Citation Choices in AI-Generated Fact-Checking: Perceptual Outcomes and Corrective Effectiveness in Political Fake News

As artificial intelligence (AI) is increasingly deployed in fact-checking, questions remain about how audiences perceive AI-generated verdicts and the sources used to justify them. Drawing on motivated reasoning and research on source credibility, this study examines how fake news congruence and AI-cited source congruence jointly shape belief correction and message- and agent-level perceptions. In a U.S.-based 2 * 3 online experiment (N = 682), participants evaluated partisan misinformation followed by AI-generated verdicts citing politically congruent, incongruent, or third-party sources. Results show that while partisan congruence of fake news systematically shaped perceptions of AI verdicts and agents, source congruence operated conditionally, enhancing corrective effectiveness under co-directional configurations (citing in-group source to debunk in-group fake news). Findings highlight the role of source citations and expectation-based mechanisms in AI-generated fact-checking.