Correcting Misinformation in an Authoritarian Context: Logical Fallacy Training via AI-Generated Short Videos in China

(with Muyao Hang and Zhejun Qiu, Univeristy of Pittsburgh)

Logical fallacies are often embedded in misinformation to mislead the public and manipulate opinion, yet scalable interventions remain understudied in authoritarian and non-English contexts. We test an AI-generated short-video intervention that trains participants to recognize three logical fallacies commonly used in misinformation: false causality, hasty generalization, and personal attack. In a preregistered survey experiment conducted in China in August 2025 (N=2,421), participants viewed either a logical fallacy training video or a placebo video before evaluating misinformation containing the corresponding fallacy. Results show that the intervention reduced endorsement of misinformation overall. However, the effects varied by fallacy type: training reduced endorsement of misinformation containing hasty generalization and personal attack, but not false causality. We find little evidence that motivated reasoning weakened the intervention. Although nationalist participants were more likely to endorse nationalist misinformation, they were not less responsive to the treatment than non-nationalists. The intervention was also more effective among participants with lower cognitive reflection ability, narrowing the gap in misinformation endorsement between low- and high-CRT participants. These findings extend misinformation-intervention research to an understudied authoritarian context and highlight the promise of scalable, politically neutral, reasoning-based interventions.