When the AI author is not disclosed: how cognitive dispositions affect audience perceptions of AI-generated news across topics

Abstract Without explicit cues that specify the AI-authorship, how would individuals evaluate AI-generated news? This study examines this question by focusing on user-level characteristics, encompassing cognitive dispositions, attitudinal orientations, and evaluative competencies. Our survey experiment randomly assigned participants to read a news article—for which the AI authorship was not disclosed—on one of three topics (banking, politics, and entertainment). Results highlight the double-edged roles of people’s actively open-minded thinking, media literacy, and fake news awareness: on one hand, these factors are associated with a more in-depth processing of AI-generated news, suggesting their positive impact on promoting a critical evaluation of AI-generated news; on the other hand, however, these cognitive factors are positively associated with AI-news’s perceived quality or credibility, indicating that readers with these cognitive capacities may simultaneously overestimate the trustworthiness of AI-generated news. These findings call for the media literacy education to go beyond teaching readers to spot misinformation or look for AI disclaimers; instead, it’s urgent to help people establish clear expectations for what human-generated journalism looks like.