In order for a robot to be perceived as a socially intelligent agent, apart from being capable of displaying different personality traits and communication styles, it needs to be endowed with mechanisms that can mimic those capabilities. Backchannelling behaviour is one of those. In this chapter, the implications of robot backchannelling, namely, the social cues conveyed by the robot to provide an instantaneous response to humans, are investigated. Firstly, a backchannelling signal is modelled in terms of non-lexical and non-verbal social features as a result of the naturalistic observations carried out in situ with patients. Secondly, a user study in the real world with 114 participants is conducted to evaluate whether the signal can be grasped. Finally, its legibility and impact on patients’ performance, cognitive workload, perception of the robot, and overall experience with it are assessed with patients during cognitive training therapy (N \( = \) 16). According to the results, patients are able to recognise such feedback and perceive the robot as more intelligent, efficient, and stimulating, but they commit more mistakes. The chapter concludes by discussing the implications of the findings when deploying robots in sensitive roles and possible solutions to address unexpected behaviours that arose during the study.