Objectives: We undertook a cluster-randomised controlled trial exploring the effect of a therapeutic companion robot (PARO) compared to a look-alike plush toy and usual care on dementia symptoms of long-term care residents. Complementing the reported quantitative outcomes, this paper provides critical reflection and commentary on individual participant responses to PARO, observed through video recordings, with a view to informing clinical practice and research. Method: A descriptive, qualitative design with five participants selected from the PARO intervention arm of the trial. The trial is registered with the Australian New Zealand Clinical Trials Registry (ACTRN12614000508673). Results: The five participants and their responses to PARO are presented in terms of three issues: i.) Different pre-intervention clinical presentations and different responses; ii.) Same individual, different response – the need for continual assessment and review; and iii.) The ethics of giving and retrieving PARO. Implications for clinical practice and future research are discussed in relation to each issue. Conclusion: The findings suggest that one approach does not fit all, and that there is considerable variation in responses to PARO. A number of recommendations are discussed to aid the delivery of psychosocial interventions with PARO in practice, as well as to guide future research.