Advanced Prompting Techniques | LLM Fundamentals

You have learned the basics. You can write clear prompts, give examples, and iterate on responses. That puts you ahead of 80% of AI users.

But what about the hard problems? The ones where basic prompting falls short?

Complex reasoning that requires multiple steps
Tasks that need verification and fact-checking
Workflows that span multiple AI interactions
Problems that benefit from multiple perspectives

This lesson teaches you the advanced techniques that researchers and power users employ. These methods can turn an AI from a helpful assistant into a powerful reasoning partner.

What You Will Learn

Chain-of-Thought prompting for complex reasoning
Self-consistency for more reliable answers
Prompt chaining for multi-step workflows
Tree of Thoughts for exploring multiple solutions
Retrieval-Augmented Generation (RAG) concepts
Meta-prompting: getting AI to write better prompts
When to use each technique (and when not to)

Why Advanced Techniques Matter

Basic prompting works well for straightforward tasks:

"Summarize this article"
"Write an email to..."
"Explain X in simple terms"

But it struggles with:

Multi-step math and logic problems
Tasks requiring external knowledge
Complex analysis with many variables
Problems where the AI needs to consider alternatives

Real-world example:

A Google Research study tested different prompting methods on math word problems. Results:

Method	Accuracy
Standard prompting	17.7%
Chain-of-thought prompting	78.7%

Same AI. Same problems. 4x better results with better prompting.

These techniques are not just academic—they have practical impact on what you can accomplish with AI.

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

Chain-of-thought prompting asks the AI to show its reasoning step by step. This dramatically improves performance on complex problems.

Why It Works

When you ask someone to "think out loud," they:

Break problems into smaller pieces
Catch their own mistakes
Consider intermediate steps

AI works the same way. Forcing explicit reasoning prevents the AI from jumping to conclusions.

How to Use It

Method 1: Simple instruction

Add one of these phrases to your prompt:

"Think step by step"
"Let's work through this carefully"
"Show your reasoning before answering"
"Walk me through your thought process"

Example:

Without CoT:

plaintext
Roger has 5 tennis balls. He buys 2 more cans of tennis balls. 
Each can has 3 tennis balls. How many tennis balls does he have now?

AI might answer "11" or get confused

With CoT:

plaintext
Roger has 5 tennis balls. He buys 2 more cans of tennis balls. 
Each can has 3 tennis balls. How many tennis balls does he have now?

Think step by step.

AI response:

Step 1: Roger starts with 5 tennis balls. Step 2: He buys 2 cans. Step 3: Each can has 3 balls, so 2 × 3 = 6 new balls. Step 4: Total = 5 + 6 = 11 tennis balls.

Answer: 11 tennis balls

Method 2: Provide a Reasoning Example

Show the AI how to reason through a similar problem:

plaintext
Q: A store has 15 apples. They sell 8 and receive a shipment 
of 20 more. How many apples do they have?

A: Let me think step by step.
1. Starting apples: 15
2. After selling 8: 15 - 8 = 7
3. After receiving 20: 7 + 20 = 27
Answer: 27 apples

Q: Roger has 5 tennis balls. He buys 2 more cans of tennis balls. 
Each can has 3 tennis balls. How many tennis balls does he have now?

A:

The AI will follow your reasoning pattern.

When to Use Chain-of-Thought

Great For	Not Necessary For
Math word problems	Simple factual questions
Logic puzzles	Creative writing
Code debugging	Summarization
Complex analysis	Translation
Multi-step planning	Formatting tasks

Chain-of-Thought Variations

Zero-shot CoT: Just add "Let's think step by step"—no examples needed.

Few-shot CoT: Provide examples of step-by-step reasoning before your question.

Instructed CoT: Explicitly describe how you want the reasoning structured:

plaintext
Solve this problem. Structure your answer as:
1. Identify what we know
2. Identify what we need to find
3. Plan the approach
4. Execute step by step
5. Verify the answer
6. State the final answer

Technique 2: Self-Consistency

Self-consistency generates multiple answers and picks the most common one. It is like asking several experts and going with the consensus.

Why It Works

AI outputs have randomness. The same prompt might yield different answers. Self-consistency uses this to its advantage:

Generate 5-10 responses
Each takes a slightly different reasoning path
The most frequent answer is usually correct

How to Use It

Manual method:

Run the same prompt 5 times
Note each answer
Pick the answer that appears most often

Prompt-based method:

plaintext
Solve this problem 5 different ways, showing your reasoning each time. 
Then tell me which answer appeared most frequently.

Problem: [your problem here]

Example

plaintext
I am going to ask you to solve a logic puzzle 3 times. 
Each time, approach it fresh and show your reasoning.
After all 3 attempts, tell me the consensus answer.

Puzzle: If all Bloops are Razzies, and all Razzies are Lazzies, 
are all Bloops definitely Lazzies?

Attempt 1:
Attempt 2:
Attempt 3:

Consensus:

AI generates three reasoning paths, likely all concluding "Yes, all Bloops are Lazzies" through slightly different logic.

