Structured Markdown Outputs

When a human reads the output, structure beats prose. Headers let readers scan, tables let them compare, lists let them act. This skill is about shaping that structure so it's predictable across calls.

For code-consumed outputs (parsing, validation, downstream pipelines), use prompt-engineering-output-json or prompt-engineering-output-xml instead. Markdown is fragile to parse; treat it as a presentation layer, not a contract.

When to use

• Output renders directly to a human (chat UI, doc page, README, email).
• The reader needs to scan (headers) or compare (tables) more than read straight through.
• A downstream renderer supports markdown (Slack, GitHub, Notion, most chat UIs).
• The output mixes prose with enumerated items, decisions, or trade-off analysis.

Choosing the right structure