Hallucination Half-Life

Hallucination Half-Life is the interval between an AI model first stating something false about a client and the moment that falsehood stops propagating through downstream answers. Models do not correct themselves on a schedule; they inherit stale or wrong claims from indexed pages, then restate them with fresh confidence. For agencies, the practical unit of work is not the correction itself but the propagation window: how many weeks a wrong claim keeps resurfacing across ChatGPT, Perplexity, and Google AI Overviews after the source page is fixed. A single misstated service area or pricing figure can outlive the retainer that produced it. The framework forces a measurement step before any remediation spend, because the half-life varies by claim type, citation density, and how many third-party pages repeat the error. Without that measurement, agencies bill for content edits that never reach the model layer.