When to Use Self-Consistency

Use When	Skip When
High-stakes decisions	Quick, simple questions
Math and logic problems	Creative tasks (variety is good)
Factual accuracy matters	Time is limited
You need confidence in the answer	One answer is clearly sufficient

Practical tip: For important decisions, generate 3-5 responses and look for agreement. If answers differ significantly, the AI is uncertain—seek additional sources.

Technique 3: Prompt Chaining

Prompt chaining breaks complex tasks into a sequence of simpler prompts, where each step feeds into the next.

Why It Works

Complex tasks often fail because they overwhelm the AI. Chaining:

Keeps each step focused
Allows verification at each stage
Makes debugging easier
Produces more reliable results

The Pattern

plaintext
Task → Step 1 → Output 1 → Step 2 → Output 2 → Step 3 → Final Output

Example: Writing a Research Report

Instead of: "Write a research report on renewable energy trends"

Chain it:

Step 1: Outline

plaintext
Create a detailed outline for a research report on renewable 
energy trends in 2024. Include 5 main sections with 3 subtopics each.

Step 2: Research points

plaintext
Based on this outline:
[paste outline]

For each section, list 3-5 key facts or statistics that should 
be included. Mark any claims that would need citations.

Step 3: Draft sections

plaintext
Using this outline and research points:
[paste both]

Write section 2: Solar Energy Developments. Include the key 
facts listed. Keep it around 400 words.

Step 4: Review and refine

plaintext
Review this draft section for:
- Accuracy (flag anything that seems wrong)
- Flow and readability
- Missing information
Suggest specific improvements.

Practical Chaining Patterns

The Expand-Then-Refine Chain:

Generate initial ideas broadly
Select the best ideas
Develop the selected ideas in detail
Polish and finalize

The Analyze-Then-Act Chain:

Analyze the situation/data
Identify options
Evaluate each option
Recommend and implement

The Create-Then-Critique Chain:

Create initial version
Critique it (use a different persona)
Revise based on critique
Final review

Example: Business Decision Chain

Step 1: Analyze

plaintext
I am considering opening a coffee shop in downtown Austin. 
Analyze the key factors I should consider. List them as 
categories with specific questions under each.

Step 2: Research (for each category)

plaintext
Based on these factors you identified:
[paste factors]

For "Competition Analysis," what specific information should 
I gather? How would I find this information?

Step 3: Synthesize

plaintext
I have gathered this information:
[paste your research]

Synthesize this into a SWOT analysis for the coffee shop decision.

Step 4: Decide

plaintext
Based on this SWOT analysis:
[paste SWOT]

Give me a recommendation with:
- Your verdict (go/no-go/conditional)
- Top 3 reasons for this recommendation
- Key risks and how to mitigate them
- First 3 steps if I decide to proceed

When to Use Chaining

Use Chaining For	Single Prompt Is Fine For
Research reports	Quick summaries
Complex analysis	Simple explanations
Multi-part creative work	Short-form content
Decision frameworks	Straightforward decisions
Anything over 1000 words	Brief responses

Technique 4: Tree of Thoughts (ToT)

Tree of Thoughts explores multiple reasoning paths simultaneously, then evaluates which path is most promising.

Why It Works

Chain-of-thought follows one path. But some problems have multiple valid approaches. ToT:

Generates several different approaches
Evaluates the promise of each
Pursues the most promising
Can backtrack if needed

The Pattern

plaintext
Problem
├── Approach A
│   ├── Promising? Evaluate...
│   └── Continue or abandon
├── Approach B
│   ├── Promising? Evaluate...
│   └── Continue or abandon
└── Approach C
    ├── Promising? Evaluate...
    └── Continue or abandon

How to Use It

plaintext
Problem: [Describe your problem]

Generate 3 different approaches to solve this problem. 
For each approach:
1. Describe the approach in one sentence
2. List the first 2-3 steps
3. Evaluate: What are the strengths and weaknesses?
4. Rate potential success (low/medium/high)

Then select the most promising approach and develop it fully.

Example: Creative Problem Solving

plaintext
Challenge: Increase customer retention for a subscription 
meal kit service where 40% of customers cancel after 3 months.

Generate 3 different strategic approaches. For each:
1. Name the approach (2-3 words)
2. Core idea (1-2 sentences)
3. First 3 actions you would take
4. Main risk or challenge
5. Likelihood of significant impact (low/medium/high)

After presenting all three, recommend which to pursue first and why.

When to Use Tree of Thoughts

Great For	Overkill For
Strategic decisions	Routine tasks
Creative challenges	Questions with clear answers
Problems with multiple valid solutions	Time-sensitive requests
When you need to explore options	Simple preferences

Technique 5: Retrieval-Augmented Generation (RAG)

RAG combines AI with external knowledge sources. Instead of relying only on what the AI was trained on, you provide relevant information in the prompt.

Why It Works

AI models have a knowledge cutoff date and can hallucinate facts. RAG:

Provides accurate, up-to-date information
Grounds responses in real sources
Reduces hallucination dramatically
Enables AI to answer questions about your specific data

How to Use It (Simple Version)

Even without fancy tools, you can do manual RAG:

Step 1: Find relevant information

Search the web
Check your documents
Gather relevant data

Step 2: Include it in your prompt

plaintext
Based on the following information, answer my question.

