Selected citations represent URLs directly attributed in the final generated answer text through inline annotations, confirming their active role in shaping the model's factual claims. In contrast, unselected citations consist of grounding chunks and browsed search results that the model evaluated during web retrieval but ultimately omitted from the final response. Tracking the relationship between these two categories establishes a domain's trust ratio, pinpointing which sources an AI model actively relies upon versus those it considers but discards.