JMIR Publications

ABSTRACT

Background:

Adolescent girls and young women (AGYW) in South Africa face overlapping sexual and reproductive health challenges, including high unintended pregnancy rates, disproportionate HIV burden, and limited access to youth-friendly services. AI-powered digital health tools can bridge gaps in accessible, stigma-free sexual and reproductive health (SRH) support, yet AGYW’s experiences with these technologies in the Global South remain under-researched. We explored experiences of acceptability and safety of two AI-powered chatbots for HIV prevention, SRH, and wellbeing among AGYW.

Objective:

This study explored the acceptability and safety of two AI-powered chatbots for HIV prevention and SRH among adolescent girls and young women in South Africa, trained on South African vernacular and national clinical guidelines.

Methods:

Convergent parallel mixed-methods research was embedded within combination HIV prevention programming across five high-burden districts in two provinces of South Africa (June-November 2025). Three focus group discussions (N=27) explored acceptability and safety among AGYW peer providers (aged 20-32) across three urban, peri-urban, and rural sites. Reflexive thematic analysis was used. A digital survey of AGYW service users (N=961, aged 18-24) measured acceptability of the two AI-powered HIV/SRH chatbots using the validated Acceptability of Intervention Measure (AIM), with descriptive statistics and correlation analyses examining sociodemographic factors associated with acceptability.

Results:

Survey participants (mean age 21.5 years; 73.1% completed secondary education) reported high acceptability: 94.6% reported satisfaction, 91.8% would use again, and 93.3% would recommend the AI-powered tool to others. Median AIM score was 17 (IQR 16-20; Cronbach's α=0.903). Most reported using chatbots for sexual health information (59.2%) and reproductive health (50.1%). 86.8% of service users trusted the chatbots’ information accuracy. Qualitative safety concerns amongst peer service providers appeared to be influenced by prior social media experiences, including worry about personal data handling. Qualitative findings revealed nuanced perceptions and concerns: AGYW sought deeper relational qualities beyond information provision, expressing unmet expectations around empathetic tone and validation. Geographic variation in acceptability emerged, with rural/peri-urban participants showing greater concerns about data privacy, though this could be due to variation in tool exposure or residence.

Conclusions:

While quantitative metrics indicate high acceptability of AI-powered SRH chatbots among South African AGYW, qualitative insights reveal important considerations around contextual relevance, empathy, and digital literacy gaps. Findings underscore the need for differentiated implementation strategies that address geographic, social and individual variations, strengthen data safety features, and balance innovation with ethical guardrails, ensuring AGYW-centered AI integration in HIV and SRH programming.