Key Social Norms Take-aways This study demonstrates that digital, behaviorally informed interventions can influence deeply rooted gender norms at scale. It is one of the first randomized trials to show that a chatbot can both improve gender attitudes and reduce intimate partner violence exposure among young women. The findings expand the evidence base on how technology can be harnessed to shift norms around relationship power, emotional abuse, and control, domains traditionally addressed through in-person interventions. This research highlights the potential of scalable digital tools to complement or extend existing violence prevention strategies, particularly in low-resource settings where access to in-person programming may be limited.
Objectives The study aimed to test whether two digital chatbots could improve gender attitudes and reduce intimate partner violence among young women in South Africa, and to assess their effects on identification of unhealthy relationship behaviors and depressive symptoms.
Methods An individually randomized controlled trial was conducted with over 19,000 women aged 18–24, recruited through Facebook and engaged via WhatsApp. Participants were randomly assigned to one of four arms: a pure control, an information-only chatbot, a gamified behaviorally-informed chatbot, or a narrative chatbot. Outcomes were measured using adapted versions of validated scales for gender attitudes, intimate partner violence, identification of unhealthy behaviors, and depressive symptoms. Analyses used ordinary least squares regression with robust standard errors and multiple comparison adjustments.
Intervention Details The gamified chatbot guided users through six modules addressing relationship power, communication, consent, safety planning, and emotional coping, with motivational elements like emoji rewards and interactive decision-making. The narrative chatbot told the story of a young woman navigating a difficult relationship, encouraging users to help her make safer choices. Both chatbots were grounded in behavior change theory and aimed to promote reflection and practical skill-building in intimate relationships.
Results All chatbot versions improved gender equitable attitudes compared to the control, with the strongest effects observed in the gamified arm. Only the gamified chatbot reduced past-month intimate partner violence, with a 7 percent reduction in reported exposure. It also increased users’ ability to identify unhealthy relationship behaviors. Neither chatbot affected depressive symptoms. The narrative chatbot, while improving attitudes, did not reduce violence exposure.
Conclusions Behaviorally-informed digital chatbots are a promising, scalable intervention for shifting gender norms and reducing violence among young women. Interactive and engaging formats that build skills and foster reflection appear more effective than passive, narrative-based approaches. These findings highlight the potential of digital platforms to support safer relationships in low-resource settings.