Socially assistive robots could help to support people's well-being in contexts such as art therapy where human therapists are scarce, by making art such as paintings together with people in a way that is emotionally contingent and creative. However, current art-making robots are typically either contingent, controlled as a tool by a human artist, or creative, programmed to paint independently, possibly because some complex and idiosyncratic concepts related to art, such as emotion and creativity, are not yet well understood. For example, the role of personalized taste in forming beauty evaluations has been studied in empirical aesthetics, but how to generate art that appears to an individual to express certain emotions such as happiness or sadness is less clear. In the current article, a collaborative prototyping/Wizard of Oz approach was used to explore generic robot art-making strategies and personalization of art via an open-ended emotion profile intended to identify tricky concepts. As a result of conducting an exploratory user study, participants indicated some preference for a robot art-making strategy involving "visual metaphors" to balance exogenous and endogenous components, and personalized representational sketches were reported to convey emotion more clearly than generic sketches. The article closes by discussing personalized abstract art as well as suggestions for richer art-making strategies and user models. Thus, the main conceptual advance of the current work lies in suggesting how an interactive robot can visually express emotions through personalized art; the general aim is that this could help to inform next steps in this promising area and facilitate technological acceptance of robots in everyday human environments.