Suppose an AI assistant drafts a product description that sounds persuasive but may not fit a brand’s voice. Reflective prompting asks the model to review its own draft, identify possible weaknesses, and suggest or make revisions. It turns a single response into a feedback loop. section 7.6
A review-and-revise prompt
“Review your answer for potential mistakes or missing details, and revise accordingly.” A more targeted request could ask the model to check a product description for persuasiveness and brand consistency. section 7.4 section 7.6
Reflexion is a formalized form of self-reflection. Reviewing and revising can be useful for code generation and open-ended writing, where refinement is expected. But a model may invent a flaw in a correct answer and introduce an error while trying to fix it. When a full rewrite is unnecessary, verification prompting offers a lighter alternative: ask for a final accuracy check and change the answer only if an obvious problem appears. section 7.4
Self-reflection does not replace evaluating the result. Prompt development still involves drafting, testing, assessing model behavior, and revising instructions; the model’s critique is another input to that process, not proof that the final answer is correct. section 7.4 section 7.6
