Tell Me Why, When, and How: Effects of Personality, Evidence, and Context on Older Adults' Perceptions of Conversational AI Explanations

Abstract

Large Language Model-based voice assistants (LLM-VAs) have shown potential to support older adults aging in place through proactive reminders, health information, and everyday assistance. As LLM-VAs become more conversational, designing explanations of their behavior requires understanding what information they communicate and how they deliver it. While prior work has studied older adults' explainability requirements, relatively little is known about their perceptions of explanations when an LLM-VA's conversational personality varies. To address this gap, we conducted a mixed-design study with 140 older adults, examining two dimensions of explanation design: evidential grounding (source of information) and the VA's conversational personality (agreeableness and extraversion), across two contexts of use. Our results show that both dimensions influenced older adults' perceptions in distinct but interdependent ways, challenging one-size-fits-all explanation strategies. We conclude with practical design implications for coordinating explanation content, conversational delivery, and context in future LLM-VAs supporting aging in place.