Course contentsShow
← Machine Learning and Deep Learning
Lesson 1866 of 3,53840. Advanced Prompting TechniquesFree lesson

Anatomy of Effective Reasoning Examples

Designing demonstration examples that show clear, logical steps from problem to solution with explicit intermediate reasoning.

Anatomy of Effective Reasoning Examples

What You'll Learn

You'll discover how to craft demonstration examples that guide models through clear, logical reasoning paths—not just show the final answer.

The Core Idea

When using few-shot Chain-of-Thought prompting, the quality of your reasoning examples matters as much as their presence. An effective reasoning example isn't just "problem → answer"—it's a transparent window into the thought process itself.

Think of it like teaching math to a student. Showing "5 × 12 = 60" helps less than showing: "5 × 12 means 5 groups of 12. I can break 12 into 10 + 2, so 5 × 10 = 50 and 5 × 2 = 10, then 50 + 10 = 60." The intermediate steps make the pattern learnable.

Key Components of Strong Reasoning Examples

1. Explicit Intermediate Steps

Break the problem into clear sub-problems. Each step should follow naturally from the previous one.

2. Reasoning Transparency

State why you're taking each step. Don't just calculate—explain what you're calculating and why it matters.

3. Problem Decomposition

Show how you break complex problems into simpler parts. This teaches the model your problem-solving strategy.

4. Consistent Structure

Use the same reasoning pattern across examples so the model recognizes the template: identify knowns → determine what's needed → plan approach → execute → verify.

The Bad vs. Good Example

Bad: "The store has 23 apples. After selling some, 8 remain. They sold 15 apples."

Good: "The store has 23 apples initially. After selling some, 8 remain. To find how many were sold, I need to subtract: final amount from initial amount. So 23 - 8 = 15 apples sold."

Key Takeaway: Effective reasoning examples make the invisible visible—they externalize the logical steps, decision points, and intermediate conclusions that bridge problem to solution, teaching the model how to think, not just what to output.