Course contentsShow
← AI Engineering
Lesson 368 of 1,8869. Retrieval-Augmented Generation (RAG) - BasicsFree lesson

Citation Quality Metrics

Measuring citation accuracy, coverage, and relevance to evaluate RAG system reliability and source usage.

Citation Quality Metrics

You'll learn: How to systematically measure and evaluate whether your RAG system cites sources accurately, completely, and relevantly.

What Are Citation Quality Metrics?

Just like a research paper can have good or bad citations, your RAG system's source references need evaluation. Citation quality metrics are standardized measurements that help you assess whether your system is attributing information correctly, covering all sources it should, and only citing relevant material.

Without these metrics, you're flying blind—you won't know if your system hallucinates sources, ignores important references, or overwhelms users with irrelevant citations.

The Three Core Dimensions

Citation Accuracy

Does each citation actually support the generated statement? This measures whether the system is "making up" sources or correctly linking claims to retrieved chunks. Think of it as fact-checking: did the model really get that information from where it claims?

Citation Coverage

Are all relevant sources cited? Even if the system uses three documents to answer a question, did it acknowledge all three? Missing citations means losing transparency and potentially legal compliance.

Citation Relevance

Are the citations helpful or noisy? A system that cites 10 documents for a simple answer when only 2 are truly relevant creates cognitive overload. This metric filters out "citation spam."

Measuring in Practice

You typically compute these as percentages or scores during evaluation. For example: "85% citation accuracy" means 85% of citations pointed to chunks that actually contained the claimed information. You'll compare generated citations against your ground truth data or human annotators.

Key Takeaway: Citation quality metrics—accuracy, coverage, and relevance—let you objectively measure whether your RAG system is a reliable, transparent source of information rather than a black box that occasionally guesses where facts came from.