How would you answer an interview scenario involving Prompt design for LLM applications?

Given an extraction prompt that returns inconsistent JSON, explain how you would redesign the prompt and validate the response.