Total: 1
Digital portals in retirement communities often create physical and cognitive barriers for older adults, leading to digital avoidance. Generative AI offers a solution by enabling natural language interaction, yet its adoption is hindered by the opaque, "Black Box" nature of these systems and lingering usability challenges. To address this, we evaluated a voice-enabled Large Language Model (LLM) chatbot at a continuing care retirement community in the Pacific Northwest. Through a mixed-methods Co-Design and Literacy Workshop (N=25), we applied a "Glass Box" approach combining multimodal accessibility with intentional AI education. The intervention significantly improved participants' technical understanding (p=0.004) and perceived transparency (p=0.001), shifting their interaction model from blind trust to informed reliance prioritizing verifiable evidence. While voice input reduced cognitive load, usability scores dropped significantly for users aged 80 and older (r=-0.50), indicating that truly age-inclusive AI must evolve beyond touch-based interfaces toward zero-touch navigation.