Research-backed frameworks for effective AI communication. These patterns are proven techniques to understand and adapt for your specific needs.
Select a category to filter patterns by type, or browse all available frameworks.
Encourage step-by-step reasoning by asking AI to "think through" problems before answering. This approach significantly improves accuracy on complex tasks.
"Let's think through this step by step..." or "Walk me through your reasoning."
Provide a few examples of the input-output format you want before your actual request. This leverages the AI's pattern recognition capabilities.
"Example 1: [input] → [output]. Example 2: [input] → [output]. Now do: [your input]"
Assign AI a specific role or expertise to shape its responses and vocabulary. This activates domain-specific knowledge patterns.
"You are an experienced [role] who specializes in [domain]..."
Define the exact format you want (JSON, table, outline) before the request. Clear structure reduces ambiguity and improves usability.
"Return your response in this format: {title: ..., summary: ..., points: [...]}"
Ask AI to verify its own work or approach a problem multiple ways and compare. This catches errors and increases confidence in results.
"After your answer, verify it's correct." or "Solve this two different ways."
Break complex tasks into sequential prompts where each output feeds the next input. This manages complexity and maintains quality.
1. Research → 2. Outline → 3. Draft → 4. Edit (separate prompts)
Ask AI to argue against a position or find weaknesses in an idea. This surfaces blind spots and strengthens decision-making.
"Play devil's advocate and critique this plan." or "What could go wrong?"
State limitations and requirements before the main request to frame the response. Front-loading constraints prevents wasted generation.
"Constraints: 200 words max, no jargon, audience is beginners. Task: Explain..."
Ask AI to interview you first before giving advice. This prevents generic responses and yields personalized outputs.
"Before answering, ask me 5 questions to understand my situation better."
Use this guide to select the right pattern for your task type.
| Task Type | Recommended Pattern | Why It Works | Example Use |
|---|---|---|---|
| Problem Solving | Chain of Thought | Forces step-by-step reasoning | Debugging code, math problems |
| Consistent Formatting | Few-Shot Learning | Demonstrates exact output pattern | Data transformation, templates |
| Expert Knowledge | Role Prompting | Activates domain-specific responses | Technical writing, consulting |
| Data Integration | Structured Output | Ensures machine-readable format | API responses, reports |
| High-Stakes Decisions | Self-Consistency | Validates through multiple approaches | Financial analysis, diagnostics |
| Large Projects | Prompt Chaining | Breaks complexity into manageable steps | Research papers, long documents |
| Risk Assessment | Devil's Advocate | Surfaces hidden problems | Business plans, proposals |
| Targeted Content | Constraints First | Prevents off-target responses | Social media posts, emails |
| Personalized Advice | Flipped Interaction | Gathers context before responding | Career guidance, planning |
See how these patterns transform real-world AI interactions.
The Challenge: A marketing manager asked AI to "write some social media posts" and received generic, unusable content.
Pattern Applied: Constraints First + Few-Shot Learning
New Prompt: "Constraints: 280 characters max, conversational tone, include one emoji, target audience is small business owners. Examples: [example 1] [example 2]. Now write 5 posts about our new scheduling feature."
The Challenge: A developer pasted error-prone code and asked "why isn't this working?" The AI's response was superficial.
Pattern Applied: Chain of Thought + Self-Consistency
New Prompt: "Let's debug this step by step. First, trace the data flow. Then identify where the expected behavior diverges from actual. Finally, verify your diagnosis by explaining what the fix would change."
The Challenge: A professional asked "should I change careers?" and got generic advice that could apply to anyone.
Pattern Applied: Flipped Interaction
New Prompt: "I'm considering a career change. Before giving advice, interview me: ask about my current role, what I enjoy vs. what drains me, my financial situation, and my risk tolerance."
These patterns are principles, not scripts. Understand why they work, then adapt them to your needs.
Patterns are frameworks to internalize, not templates to paste. Learn the principle behind each pattern so you can adapt it naturally.
Patterns work well together. Use Role + Chain of Thought for expert analysis, or Few-Shot + Structured Output for consistent formatting.
Start with a pattern, see what works, then adjust. The best prompts come from experimentation and learning from each interaction.
Not every task needs every pattern. Simple questions don't need Chain of Thought. Match the pattern's complexity to your task's requirements.
Keep examples of prompts that worked well for you. Over time, you'll develop personal variations that fit your communication style.
These patterns are building blocks. Learn complete methodologies like CRISP, COSTAR, and ReAct to combine patterns into powerful frameworks.
Don't use Chain of Thought for "What's the capital of France?" Match complexity to need—simple questions deserve simple prompts.
Very long few-shot examples can crowd out your actual question. Keep examples concise and relevant to leave room for quality responses.
Asking AI to be "detailed but brief" or "formal but conversational" creates confusion. Be clear about your priorities.
If your first prompt doesn't work well, try a completely different pattern instead of just tweaking words. Sometimes a fresh approach works better than iteration.
Learn the methodologies that combine these patterns, or test your prompts with our Analyzer tool.