Information:
---
[paste relevant text, data, or documents]
---

Question: [your question]

Use only the information provided above. If the answer is not 
in the provided information, say so.

Example: Answering Questions About Your Company

plaintext
You are a customer support assistant for TechFlow Inc. 
Answer customer questions using ONLY the information provided below. 
If you cannot answer from this information, say "I'll need to 
check with our team on that."

COMPANY INFORMATION:
---
Pricing: Starter plan $29/month, Pro plan $79/month, Enterprise custom
Free trial: 14 days, no credit card required
Refund policy: Full refund within 30 days
Support hours: 9am-6pm EST, Monday-Friday
Data storage: All data encrypted, stored in US data centers
---

Customer question: Do you offer a free trial, and will I be 
charged automatically?

RAG for Personal Knowledge

plaintext
I am going to share my notes from a meeting. Then I will ask 
questions about it.

MEETING NOTES:
---
[paste your notes]
---

Questions:
1. What were the main action items decided?
2. Who is responsible for the Q1 budget review?
3. When is the next follow-up meeting?

When to Use RAG

Essential For	Not Necessary For
Current events (after training cutoff)	General knowledge questions
Your specific documents/data	Creative writing
Factual accuracy requirements	Opinions and advice
Company-specific information	Brainstorming
Technical documentation	Simple explanations

Technique 6: Meta-Prompting

Meta-prompting uses AI to improve your prompts. It is like having a prompt engineering consultant built in.

The Core Idea

Ask the AI to:

Critique your prompt
Suggest improvements
Rewrite it better
Generate alternative versions

Prompt Improvement Request

plaintext
I want to use AI to help me with [task]. 

Here is my current prompt:
---
[your prompt]
---

Please:
1. Identify any weaknesses or ambiguities in this prompt
2. Suggest 3 specific improvements
3. Rewrite the prompt incorporating these improvements

Example

plaintext
I want to use AI to help me write better LinkedIn posts.

Here is my current prompt:
---
Write a LinkedIn post about AI in healthcare.
---

Please improve this prompt to get better results.

AI response:

Weaknesses identified:

No target audience specified

No desired length or format

Topic is too broad

No indication of tone or style

No call-to-action or engagement goal

Improved prompt: "Write a LinkedIn post (150-200 words) about how AI is reducing diagnostic errors in radiology. Target audience: healthcare professionals and health-tech investors. Tone: informative but accessible, with one specific statistic. End with a question to encourage comments. Write in first person as a health-tech professional sharing an insight."

Prompt Generation from Scratch

plaintext
I need to accomplish this task: [describe your goal in plain English]

Generate 3 different prompts I could use to accomplish this. 
For each prompt:
1. The full prompt text
2. What makes this approach effective
3. Any potential weaknesses

Then recommend which prompt to try first.

Expert Persona for Prompt Review

plaintext
You are an expert prompt engineer who has written thousands of 
effective prompts. 

Review this prompt and grade it on:
- Clarity (1-10)
- Specificity (1-10)
- Likelihood of good results (1-10)

Then rewrite it to score 9+ on all criteria.

Prompt to review:
[your prompt]

Technique 7: Structured Output Formats

Controlling the output format precisely ensures you get usable responses. This is especially important for automation.

JSON Output

plaintext
Analyze this customer review and return the analysis as JSON.

Review: "The product arrived quickly but was smaller than expected. 
Quality seems good though. Would probably buy again."

Return JSON with this structure:
{
  "sentiment": "positive/negative/mixed",
  "key_points": ["point1", "point2"],
  "purchase_intent": "likely/unlikely/uncertain",
  "issues_mentioned": ["issue1", "issue2"] or null
}

Markdown Tables

plaintext
Compare these 3 project management tools: Asana, Monday.com, Trello.

Return as a markdown table with columns:
| Feature | Asana | Monday.com | Trello |
Include: Pricing, Best for, Learning curve, Integrations, Mobile app

Numbered Lists with Specific Format

plaintext
Give me 5 blog post ideas about sustainable living.

Format each as:
[Number]. [Catchy Title]
   Target audience: [who]
   Key angle: [unique perspective]
   Estimated length: [word count]

XML-Style Tags

plaintext
Analyze this text for writing quality.

<text>
[paste text]
</text>

Respond using these tags:
<strengths>What the writing does well</strengths>
<weaknesses>Areas for improvement</weaknesses>
<suggestions>Specific changes to make</suggestions>
<revised_sample>Rewrite of the first paragraph</revised_sample>

Combining Techniques

The real power comes from combining these methods.

Example: Comprehensive Analysis

plaintext
[ROLE]
You are a senior business analyst with 20 years of experience 
in market research.

[TASK]
Analyze whether a boutique hotel should expand into the 
co-working space trend.

