Research on social robots in dementia care has focused on their effects, for example in relation to the patients' wellbeing or the care-givers' working environment. Such approaches to social robots treat them as stable objects with a singular function. Combining social gerontology with social studies of science, the current study offers a new angle by asking: How do patients and care-givers in care homes for older people establish a shared definition of the situation in interactions involving robot animals? Drawing on ethnography and multimodal conversation analysis of 211 minutes of video recordings in two care homes in Sweden, we demonstrate the embodied work by which participants in interactions establish activities with robot animals. In contrast to the ideal of transparency in social robotics, we show that a central affordance of the robots is their vagueness, which allows for their inclusion in playful interactions. Playful framings of the robots highlight their social functions and downplay care-giver-patient asymmetries. However, situations where patients resist a playful frame actualise a dilemma of social inclusion, on the one hand, and the right to not participate in play, on the other. Showing this, the article contributes to knowledge on how people age with technology; in particular, it draws attention to the limits of an ideal of transparency when social robots are included in dementia care.