A human-centric approach for responding to online soft hate speech: the effect of AI-based persuasion on the comment modification intention of commenters

The development of digital technology has facilitated the 'democratisation' of the internet, but it has also provided opportunities for online hate speech, which can have negative effects on individuals, groups, and society. Therefore, addressing online hate speech is crucial. However, owing to the paradox between free speech and the regulation of hate speech, soft hate speech, which is morally unacceptable but not directly constrained by law, as hard hate speech is, can result in harmful consequences in more subtle forms. In response, this study proposes shifting the focus of online hate speech research to individuals, with the objective of influencing the comment modification intention of soft hate speech commenters through AI-based persuasion and its associated strategies. Specifically, Study 1 establishes baseline effects by comparing AI-based persuasion against no-persuasion and irrelevant-persuasion controls. A pilot study examines how participant perspective moderates strategy effectiveness, providing the empirical basis for Study 2’s design. Study 2 further examines the comparative effects of emotional labelling and reason-explanation strategies and the mediating role of perceived persuasiveness under higher ecological validity conditions. In summary, this study explores a human-centric approach to responding to online soft hate speech and, through controlled scenario-based experiments, provides preliminary evidence that AI-based persuasion has a positive effect on scenario-based comment modification intention, with perceived persuasiveness identified as a psychological mechanism underlying this effect.