[PROCESS]
Use this structured approach:

Step 1: Generate 3 different analytical frameworks we could 
use (Tree of Thoughts)

Step 2: Select the best framework and explain why

Step 3: Apply the framework step-by-step (Chain of Thought)

Step 4: Consider counter-arguments (Self-Consistency check)

Step 5: Synthesize into a recommendation

[FORMAT]
- Use headers for each step
- Include a summary table of pros/cons
- End with a clear recommendation and confidence level

Example: Content Creation Pipeline

Prompt 1 (Ideation with Tree of Thoughts):

plaintext
Generate 5 different angles for a blog post about remote work productivity.
Evaluate each for uniqueness and reader value.
Select the top 2.

Prompt 2 (Outline with CoT):

plaintext
For angle #1: [paste selected angle]
Think through the logical structure step by step.
Create a detailed outline with:
- Hook/introduction approach
- 5 main sections
- Key points under each section
- Conclusion and CTA

Prompt 3 (Draft with RAG):

plaintext
Using this outline:
[paste outline]

And these research notes:
[paste relevant research]

Write section 3 in full. Include specific examples and data 
from the research. Target 400 words.

Prompt 4 (Review with Meta-Prompting):

plaintext
Review this draft section:
[paste section]

As a senior editor, identify:
- Any weak arguments
- Missing examples
- Awkward phrasing
- Opportunities to be more specific

Provide a revised version addressing these issues.

Technique Selection Guide

Use this quick reference to pick the right technique:

Situation	Recommended Technique
Math or logic problem	Chain-of-Thought
Need high confidence in answer	Self-Consistency
Complex multi-part task	Prompt Chaining
Multiple possible solutions	Tree of Thoughts
Need current/specific information	RAG
Want to improve your prompting	Meta-Prompting
Need automation-friendly output	Structured Output

Decision Flowchart

plaintext
Is it a simple, straightforward request?
├── Yes → Basic prompting is fine
└── No → Does it require reasoning?
    ├── Yes → Chain-of-Thought
    │   └── Need high confidence? → Add Self-Consistency
    └── No → Is it a multi-step task?
        ├── Yes → Prompt Chaining
        └── No → Multiple approaches possible?
            ├── Yes → Tree of Thoughts
            └── No → Need external information?
                ├── Yes → RAG
                └── No → Structured Output + Basic Prompting

Common Pitfalls

Pitfall 1: Over-Engineering Simple Tasks

Bad: Using Tree of Thoughts to decide what to eat for lunch Good: Reserve advanced techniques for genuinely complex problems

Pitfall 2: Not Verifying Intermediate Steps

In prompt chaining, always check outputs before proceeding. Errors compound.

Pitfall 3: Ignoring Token Limits

Long chains and multiple examples consume context. If responses get worse, you may be hitting limits.

Pitfall 4: Expecting Perfection

Even advanced techniques are not 100% reliable. Always verify critical outputs.

Pitfall 5: Forgetting the Basics

Advanced techniques build on fundamentals. A poorly written chain-of-thought prompt is still a poor prompt.

Key Takeaways

Chain-of-Thought transforms reasoning. For any problem requiring logic or steps, ask the AI to show its work.
Self-Consistency improves reliability. When accuracy matters, generate multiple answers and look for consensus.
Prompt Chaining handles complexity. Break big tasks into focused steps, verify each one, then combine.
Tree of Thoughts explores options. When there are multiple valid approaches, evaluate before committing.
RAG grounds responses in facts. For accuracy, provide the information yourself rather than relying on the AI's knowledge.
Meta-Prompting makes you better. Use AI to improve your prompts—it is like having a tutor built in.
Combine techniques for real power. The best results often come from using multiple methods together.
Match technique to task. Not every problem needs advanced methods. Use the simplest approach that works.

Practice Exercises

Exercise 1: Chain-of-Thought

Take a complex problem from your work. Write a prompt that includes "Think step by step." Compare the result to a basic prompt.

Exercise 2: Self-Consistency

Ask the AI to solve a tricky logic puzzle 3 times. Do the answers agree? If not, what does that tell you?

Exercise 3: Prompt Chaining

Take a task you usually do in one prompt. Break it into 3-4 chained prompts. Is the final output better?

Exercise 4: Meta-Prompting

Take your most-used prompt and ask the AI to improve it. Try the improved version. Which works better?

What is Next

You have now mastered the core prompting techniques—from basics to advanced methods. In the next lesson, we cover Fine-Tuning Strategies, where you learn to customize AI models for your specific use cases.

Fine-tuning is the next level: instead of better prompts, you create a better model.

References & Further Reading

Research Papers

Wei, J., et al. (2022). Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Wang, X., et al. (2022). Self-Consistency Improves Chain of Thought Reasoning
Yao, S., et al. (2023). Tree of Thoughts: Deliberate Problem Solving with Large Language Models
Lewis, P., et al. (2020). Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

Practical Guides

Prompt Engineering Guide - Comprehensive resource with examples
OpenAI Cookbook - Practical patterns and code examples
Anthropic's Constitutional AI - Understanding AI reasoning

Tools to Practice

ChatGPT - Best for experimenting with different techniques
Claude - Excellent for long-form reasoning
Perplexity - Built-in RAG for research questions

You have learned the basics. You can write clear prompts, give examples, and iterate on responses. That puts you ahead of 80% of AI users.

