This article presents preliminary work from the ARIA project, which aims to develop emotionally intelligent social robots specifically designed for long-term interaction with people affected by Alzheimer’s. In particular, the work starts from the analysis of recent research in the same context, identifying aspects that still represent limitations in this scenario for the practical usage of social robots, i.e., the lack of long-term capabilities, strategies to personalize the interaction, and software architectures capable of properly considering the robot’s and user’s emotions in the interaction. To explore the desiderata for socially assistive robots that can play a role in this context, we conducted meetings with caregivers and experts in the field to produce a set of scenarios and guidelines that will drive the development of the robot’s software architecture. The article describes the methodology used to define these guidelines, together with preliminary results and valuable insights into the practical integration of social robotics in assisting users with activities of daily living.