Faithfulness (groundedness)
A RAG metric: the share of claims in an answer that are supported by the retrieved sources, not by the model’s memory or invention.
The usual method splits the answer into individual claims and checks each one against the retrieved context, typically with an LLM as judge; the score is supported claims divided by all claims. A low score means hallucination relative to the sources. It does not check whether the sources are right or whether the answer addresses the question: a faithful answer can still be useless, which is why it is paired with answer relevancy. Some tools call it groundedness.
Related terms
- Hallucination: A fluent, confident answer that is false or not supported by the sources.
- Answer relevancy: A RAG metric that scores how directly an answer addresses the question asked, regardless of whether it is factually correct.
- Grounding: Tying a model’s answer to specific supplied sources, such as retrieved documents or search results, instead of its training memory.
- LLM as judge: Using a model with a rubric to grade another model’s output.
- Ragas: An open-source Python library for evaluating RAG systems with metrics such as faithfulness, answer relevancy, context precision and context recall.
