Research on level 1 visual perspective taking (L1-VPT) has been debating whether L1-VPT is an implicit socially rooted or rather a non-social process. Using online versions of the dot perspective task (Samson et al., 2010) we approached this question by comparing L1-VPT for robot vs. human avatars, presuming that social perspective taking due to mentalizing should occur more strongly for the human avatars, leading to altercentric and egocentric intrusions. The experimental design was manipulated within subjects: 2 (avatar: human, robot) x 2 (avatar perspective: consistent vs. inconsistent) x 2 (task: avatar perspective vs. self-perspective). Experiment 1 (N = 121) used a male, Experiment 2 (N = 117) a female human avatar. While reaction time measures in Experiment 1 confirmed our prediction of stronger altercentric intrusions for human than for robot avatars, in Experiment 2 and for error rate based intrusions, we found no significant differences between human and robot avatars. In addition, in both experiments and for both measures significant perspective x consistency interactions were found, the consistency effect being stronger for the other perspective than for the self-perspective. This suggests an egocentric bias with participants’ successful shielding against altercentric intrusions and a generally low level of implicit mentalizing. Yet, still in line with the mentalizing approach and to be further investigated, an explorative analysis of the combined data showed subtle effects of gender congruency: When the participants’ biological and the sex of the human avatar were congruent, the levels of error-rate-based altercentric intrusions were enhanced.