The ARC framework serves as the core diagnostic engine within DEJAN's AI influence platform, systematically auditing how major language models perceive, evaluate, and cite brands across parametric and web-grounded environments:

• Association (A): Measures ungrounded parametric memory through three distinct directional probe pathways: Entity-to-Brand (E2B) to determine which brands lead category concepts, Brand-to-Entity (B2E) to uncover which attributes models organically associate with a brand, and Query-to-Brand (Q2B) to map search intent capture. These probe outputs leverage Matryoshka prefix-based hierarchy attribution to aggregate sub-brands and variant naming conventions into unified root-brand share metrics.
• Relevance (R): Evaluates direct model recommendation strength by running repeated binary evaluation probes ("yes/no") across active providers (such as Google Gemini, OpenAI GPT, and Anthropic Claude) to compute query and entity relevance scores (QRS/ERS).
• Citation (C): Mines live, search-grounded model responses to inspect search fan-out queries, parsed grounding metadata, and cited web domains. It differentiates between selected and unselected sources—distinguishing URLs actively cited in the generated answer from those merely retrieved during search—to measure owned-domain citation share against third-party authorities.

By continuously triangulating these three pillars, the system identifies critical perceptual discrepancies between an AI model's internal training weights and its live web retrieval behavior, providing actionable data for content and snippet optimization.