But what about the hard problems? The ones where basic prompting falls short?

Complex reasoning that requires multiple steps
Tasks that need verification and fact-checking
Workflows that span multiple AI interactions
Problems that benefit from multiple perspectives

This lesson teaches you the advanced techniques that researchers and power users employ. These methods can turn an AI from a helpful assistant into a powerful reasoning partner.

What You Will Learn

Chain-of-Thought prompting for complex reasoning
Self-consistency for more reliable answers
Prompt chaining for multi-step workflows
Tree of Thoughts for exploring multiple solutions
Retrieval-Augmented Generation (RAG) concepts
Meta-prompting: getting AI to write better prompts
When to use each technique (and when not to)

Why Advanced Techniques Matter

Basic prompting works well for straightforward tasks:

"Summarize this article"
"Write an email to..."
"Explain X in simple terms"

But it struggles with:

Multi-step math and logic problems
Tasks requiring external knowledge
Complex analysis with many variables
Problems where the AI needs to consider alternatives

Real-world example:

A Google Research study tested different prompting methods on math word problems. Results:

Method	Accuracy
Standard prompting	17.7%
Chain-of-thought prompting	78.7%

Same AI. Same problems. 4x better results with better prompting.

These techniques are not just academic—they have practical impact on what you can accomplish with AI.

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

Chain-of-thought prompting asks the AI to show its reasoning step by step. This dramatically improves performance on complex problems.

Why It Works

When you ask someone to "think out loud," they:

Break problems into smaller pieces
Catch their own mistakes
Consider intermediate steps

AI works the same way. Forcing explicit reasoning prevents the AI from jumping to conclusions.

How to Use It

Method 1: Simple instruction

Add one of these phrases to your prompt:

"Think step by step"
"Let's work through this carefully"
"Show your reasoning before answering"
"Walk me through your thought process"

Example:

Without CoT:

plaintext
Roger has 5 tennis balls. He buys 2 more cans of tennis balls. 
Each can has 3 tennis balls. How many tennis balls does he have now?

AI might answer "11" or get confused

With CoT:

plaintext
Roger has 5 tennis balls. He buys 2 more cans of tennis balls. 
Each can has 3 tennis balls. How many tennis balls does he have now?

Think step by step.

AI response:

Step 1: Roger starts with 5 tennis balls. Step 2: He buys 2 cans. Step 3: Each can has 3 balls, so 2 × 3 = 6 new balls. Step 4: Total = 5 + 6 = 11 tennis balls.

Answer: 11 tennis balls

Method 2: Provide a Reasoning Example

Show the AI how to reason through a similar problem:

plaintext
Q: A store has 15 apples. They sell 8 and receive a shipment 
of 20 more. How many apples do they have?

A: Let me think step by step.
1. Starting apples: 15
2. After selling 8: 15 - 8 = 7
3. After receiving 20: 7 + 20 = 27
Answer: 27 apples

Q: Roger has 5 tennis balls. He buys 2 more cans of tennis balls. 
Each can has 3 tennis balls. How many tennis balls does he have now?

A:

The AI will follow your reasoning pattern.

When to Use Chain-of-Thought

Great For	Not Necessary For
Math word problems	Simple factual questions
Logic puzzles	Creative writing
Code debugging	Summarization
Complex analysis	Translation
Multi-step planning	Formatting tasks

Chain-of-Thought Variations

Zero-shot CoT: Just add "Let's think step by step"—no examples needed.

Few-shot CoT: Provide examples of step-by-step reasoning before your question.

Instructed CoT: Explicitly describe how you want the reasoning structured:

plaintext
Solve this problem. Structure your answer as:
1. Identify what we know
2. Identify what we need to find
3. Plan the approach
4. Execute step by step
5. Verify the answer
6. State the final answer

Technique 2: Self-Consistency

Self-consistency generates multiple answers and picks the most common one. It is like asking several experts and going with the consensus.

Why It Works

AI outputs have randomness. The same prompt might yield different answers. Self-consistency uses this to its advantage:

Generate 5-10 responses
Each takes a slightly different reasoning path
The most frequent answer is usually correct

How to Use It

Manual method:

Run the same prompt 5 times
Note each answer
Pick the answer that appears most often

Prompt-based method:

plaintext
Solve this problem 5 different ways, showing your reasoning each time. 
Then tell me which answer appeared most frequently.

Problem: [your problem here]

Example

plaintext
I am going to ask you to solve a logic puzzle 3 times. 
Each time, approach it fresh and show your reasoning.
After all 3 attempts, tell me the consensus answer.

Puzzle: If all Bloops are Razzies, and all Razzies are Lazzies, 
are all Bloops definitely Lazzies?

Attempt 1:
Attempt 2:
Attempt 3:

Consensus:

AI generates three reasoning paths, likely all concluding "Yes, all Bloops are Lazzies" through slightly different logic.

