Glossary
Instruction Ambiguity
The quality of an instruction or prompt that allows for multiple reasonable interpretations, potentially leading to inconsistent or incorrect model responses.
Instruction ambiguity occurs when what you asked the model isn’t clear enough to determine a single correct response — and different reasonable interpretations lead to very different outputs. A request like “make this more professional” could mean adding formal vocabulary, removing contractions, restructuring sentences, or some combination of all three. When a model encounters an ambiguous instruction, it must choose an interpretation — and that choice may not match what you intended. For behavior architects, ambiguity testing is a useful practice: deliberately submitting instructions to different models or across different sessions to see which interpretations they choose, and then revising instructions to be more specific. Ambiguity in system prompts is particularly costly because it affects every interaction.