Older adults are increasingly turning to chatbot-based mental health support, yet adoption remains limited by barriers in accessibility, privacy, security, and trust. We present a two-phase study on the co-design of MESA-Bot (Mental and Emotional Support Assistant for Older Adults), a non-diagnostic chatbot tailored to later-life needs. In Phase I, we analyzed ten leading mental health chatbots and conducted co-design sessions with N1=10 older adults to identify challenges and inform an accessible, privacy-aware prototype. In Phase II, we evaluated MESA-Bot with N2=28 older adults using semi-structured interviews and structured technical assessments. Participants emphasized transparent consent, supportive tone, and fine-grained data control. Features such as revocable consent, role-based access control, customizable data visibility, and simplified dialogue flows increased trust and usability; 86% found MESA-Bot easy to use. We offer design insights for inclusive, trustworthy mental health technologies that integrate accessibility with verifiable privacy and security protections.
ACM CHI Conference on Human Factors in Computing Systems