Parametric memory recall evaluates what large language models remember about a brand natively without live web search, utilizing zero-shot prompts and token-level confidence scores to uncover unprompted brand associations and factual hallucination gaps.

flowchart TD
    A[Ungrounded Prompt] --> B[LLM Parametric Inference]
    B --> C[Token Logprobs & Output]
    C --> D[Extracted Brand Facts]
    E[(Canonical Brand Facts)] --> F[Discrepancy Analysis]
    D --> F
    F --> G[Confidence Scores]
    F --> H[Hallucinations & Gaps]