The parser classifies URLs as selected citations when explicitly cited in the generated answer and unselected sources when browsed during search execution but omitted from the final response text, computing a trust ratio that measures how reliably an AI model cites a retrieved source.
flowchart TD
Raw[Raw LLM Search Output] --> Parser[Citation Parser]
Parser --> Browsed[Browsed URLs<br>Retrieved but omitted]
Parser --> Selected[Inline Citations<br>Explicitly cited in text]
Browsed --> Metric[Citation Trust Ratio]
Selected --> Metric
Metric --> Insight[Source Conversion Rate]Referenced by