When to Use Self-Consistency

Use When	Skip When
High-stakes decisions	Quick, simple questions
Math and logic problems	Creative tasks (variety is good)
Factual accuracy matters	Time is limited
You need confidence in the answer	One answer is clearly sufficient

Practical tip: For important decisions, generate 3-5 responses and look for agreement. If answers differ significantly, the AI is uncertain—seek additional sources.

Technique 3: Prompt Chaining

Prompt chaining breaks complex tasks into a sequence of simpler prompts, where each step feeds into the next.

Why It Works

Complex tasks often fail because they overwhelm the AI. Chaining:

Keeps each step focused
Allows verification at each stage
Makes debugging easier
Produces more reliable results

The Pattern

plaintext
Task → Step 1 → Output 1 → Step 2 → Output 2 → Step 3 → Final Output

Example: Writing a Research Report

Instead of: "Write a research report on renewable energy trends"

Chain it:

Step 1: Outline

plaintext
Create a detailed outline for a research report on renewable 
energy trends in 2024. Include 5 main sections with 3 subtopics each.

Step 2: Research points

plaintext
Based on this outline:
[paste outline]

For each section, list 3-5 key facts or statistics that should 
be included. Mark any claims that would need citations.

Step 3: Draft sections

plaintext
Using this outline and research points:
[paste both]

Write section 2: Solar Energy Developments. Include the key 
facts listed. Keep it around 400 words.

Step 4: Review and refine

plaintext
Review this draft section for:
- Accuracy (flag anything that seems wrong)
- Flow and readability
- Missing information
Suggest specific improvements.

Practical Chaining Patterns

The Expand-Then-Refine Chain:

Generate initial ideas broadly
Select the best ideas
Develop the selected ideas in detail
Polish and finalize

The Analyze-Then-Act Chain:

Analyze the situation/data
Identify options
Evaluate each option
Recommend and implement

The Create-Then-Critique Chain:

Create initial version
Critique it (use a different persona)
Revise based on critique
Final review

Example: Business Decision Chain

Step 1: Analyze

plaintext
I am considering opening a coffee shop in downtown Austin. 
Analyze the key factors I should consider. List them as 
categories with specific questions under each.

Step 2: Research (for each category)

plaintext
Based on these factors you identified:
[paste factors]

For "Competition Analysis," what specific information should 
I gather? How would I find this information?

Step 3: Synthesize

plaintext
I have gathered this information:
[paste your research]

Synthesize this into a SWOT analysis for the coffee shop decision.

Step 4: Decide

plaintext
Based on this SWOT analysis:
[paste SWOT]

Give me a recommendation with:
- Your verdict (go/no-go/conditional)
- Top 3 reasons for this recommendation
- Key risks and how to mitigate them
- First 3 steps if I decide to proceed

When to Use Chaining

Use Chaining For	Single Prompt Is Fine For
Research reports	Quick summaries
Complex analysis	Simple explanations
Multi-part creative work	Short-form content
Decision frameworks	Straightforward decisions
Anything over 1000 words	Brief responses

Technique 4: Tree of Thoughts (ToT)

Tree of Thoughts explores multiple reasoning paths simultaneously, then evaluates which path is most promising.

Why It Works

Chain-of-thought follows one path. But some problems have multiple valid approaches. ToT:

Generates several different approaches
Evaluates the promise of each
Pursues the most promising
Can backtrack if needed

The Pattern

plaintext
Problem
├── Approach A
│   ├── Promising? Evaluate...
│   └── Continue or abandon
├── Approach B
│   ├── Promising? Evaluate...
│   └── Continue or abandon
└── Approach C
    ├── Promising? Evaluate...
    └── Continue or abandon

How to Use It

plaintext
Problem: [Describe your problem]

Generate 3 different approaches to solve this problem. 
For each approach:
1. Describe the approach in one sentence
2. List the first 2-3 steps
3. Evaluate: What are the strengths and weaknesses?
4. Rate potential success (low/medium/high)

Then select the most promising approach and develop it fully.

Example: Creative Problem Solving

plaintext
Challenge: Increase customer retention for a subscription 
meal kit service where 40% of customers cancel after 3 months.

Generate 3 different strategic approaches. For each:
1. Name the approach (2-3 words)
2. Core idea (1-2 sentences)
3. First 3 actions you would take
4. Main risk or challenge
5. Likelihood of significant impact (low/medium/high)

After presenting all three, recommend which to pursue first and why.

When to Use Tree of Thoughts

Great For	Overkill For
Strategic decisions	Routine tasks
Creative challenges	Questions with clear answers
Problems with multiple valid solutions	Time-sensitive requests
When you need to explore options	Simple preferences

Technique 5: Retrieval-Augmented Generation (RAG)

RAG combines AI with external knowledge sources. Instead of relying only on what the AI was trained on, you provide relevant information in the prompt.

