Skip to content
Technical GlossaryNatural Language Processing

Summary Faithfulness

A quality dimension describing how faithfully a generated summary remains grounded in the source text.

Summary faithfulness is a key concept showing that fluent output alone is not enough in generative summarization systems. A summary may be concise and impressive, but if it introduces unsupported information, it becomes operationally risky. For this reason, summary quality should be evaluated not only by overlap metrics such as ROUGE, but also by factual accuracy and grounding in the source. It is indispensable for enterprise use.