JMIR Publications

ABSTRACT

Background:

Gender-based violence is a major public health issue with substantial psychological, social, and health consequences for women worldwide and in Italy specifically. Digital interventions based on conversational agents may offer scalable opportunities for prevention. Recent approaches increasingly combine structured psychoeducational content with AI-enabled conversational components to enhance personalization and user engagement.

Objective:

This pilot study aimed to evaluate the usability, acceptability, and perceived impact of a Telegram-based chatbot intervention for gender-based violence prevention in a sample of young male users, and to identify key human-factor and design issues relevant to the adaptation of the intervention for this population.

Methods:

A mixed methods pilot study was conducted with 11 male participants recruited in Trento, Italy. Participants tested a Telegram-based hybrid chatbot integrating predefined prevention pathways with AI-supported conversational interactions addressing healthy and unhealthy relationship dynamics, warning signs of abuse, self-reflection, and help-seeking. Quantitative evaluation integrated a semantic differential, selected dimensions from the user version of the Mobile Application Rating Scale (uMARS), and items derived from the Comprehensive Evaluation Scale for LLM-Powered Counseling Chatbots (CES-LCC). One-sample Wilcoxon signed-rank tests were used to compare ratings against the neutral midpoint of the Likert scale (μ=3), and rank-biserial correlations were reported as effect sizes. Qualitative feedback was collected through semi-structured interviews and analyzed thematically.

Results:

Quantitative findings showed a selective rather than uniformly positive pattern. In the semantic differential, only single-interaction duration was significantly above the neutral midpoint (mean 3.77, median 3.50, W=34.0, P=.028, r=0.889). In the uMARS-derived dimensions, no domain differed significantly from neutrality. In the CES-LCC-derived dimensions, perceived helpfulness/credibility of information (PHI: mean 3.91, median 4.00, W=52.5, P=.012, r=0.909), language quality (LQ: mean 4.42, median 4.67, W=65.0, P=.005, r=0.970), and trust (T: mean 3.64, median 3.67, W=56.0, P=.044, r=0.697) were significantly above the midpoint. At the aggregate level, Communication (mean 3.58, median 3.70, W=59.5, P=.021, r=0.803) and Session Structure (mean 3.77, median 3.77, W=34.0, P=.028, r=0.889) were significantly above neutral.

Conclusions:

The intervention showed promising features in terms of communication quality, language quality, trust, and perceived informational credibility among young male users. At the same time, the findings highlight clear redesign priorities concerning personalization, transparency, conversational memory, interaction flexibility, and gender-sensitive framing.