Humans excel at understanding social cues in communication, but robots struggle. Social cues are crucial for humans to interpret the intentions of their communication partners. Research indicates that we typically interpret the actions of anthropomorphic robots analogously to their human counterparts, paving a clear path to the design of appropriate social cues. For non-anthropomorphic robots, however, it is an open question how humans interpret social cues with different output modalities and in different contexts. Our study investigates whether social cues signaled by typical non-anthropomorphic modalities such as lights, sounds, and gestures are consistently interpreted across people and contexts. We, therefore, conducted a contextual investigation in a hospital, derived scenarios from co-design workshop, and tested 103 cues collected from the literature in a large online survey (N = 1545). Our results demonstrate that most human interpretations vary by context, highlighting the need to design dynamic and adaptive social cues for interactive robotic systems.