Large language model-supported companion robots for loneliness in older people: A UK–Japan qualitative study integrating focus groups and in-home deployment

Background: Loneliness is a critical social issue in older adults, with clinical implications. Conversational companion robots have been studied as one approach to ameliorating this issue. Large language models (LLMs) can enable flexible conversational ability in companion robots, but acceptability and suitability remain uncertain. We explored older people’s expectations and concerns regarding an LLM-supported companion robot for loneliness support. Methods: We conducted a UK–Japan qualitative study comprising hands-on focus groups for community-dwelling older adults in London (n=17) and a one-week in-home use with follow-up interviews in Osaka among outpatients with mild cognitive impairment (MCI; n=8). Transcripts were analysed using reflexive thematic analysis; for cross-site reporting, Japan themes/codes were mapped onto the thematic structure generated from the larger UK dataset. Descriptive questionnaire measures and at-home conversational log metrics were collected to contextualise qualitative findings. Results: Participants saw value of the companion robot as a support for older people with loneliness but emphasised that acceptability depends on interaction mechanics and user agency. Three cross-context themes were identified: (1) Practical use and functionality (response latency, turn-taking, desired features, and controllability in home use); (2) Emotional connection and engagement (social presence alongside perceived limits in conversational fit and depth); and (3) Ethical and societal reflections (privacy/data governance, access, and concerns about substituting for human contact). Conclusions: LLM-supported companion robots may provide acceptable low-intensity support for some older people, including those with MCI, provided that usability, user-adjustable control and ethical governance are prioritised. Longer deployments are needed to evaluate potential sustained benefit and burden.