Reflection Prompting

Introduction

Reflection asks the model to evaluate its own first answer and improve it. It's a one-prompt critic that catches obvious errors before they ship.

Beginner analogy: Like re-reading your own essay before submitting. Self-edit catches half the bugs.

Understanding the topic

Core concepts to understand:

• Two-turn pattern: answer → critique → revise.
• Often run in a single prompt with sections.
• Catches hallucinations, format breaks, missed constraints.
• Adds latency & cost — use on critical paths.

Syntax reference

Visual workflow / architecture:

Step 1: Generate answer.
Step 2: Critique answer for:
        - factual errors
        - missed constraints
        - poor format
Step 3: Rewrite, fixing all critiques.
Output: final revised answer only.

Real-world use

Coding assistants use reflection to retry compilation errors before showing code to the user.

Best practices

• Limit to 1 reflection round — diminishing returns after that.
• Reflect against a checklist, not free-form.
• Hide reflection from end users.

Hands-on exercise

Interview preparation — practice these questions:

• Q1. What is reflection prompting?
• Q2. When does it backfire?
• Q3. How many rounds of reflection?
• Q4. Reflection vs human review?