reflection prompting

**Reflection prompting** is the **prompting technique that asks the model to review and critique its own output before producing a revised answer** - it introduces a self-correction loop that often improves final quality. **What Is Reflection prompting?** - **Definition**: Two-stage or multi-stage prompt pattern of generate, critique, and refine. - **Review Focus**: Can target factual errors, logic gaps, formatting defects, or policy compliance issues. - **Execution Modes**: Single-model self-review or separate critic and generator roles. - **Task Fit**: Especially useful for coding, analytical writing, and high-precision structured outputs. **Why Reflection prompting Matters** - **Quality Improvement**: Self-audit frequently catches issues missed in first-pass generation. - **Reliability Gain**: Iterative refinement reduces obvious errors and inconsistencies. - **Process Transparency**: Reflection output provides rationale for revisions. - **Alignment Support**: Critique stage can enforce style, safety, and domain constraints. - **Cost Tradeoff**: Extra passes increase latency but can reduce downstream correction effort. **How It Is Used in Practice** - **Critique Template**: Define explicit review criteria and severity tags for detected issues. - **Revision Rules**: Require second pass to address each critique point directly. - **Stop Conditions**: Limit iteration count and use quality thresholds to control runtime. Reflection prompting is **a practical self-improvement loop for LLM outputs** - structured review and revision cycles improve correctness and robustness in production prompt workflows.