Supporting bystander intervention through conversational AI: Experimental evidence from undergraduate samples.

Across campuses and workplaces, people are receiving bystander intervention training to empower them to support victims and mitigate situations of harm. But even after this training, many people remain insecure about their ability to intervene and hesitate to act when intervention is needed. One promising solution to bystanders' ongoing inaction is to provide a safe, judgment-free environment where people can repeatedly practice skills and receive personalized feedback. Here, we explore that solution through a chatbot-based bystander intervention practice tool, where an AI agent guides users through evidence-based training. Across two preregistered experiments with undergraduates-a pilot study conducted on a local college campus followed by a larger online study-we found preliminary evidence that chatbot-guided training may strengthen several motivational outcomes associated with bystander intervention, including bystander efficacy, attitudes, psychological safety, and norm perceptions. This work provides preliminary evidence of the impact of chatbot-guided bystander intervention training. The findings highlight the potential of AI as a pedagogical tool for supporting bystander intervention and provide an empirical foundation for future research on AI-assisted intervention training.