生成式人工智能标签与新闻信任悖论：信源权威性的调节作用

Artificial Intelligence Labels and the News Trust Paradox: The Moderating Role of Source Authority

摘要: 生成式人工智能正深度嵌入新闻生产与传播的全链条，AI内容标识制度随之落地。然而，标识能否修复公众信任，仍是一个有待检验的实证问题。本研究基于预期违背理论，采用2 (内容来源标签：AI生成vs.人类撰写) × 2 (信源权威性：权威媒体vs.自媒体)被试间实验设计，系统考察了内容来源标签与信源权威性对公众新闻信任的交互效应及其心理机制。研究发现：AI标签对新闻可信度存在显著的负面主效应，但这一效应高度依赖于信源权威性，当权威媒体标注AI生成时，受众感知可信度大幅下降；当自媒体标注AI生成时，信任折损并不显著，此即“信任悖论”的实证确认。有调节的中介分析表明，感知信任在“AI标签→信息接受意愿”之间发挥中介作用，且该中介效应仅在权威媒体情境下成立。三重交互分析进一步揭示，既有新闻信任构成第二层边界条件，高信任群体受“信任悖论”的冲击尤为剧烈。本研究为“信任悖论”提供了机制层面的实证解释，拓展了预期违背理论在AI传播情境中的应用边界，并为AI内容标识制度的优化与权威媒体的智能化转型提供了实证依据。

Abstract: Generative artificial intelligence (AI) is deeply embedded across the entire chain of news production and dissemination, and AI content labeling systems have been implemented accordingly. Whether labeling can restore public trust, however, remains an empirical question. Drawing on Expectancy Violations Theory, this study uses a 2 (content source label: AI-generated vs. human-written) × 2 (source authority: authoritative media vs. self-media) between-subjects experimental design to systematically examine the interaction effect of content source labels and source authority on public news trust and its psychological mechanism. The study finds that AI labels have a significant negative main effect on news credibility, but this effect is highly contingent on source authority: when authoritative media label content as AI-generated, perceived credibility drops sharply; when self-media do so, the trust penalty is not significant. This empirically confirms the “trust paradox.” Moderated mediation analysis shows that perceived trust mediates the path from AI labels to willingness to accept information, and this mediation effect holds only in the authoritative media condition. Three-way interaction analysis further reveals that prior news trust constitutes a second-layer boundary condition, with high-trust groups being particularly severely affected by the “trust paradox.” This study provides a mechanism-level empirical explanation of the “trust paradox,” extends the applicability boundary of Expectancy Violations Theory in AI communication contexts, and offers empirical evidence for optimizing AI content labeling systems and the intelligent transformation of authoritative media.