Forcing a model to reply in a strict machine-readable format rather than free text, so its answer can be fed directly into other software without anything having to interpret it. Essential for anything automated, since prose replies that are almost the right shape will break a program.
Technical view
Implemented either by prompting alone, which is unreliable, or by constrained decoding, where the sampling process is restricted at each step to tokens that keep the output valid against a schema. The latter guarantees syntactic validity but not semantic correctness: a model can return well-formed output containing the wrong values.
Related
Part of AI Terminology Index, section: Controlling and Extending Behaviour at Run Time.