Parametric memory reflects an AI model's internalized training weights, capturing unprompted brand associations, sentiment, and core entity concepts without connecting to outside tools. In contrast, grounded memory executes live web search and document retrieval to pull real-time information, citation links, and extracted text snippets from current websites. Setting these two mechanisms against each other allows systems like a Fact & Gap Checker to evaluate brand recall accuracy against live ground truth. This comparison highlights hallucinations and factual contradictions in model memory while identifying knowledge gaps where a brand exists on the web but remains absent from model recall.