A structured extraction prompt should define failure
A useful extraction prompt does more than describe the desired fields. It defines the schema, states which source text supports each value, explains how to represent missing information, and forbids guessing.
Ask for `null` when the document does not contain a value. Use enums for bounded categories, strict types for dates and numbers, and a separate evidence field when auditability matters. Validate the output in code even when the model supports schema-constrained responses.
The most important instruction is the failure rule: if the source does not support a value, the model must not manufacture one to satisfy the schema.
- Category
- Data
- Platform
- Web