Technique: Evaluation Framework Prompting
LLMs can act as judges and evaluators when given explicit scoring criteria. This technique provides a rubric — a set of dimensions with clear definitions — and asks the model to score content against each dimension independently. This mirrors how human experts evaluate work and is the foundation of “LLM-as-Judge” evaluation systems.
The key to effective evaluation prompting is making criteria specific and measurable. “Is it good?” produces vague results. “Rate clarity on a 1-10 scale where 1 means jargon-heavy and 10 means a non-expert could understand it” produces actionable feedback. Always ask for justification alongside scores — the reasoning is often more valuable than the number.
When to use: Quality assurance for AI-generated content, peer review automation, grading rubrics, A/B testing analysis, content moderation, and building evaluation pipelines for prompt optimization.