Automatic imitation, demonstrated frequently with simple finger movements, refers to the tendency of individuals to involuntarily imitate observed task-irrelevant movement of others. In a meta-analysis (Cracco et al., 2018) it was revealed that automatic imitation, contingent upon the similarity between the model and imitator, lacks unequivocal findings concerning top-down modulation of similarity. In this online study, we manipulated top-down beliefs in a between-subject design, comparing human to robot imitation using novel multiframe stimuli in a third-person perspective. We predicted an increased imitative compatibility effect for the human than for the robotic hand, but no significant difference between these two agents in spatial compatibility. Our results (N=202) showed significant overall, spatial, and imitative compatibility effects for both, human and robotic hand stimuli. We found no significant difference in automatic imitation between the robot and human conditions neither in terms of reaction times, nor in error rates. Additionally, we found no significant difference between human and robotic hand stimuli in terms of perceived similarity and intentionality ratings, and no correlation with imitative and spatial compatibility effects. However, exploratory analysis showed that mood seems to correlate negatively with spatial compatibility. While online studies appear suitable for investigating automatic imitation per se, our results suggest that potential top-down modulation effects were not detectable in our online study.