Successful cases of sales promotion using a self-recommendation robot (SRR) are increasing. SRR is a robot system that moves a product being promoted and makes it appear as if it were a robot itself. Namely, for the robot, recommending itself is the same as recommending the product. In the case of actual sales promotion, the SRR is expected to express emotions such as “I am happy to see you,” as in the case of sales promotions by conventional robots. However, since the SRRs do not have arms, legs, or faces, their behaviors and appearances differ from humanoid or anthropomorphic robots. Hence, it is impossible to apply the findings on the emotional expressions of conventional robots to the SRRs. There has still been insufficient research on the emotional expressions of the SRR, and speech and interaction design have been based on the knowledge and senses of individual developers. Therefore, we conducted an impression evaluation survey to clarify the factors that form the impression of the SRR. Referring to Mehrabian’s experiments, we evaluated SRR’s impressions by dividing emotional expressions into “motion (visual) information”, “vocal (auditory) information” and “verbal information” in our experiment. The results showed that vocal Information has a significant impact on forming the impression of the SRR.