Write a Prompt That Produces the Format You Actually Need

An answer may contain the right information in the wrong shape. If I need a bug triage checklist, three pages of prose slow me down. If I need data for an app, a pretty markdown table may be difficult to parse. When another system will consume the response, Structured Outputs: When a Schema Beats More Prompt Words explains where a schema beats formatting instructions.

Before prompting, ask what you will do with the answer: Read and decide: use a short comparison with criteria. Act step by step: use a checklist with owners or order. Run code: request code, tests, and assumptions separately. Feed an application: use schema-constrained structured output if the provider supports it. For example, this vague prompt often produces more than I can use: Review my playtest notes. A more useful version: Read the playtest notes below. Return a table with Issue, Evidence from notes, Severity (high/medium/low), and Next test. Do not invent a severity from frequency alone; explain any uncertainty in one sentence after the table. The table turns a discussion into actions. The “evidence” column prevents confident labels without support.

Asking for JSON does not guarantee valid JSON, and valid JSON does not guarantee correct information. If software depends on the result, use the provider’s structured-output feature where possible, then validate the fields and the values in your own code. Google’s structured-output docs and OpenAI’s structured-output docs explain provider-supported schema approaches.

Avoid specifying twenty presentation rules when four fields would do. A format is useful if it lowers the effort of your next step. If you have to manually rework the output, revise the prompt around that friction. Try it: Pick one AI answer you copied into a document or spreadsheet. What manual edits did you make? Prompt for that final form next time and compare the time needed to use it. Keep learning: Structured Outputs: When a Schema Beats More Prompt Words, What to Put in a Reusable Prompt Template, and How to Tell Whether an AI Answer Is Any Good.