Why It Works

AI models have a knowledge cutoff date and can hallucinate facts. RAG:

Provides accurate, up-to-date information
Grounds responses in real sources
Reduces hallucination dramatically
Enables AI to answer questions about your specific data

How to Use It (Simple Version)

Even without fancy tools, you can do manual RAG:

Step 1: Find relevant information

Search the web
Check your documents
Gather relevant data

Step 2: Include it in your prompt

plaintext
Based on the following information, answer my question.

Information:
---
[paste relevant text, data, or documents]
---

Question: [your question]

Use only the information provided above. If the answer is not 
in the provided information, say so.

Example: Answering Questions About Your Company

plaintext
You are a customer support assistant for TechFlow Inc. 
Answer customer questions using ONLY the information provided below. 
If you cannot answer from this information, say "I'll need to 
check with our team on that."

COMPANY INFORMATION:
---
Pricing: Starter plan $29/month, Pro plan $79/month, Enterprise custom
Free trial: 14 days, no credit card required
Refund policy: Full refund within 30 days
Support hours: 9am-6pm EST, Monday-Friday
Data storage: All data encrypted, stored in US data centers
---

Customer question: Do you offer a free trial, and will I be 
charged automatically?

RAG for Personal Knowledge

plaintext
I am going to share my notes from a meeting. Then I will ask 
questions about it.

MEETING NOTES:
---
[paste your notes]
---

Questions:
1. What were the main action items decided?
2. Who is responsible for the Q1 budget review?
3. When is the next follow-up meeting?

When to Use RAG

Essential For	Not Necessary For
Current events (after training cutoff)	General knowledge questions
Your specific documents/data	Creative writing
Factual accuracy requirements	Opinions and advice
Company-specific information	Brainstorming
Technical documentation	Simple explanations

Technique 6: Meta-Prompting

Meta-prompting uses AI to improve your prompts. It is like having a prompt engineering consultant built in.

The Core Idea

Ask the AI to:

Critique your prompt
Suggest improvements
Rewrite it better
Generate alternative versions

Prompt Improvement Request

plaintext
I want to use AI to help me with [task]. 

Here is my current prompt:
---
[your prompt]
---

Please:
1. Identify any weaknesses or ambiguities in this prompt
2. Suggest 3 specific improvements
3. Rewrite the prompt incorporating these improvements

Example

plaintext
I want to use AI to help me write better LinkedIn posts.

Here is my current prompt:
---
Write a LinkedIn post about AI in healthcare.
---

Please improve this prompt to get better results.

AI response:

Weaknesses identified:

No target audience specified

No desired length or format

Topic is too broad

No indication of tone or style

No call-to-action or engagement goal

Improved prompt: "Write a LinkedIn post (150-200 words) about how AI is reducing diagnostic errors in radiology. Target audience: healthcare professionals and health-tech investors. Tone: informative but accessible, with one specific statistic. End with a question to encourage comments. Write in first person as a health-tech professional sharing an insight."

Prompt Generation from Scratch

plaintext
I need to accomplish this task: [describe your goal in plain English]

Generate 3 different prompts I could use to accomplish this. 
For each prompt:
1. The full prompt text
2. What makes this approach effective
3. Any potential weaknesses

Then recommend which prompt to try first.

Expert Persona for Prompt Review

plaintext
You are an expert prompt engineer who has written thousands of 
effective prompts. 

Review this prompt and grade it on:
- Clarity (1-10)
- Specificity (1-10)
- Likelihood of good results (1-10)

Then rewrite it to score 9+ on all criteria.

Prompt to review:
[your prompt]

Technique 7: Structured Output Formats

Controlling the output format precisely ensures you get usable responses. This is especially important for automation.

JSON Output

plaintext
Analyze this customer review and return the analysis as JSON.

Review: "The product arrived quickly but was smaller than expected. 
Quality seems good though. Would probably buy again."

Return JSON with this structure:
{
  "sentiment": "positive/negative/mixed",
  "key_points": ["point1", "point2"],
  "purchase_intent": "likely/unlikely/uncertain",
  "issues_mentioned": ["issue1", "issue2"] or null
}

Markdown Tables

plaintext
Compare these 3 project management tools: Asana, Monday.com, Trello.

Return as a markdown table with columns:
| Feature | Asana | Monday.com | Trello |
Include: Pricing, Best for, Learning curve, Integrations, Mobile app

Numbered Lists with Specific Format

plaintext
Give me 5 blog post ideas about sustainable living.

Format each as:
[Number]. [Catchy Title]
   Target audience: [who]
   Key angle: [unique perspective]
   Estimated length: [word count]

XML-Style Tags

plaintext
Analyze this text for writing quality.

<text>
[paste text]
</text>

Respond using these tags:
<strengths>What the writing does well</strengths>
<weaknesses>Areas for improvement</weaknesses>
<suggestions>Specific changes to make</suggestions>
<revised_sample>Rewrite of the first paragraph</revised_sample>

Combining Techniques

The real power comes from combining these methods.

Example: Comprehensive Analysis

plaintext
[ROLE]
You are a senior business analyst with 20 years of experience 
in market research.

[TASK]
Analyze whether a boutique hotel should expand into the 
co-working space trend.

