Parametric memory auditing probes what language models remember natively from their training weights by executing multiple ungrounded iterations across major platforms like Google Gemini, OpenAI GPT, and Anthropic Claude. This process measures the stability, rank order, and token-level confidence of unprompted brand associations, extracting raw statements and distilling them into a canonical list of brand claims. Conversely, grounded memory auditing disables parametric assumptions, using recursive web search and URL context tools to establish an objective baseline of factual, citation-backed statements currently discoverable on the live web.

Both data streams are then processed by the automated Fact & Gap Checker, which cross-references unprompted recall against real-time research to evaluate brand positioning and uncover factual vulnerabilities:

• Confirms: Verifies that recalled parametric claims match live web evidence, marking them as accurate and binding verifiable source URLs to the claim.
• Contradicts: Detects hallucinations, outdated legacy data, or competitor misattributions in parametric memory, providing direct source URLs that refute the model's false recall.
• Gaps: Surfaces critical brand facts, services, or credentials that exist on the live web but are completely missing from model memory, pinpointing blind spots where search-engine grounding is mandatory for brand discovery.
• Human Rulings & Regrounding: Preserves manual overrides so automated scans never overwrite human-vetted truths, while tracking how newly discovered live evidence continually reinforces or shifts existing knowledge gaps over time.