The ARC framework establishes a structured methodology for diagnosing and auditing brand salience across major AI models in both ungrounded and web-connected modes. Its Association engine maps multi-directional memory pathways—Entity-to-Brand (E2B), Brand-to-Entity (B2E), and Query-to-Brand (Q2B)—to evaluate brand salience and hierarchical parent-brand relationships in parametric memory. Simultaneously, Relevance probes score categorical alignment by quantifying direct binary recommendation rates across repeated query sampling. Concurrently, the Citation mining pipeline analyses live search-grounded responses to track which specific URLs, publisher domains, and competitor entities are surfaced during real-time retrieval, pinpointing discrepancies between an AI's internal training recall and external web citations.