks into generate-critique-refine cycles produces higher quality than trying to get it perfect in one shot.
Pattern 6: Multi-Perspective Analysis
Use when: You need balanced analysis of complex decisions or controversial topics.
Structure:
I need to analyze [topic/decision] from multiple perspectives. Please respond in three sections: ## Perspective 1: [Stakeholder/viewpoint A] Assume you are [this perspective]. What matters most? What would you advocate for and why? ## Perspective 2: [Stakeholder/viewpoint B] Assume you are [this perspective]. What matters most? What would you advocate for and why? ## Perspective 3: [Stakeholder/viewpoint C] Assume you are [this perspective]. What matters most? What would you advocate for and why? ## Synthesis Now step back and analyze: Where do these perspectives align? Where do they conflict? What's the path forward that best addresses all concerns?
Example:
I need to analyze our decision to switch from annual to monthly contracts from multiple perspectives. ## Perspective 1: Sales Team Assume you are the VP of Sales. What matters most? How does this change affect your team and targets? ## Perspective 2: Finance Team Assume you are the CFO. What matters most? How does this change affect revenue recognition and forecasting? ## Perspective 3: Customer Success Team Assume you are the VP of Customer Success. What matters most? How does this change affect retention and expansion? ## Synthesis Now step back and analyze: Where do these perspectives align? Where do they conflict? What's the path forward that best addresses all concerns?
Why it works: Forcing explicit perspective-taking surfaces considerations you might miss. The synthesis step integrates the insights into actionable guidance.
Choosing the Right Combination
| Scenario | Recommended Pattern |
|---|---|
| Strategic advice | Role + Chain-of-Thought |
| Consistent content creation | Few-Shot + Formatting |
| Complex decisions | COSTAR + Chain-of-Thought |
| High-quality creative | Role + Few-Shot + Formatting |
| Important documents | Iterative Refinement |
| Controversial topics | Multi-Perspective Analysis |
Common Mistakes
❌ Mistake: Using all techniques for every prompt ✅ Fix: Match techniques to actual needs—simpler prompts for simpler tasks
❌ Mistake: Techniques contradict each other ✅ Fix: Ensure role, tone, and examples all align
❌ Mistake: So many elements that the AI gets confused ✅ Fix: Prioritize and streamline—clear beats comprehensive
❌ Mistake: Not adjusting when combinations don't work ✅ Fix: Remove elements one at a time to find what's causing issues
Practice Exercise
Try building a combined prompt for this scenario:
Task: Create a presentation talking points document for announcing a company restructuring to employees.
Consider what you need:
- Expert perspective (HR, executive communication)?
- Careful reasoning (sensitive topic)?
- Specific format (talking points structure)?
- Example to demonstrate style?
- Multiple perspectives (manager vs. IC audience)?
Build your prompt using the appropriate combination:
[Your combined prompt here]
Compare your result to this approach:
You are an experienced HR communications director who has guided companies through multiple restructurings. You're known for messages that are honest, empathetic, and clear. CONTEXT: We're announcing a restructuring that eliminates 20 positions (10% of staff), reorganizes two departments, and promotes 3 people to new leadership roles. Affected employees will be notified privately before the announcement. OBJECTIVE: Create talking points for leaders to use when announcing this to their teams. AUDIENCE: Team leaders who are nervous about delivering this message. Their teams will have mixed reactions. Think through what makes restructuring communications succeed or fail: 1. What must be communicated clearly? 2. What questions will employees have? 3. What tone threads the needle between honest and stable? Then provide: ## Opening (2-3 sentences to set context) ## Key Messages (5 bullets to cover) ## Expected Questions & Answers (5 Q&As) ## What NOT to Say (3-4 things to avoid) ## Closing (how to end the conversation)
Key Takeaways
- Single techniques solve single problems; complex tasks need combinations
- Role + Chain-of-Thought for expert reasoning you can verify
- Few-Shot + Formatting for consistent, specific outputs
- COSTAR + Chain-of-Thought for comprehensive, well-reasoned work
- Iterative refinement for when quality matters more than speed
- Multi-perspective for balanced analysis of complex decisions
- Match techniques to needs—don't over-engineer simple prompts
Conclusion
The techniques you've learned are tools. Individually, each solves specific problems. Combined thoughtfully, they handle any complex AI task you can imagine.
The key is diagnosis: What does this task actually need? Expert perspective? Careful reasoning? Specific format? Multiple viewpoints? Once you identify the needs, you know which techniques to combine.
Start experimenting with combinations on your next complex task. Notice which pairings work well for your common use cases. Build your own patterns that you can reuse.
This is where prompt engineering becomes powerful—not in mastering individual techniques, but in combining them fluidly to meet whatever challenge you face.
Master the techniques. Combine them with purpose. Handle anything.
Sponsored Content
Interested in advertising? Reach automation professionals through our platform.