To establish whether humans possess a rapid implicit mentalizing system enabling efficient social interaction, researchers have extensively investigated spontaneous visual perspective taking using paradigms such as the dot perspective task. Yet the validity of this task has been challenged by the submentalizing account, which attributes the observed self-consistency effects (altercentric interferences) to domain-general attentional orienting. We devised a novel mirror-reflection paradigm that equates visual information between participant and avatar while preserving the avatar's directional cue to isolate the contributions of directional attention versus visual content alignment. In two within-subjects experiments (n = 50 and 53), participants judged the number of targets visible either to themselves or to an avatar whose body orientation was consistent or inconsistent with target location. Crucially, mirrors ensured that the avatar always shared the participant's visual access; in control (blackboard) scenes, visual access could differ. Robust consistency effects emerged and, critically, the magnitude of the altercentric interferences was not modulated by the scene type. A blocked-perspective replication (Experiment 2) ruled out perspective-switching costs as an alternative explanation. These findings demonstrate that directional attentional orienting, rather than spontaneous perspective-taking, underlies performance in the dot-perspective task, providing compelling evidence for the submentalizing account and cautioning against the use of the task as a pure index of implicit theory of mind.