Self-reflection is a follow-up prompt that asks a model to critique its own output immediately after generating it. The critique can expose gaps and assumptions; paired with refinement, it forms an evaluation-then-improvement workflow. Use it when quality matters enough to justify another step, such as for complex analysis, strategic recommendations, or high-stakes outputs. Skip it for simple factual queries or urgent work. section 4.6
A practical workflow is to:
- Generate the initial answer, then ask the model to identify what it did well and what it may have overlooked. section 4.6.1
- Use the critique to refine the original output while preserving its strengths. section 4.8.1
- Choose self-reflection for internal critique; use inversion when you want gaps exposed through external challenge. section 4.6.1
Improving API documentation
Given minimal POST /orders documentation, ask: “What important elements might be missing that would help a developer use this endpoint correctly?” The critique identifies omissions such as the request schema, authentication, status codes, and error responses. A refinement prompt can then add the missing details, including a request structure and response codes. section 4.8.1
