Aimee, an AI-powered HIV prevention and sexual and reproductive health intervention for adolescent girls and young women in South Africa: Design, pilot study methods, and feasibility

Read the pre-print publication

Background

Adolescent girls and young women (AGYW) in South Africa face persistent structural, stigma-related, and logistical barriers to HIV prevention services. WhatsApp-based conversational AI may extend confidential, around-the-clock support beyond facility settings, but requires human oversight and robust safety governance to be responsibly deployed at scale. This paper describes the design of Aimee, a multimodal, provider-supported AI companion embedded within Shout-It-Now’s community-based HIV prevention programme, and assesses its feasibility among AGYW in Gauteng.

Methods

This single-arm, longitudinal, mixed-methods pilot was conducted between 17 March and 30 November 2025 across the Ekurhuleni and Tshwane districts. The primary outcome was feasibility, assessed via a structured, multi-stage engagement funnel spanning onboarding, conversation depth, return engagement, HIV self-test (HIVST) completion, risk stratification, and clinical triage, alongside AI system accuracy, service uptake, and provider operational performance as secondary outcomes. Acceptability was also assessed through focus groups and provider interviews, with findings reported elsewhere.

Results:

Of 9,958 participants who initiated contact with Aimee, 9,310 (93.5%) enrolled, forming the analytic cohort. Within this cohort, 80.9% sent at least five messages and 58.6% returned on at least two distinct days. Engagement narrowed at higher-intensity milestones with 4.5% of the cohort uploading an HIVST image. Provider tasks were generated for 47.9% of participants, with 93.0% of these receiving outreach; safety-flagged tasks were completed substantially faster (median 13.0–15.3 hours) than routine follow-up (47.3 hours).

Conclusions:

Aimee was feasible to deliver at meaningful scale within a real-world, resource-constrained HIV prevention programme, with high early engagement and operationally sound provider safety workflows. Engagement at higher-intensity milestones, including HIVST uptake, should be interpreted alongside significant service disruptions during the study period. These findings establish a methodological and governance foundation for a forthcoming series of papers reporting on Aimee’s acceptability, safety, and effectiveness.