A data scientist is designing a prompt to extract structured information (e.g., JSON) from text using an instruct model on OCI Generative AI. The model sometimes outputs additional text beyond the JSON, breaking parsing. Which prompt engineering technique is most effective to enforce structured output?