Factual hallucination gaps are critical discrepancies between an LLM's ungrounded parametric memory and verified, real-world source data. Systems like the Fact & Gap Checker and the Veracity Engine systematically isolate these divergence points by contrasting live-researched web sources against raw model outputs.

Through these engines, platforms analyze three key failure modes in generated claims: - Omissions (Lost facts): Important, verifiable brand attributes or offerings that the model fails to recall entirely. - Distortions: Source claims that are misrepresented, taken out of context, or altered during generation. - Fabrications: Entirely hallucinated claims lacking any grounding in source material.

By quantifying the proportion of source facts that survive intact versus those that are lost or distorted, these tools assign precise fidelity scores to pinpoint where generative models need targeted optimization.