Perceived quality of AI-generated and human-written news: a cross-national experimental study among journalism students in China and Vietnam

Abstract

Introduction Generative artificial intelligence is reshaping news production, yet how journalism students evaluate AI-generated content and whether cultural-psychological factors systematically shape these evaluations remain poorly understood, particularly in non-Western contexts. Methods This study employed a 2 (Country: China/Vietnam) × 2 (Author Type: AI/Human) × 4 (Topic: Nature/Health/Society/Finance) mixed-design ANOVA with 140 journalism and communication students under blind conditions. Participants evaluated news quality across seven dimensions without knowing whether articles were AI-generated or human-written. Results AI-generated news received significantly higher overall ratings than human-written news. This advantage was not uniform: a significant Country × Authorship interaction indicated that Chinese participants showed a substantially larger preference for AI-generated news than Vietnamese participants, whose evaluations of the two content types did not differ significantly; a significant Authorship × Topic interaction indicated that the AI advantage was more pronounced for Nature and Society news than for Health and Finance news; and the three-way interaction was not significant. AI-generated news outperformed human-written news specifically on structural dimensions (perceived headline relevance, coverage adequacy, readability, and punctuation and sentence structure) but not on perceived factual accuracy or source usage. Gender predicted evaluations of both news types without a significant authorship-specific asymmetry, whereas news interest and news trust predicted evaluations of human-written news more strongly than AI-generated news, a pattern confirmed by direct interaction tests. Discussion These patterns are discussed in relation to culturally grounded orientations toward institutional authority and technology, including the authority heuristic and techno-optimism, although these constructs were inferred from country-level cultural indices rather than directly measured among individual participants. A post hoc textual analysis found that AI-generated articles were not lexically less diverse than human-written articles, which complicates homogenization-based accounts of the cross-national convergence observed in AI-generated news evaluations. These findings extend cross-cultural psychology by illustrating both the promise and the limits of using country as a proxy for culturally embedded psychological constructs in research on AI-generated content.