A qualitative study assessing the acceptability of a multi-agent AI Chatbot for providing HIV and mental health support among men who have sex with men and transgender women in KwaZulu-Natal, South Africa.

The AI chatbot was generally acceptable to MSM and transgender women, with privacy, convenience, and human-like interaction identified as key facilitators of acceptability.

• Study conducted at the Aurum POP INN clinic in Pietermaritzburg, KwaZulu-Natal, South Africa.
• Ten participants engaged in in-depth interviews after interacting with the chatbot.
• An additional 34 participants experienced both chatbot and in-person counselling through a randomised crossover design and then participated in four focus group discussions.
• Analysis was guided by the Unified Theory of Acceptance and Use of Technology and the Acceptability of Healthcare Interventions Framework.
• Acceptability was enhanced by associations with modernity and anonymity.

Transgender women and men who have sex with men are disproportionately affected by HIV and mental health challenges, and gaps in mental health service delivery present challenges for scalability in public health systems.

• Mental well-being influences uptake and adherence to HIV prevention and treatment.
• The study population consisted of TGW and MSM, who represent key populations in HIV care.
• The chatbot was designed as a novel, scalable solution to expand access to mental health support for these populations.

The multi-agent AI chatbot was designed to simulate supportive counselling based on the Inuka model.

• The chatbot was piloted with TGW and MSM at a dedicated clinic in Pietermaritzburg.
• A qualitative study design was used, combining in-depth interviews and focus group discussions.
• The randomised crossover design allowed 34 participants to experience both chatbot and in-person counselling for comparison.

Trust, usability, and accessibility were identified as factors that improved engagement with the AI chatbot.

• Participants valued the chatbot's privacy and anonymity features.
• Human-like interaction was specifically noted as a positive attribute of the chatbot.
• Associations with modernity contributed to enhanced acceptability among participants.

Key barriers to chatbot acceptability included slow response times, limited rapport, and repetitive messaging.

• Slow response times were identified as a barrier to engagement.
• Limited rapport between participants and the chatbot was noted as a challenge compared to in-person counselling.
• Repetitive messaging was flagged as reducing the quality of the chatbot interaction experience.

AI chatbots were concluded to offer a promising, scalable approach to supporting mental health among key populations in HIV care.

• The study supports AI-driven chatbots as a potential solution to address gaps in mental health service delivery.
• Scalability was emphasized as a key advantage of the chatbot approach in public health systems.
• The findings suggest chatbots could complement existing HIV care services for MSM and TGW.