[PROCESS]
Use this structured approach:

Step 1: Generate 3 different analytical frameworks we could 
use (Tree of Thoughts)

Step 2: Select the best framework and explain why

Step 3: Apply the framework step-by-step (Chain of Thought)

Step 4: Consider counter-arguments (Self-Consistency check)

Step 5: Synthesize into a recommendation

[FORMAT]
- Use headers for each step
- Include a summary table of pros/cons
- End with a clear recommendation and confidence level

Example: Content Creation Pipeline

Prompt 1 (Ideation with Tree of Thoughts):

plaintext
Generate 5 different angles for a blog post about remote work productivity.
Evaluate each for uniqueness and reader value.
Select the top 2.

Prompt 2 (Outline with CoT):

plaintext
For angle #1: [paste selected angle]
Think through the logical structure step by step.
Create a detailed outline with:
- Hook/introduction approach
- 5 main sections
- Key points under each section
- Conclusion and CTA

Prompt 3 (Draft with RAG):

plaintext
Using this outline:
[paste outline]

And these research notes:
[paste relevant research]

Write section 3 in full. Include specific examples and data 
from the research. Target 400 words.

Prompt 4 (Review with Meta-Prompting):

plaintext
Review this draft section:
[paste section]

As a senior editor, identify:
- Any weak arguments
- Missing examples
- Awkward phrasing
- Opportunities to be more specific

Provide a revised version addressing these issues.

Technique Selection Guide

Use this quick reference to pick the right technique:

Situation	Recommended Technique
Math or logic problem	Chain-of-Thought
Need high confidence in answer	Self-Consistency
Complex multi-part task	Prompt Chaining
Multiple possible solutions	Tree of Thoughts
Need current/specific information	RAG
Want to improve your prompting	Meta-Prompting
Need automation-friendly output	Structured Output

Decision Flowchart

plaintext
Is it a simple, straightforward request?
├── Yes → Basic prompting is fine
└── No → Does it require reasoning?
    ├── Yes → Chain-of-Thought
    │   └── Need high confidence? → Add Self-Consistency
    └── No → Is it a multi-step task?
        ├── Yes → Prompt Chaining
        └── No → Multiple approaches possible?
            ├── Yes → Tree of Thoughts
            └── No → Need external information?
                ├── Yes → RAG
                └── No → Structured Output + Basic Prompting

Common Pitfalls

Pitfall 1: Over-Engineering Simple Tasks

Bad: Using Tree of Thoughts to decide what to eat for lunch Good: Reserve advanced techniques for genuinely complex problems

Pitfall 2: Not Verifying Intermediate Steps

In prompt chaining, always check outputs before proceeding. Errors compound.

Pitfall 3: Ignoring Token Limits

Long chains and multiple examples consume context. If responses get worse, you may be hitting limits.

Pitfall 4: Expecting Perfection

Even advanced techniques are not 100% reliable. Always verify critical outputs.

Pitfall 5: Forgetting the Basics

Advanced techniques build on fundamentals. A poorly written chain-of-thought prompt is still a poor prompt.

Key Takeaways

Chain-of-Thought transforms reasoning. For any problem requiring logic or steps, ask the AI to show its work.
Self-Consistency improves reliability. When accuracy matters, generate multiple answers and look for consensus.
Prompt Chaining handles complexity. Break big tasks into focused steps, verify each one, then combine.
Tree of Thoughts explores options. When there are multiple valid approaches, evaluate before committing.
RAG grounds responses in facts. For accuracy, provide the information yourself rather than relying on the AI's knowledge.
Meta-Prompting makes you better. Use AI to improve your prompts—it is like having a tutor built in.
Combine techniques for real power. The best results often come from using multiple methods together.
Match technique to task. Not every problem needs advanced methods. Use the simplest approach that works.

Practice Exercises

Exercise 1: Chain-of-Thought

Take a complex problem from your work. Write a prompt that includes "Think step by step." Compare the result to a basic prompt.

Exercise 2: Self-Consistency

Ask the AI to solve a tricky logic puzzle 3 times. Do the answers agree? If not, what does that tell you?

Exercise 3: Prompt Chaining

Take a task you usually do in one prompt. Break it into 3-4 chained prompts. Is the final output better?

Exercise 4: Meta-Prompting

Take your most-used prompt and ask the AI to improve it. Try the improved version. Which works better?

What is Next

Fine-tuning is the next level: instead of better prompts, you create a better model.

References & Further Reading

Research Papers

Wei, J., et al. (2022). Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Wang, X., et al. (2022). Self-Consistency Improves Chain of Thought Reasoning
Yao, S., et al. (2023). Tree of Thoughts: Deliberate Problem Solving with Large Language Models
Lewis, P., et al. (2020). Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

Practical Guides

Prompt Engineering Guide - Comprehensive resource with examples
OpenAI Cookbook - Practical patterns and code examples
Anthropic's Constitutional AI - Understanding AI reasoning

Tools to Practice

ChatGPT - Best for experimenting with different techniques
Claude - Excellent for long-form reasoning
Perplexity - Built-in RAG for research questions