Unselected sources serve as critical model-behaviour metadata indicating which retrieved web pages an AI model considered during search grounding but declined to reference directly in its final synthesized answer. Tracking these ignored documents enables the calculation of a domain's selection percentage and Trust Ratio metric (selected versus unselected citations) to uncover which web properties models deem authoritative versus untrustworthy. Analyzing this drop-off on a per-provider basis reveals systemic model preferences and highlights primary competitor brand matches that are repeatedly retrieved yet excluded from citations.