Abstract
Generative answer systems increasingly mediate access to news through synthesized responses accompanied by a small set of citations. Yet citation presence does not establish that a claim’s epistemic lineage remains recoverable. This article develops a journalism-specific framework for claim-level provenance auditing in consumer-facing AI-mediated news. Journalism specificity lies not in treating provenance as unique to news, but in distinguishing journalistic reporting origin from evidentiary origin and in coding attribution, syndication, non-public evidence, temporal updating, and material epistemic qualification as constitutive source relations. The framework separates five analytically distinct dimensions: citation addressability, claim-source support, reporting-origin traceability, evidence-origin traceability, and epistemic-status retention. It also treats source displacement as two independently codable profiles—reporting-origin displacement and evidence-origin displacement—so that both can occur in the same claim lineage. The framework is operationalized through an independent provenance-dossier method, explicit coding values and decision rules, and three constructed worked cases that demonstrate application without being presented as empirical validation. Rather than collapsing the dimensions into a single score, the approach preserves their diagnostic value and is designed to support reproducible human-coded audits and subsequent empirical validation. The paper contributes a conceptual vocabulary and methodological protocol for studying how generative systems reorganize the visibility of journalistic work, evidence, and uncertainty at the point of news consumption.