All terms

Rubric Grading

Scoring an AI's response against a specific, predefined checklist of criteria rather than a general "is this good" judgment.

A rubric breaks quality down into concrete, checkable items - did the response cite a source, did it address every part of the question, did it avoid a specific known error. Rubric-based evaluation is more consistent across different graders than open-ended judgment, which is why most serious evaluation platforms use it.

Example

A rubric might list five yes/no checks - "addressed all parts of the prompt," "no factual errors," "appropriate tone," "cited a source where needed," "correct final answer" - and you tick each one independently.

Common questions

Most platforms have a free-text or "other issues" field for exactly this - use the rubric as the primary structure, not the only thing you can flag.

Related terms