In interpersonal interactions, linguistic information is complemented by non-linguistic information originating largely from facial expressions. The study of online face-to-face social interaction thus entails investigating the multimodal simultaneous processing of oral and visual percepts. Moreover, gaze in and of itself functions as a powerful communicative channel. In this respect, gaze should not be examined as a purely perceptive process but also as an active social performance. We designed a task involving multimodal deciphering of social information based on virtual characters, embedded in naturalistic backgrounds, who directly address the participant with non-literal speech and meaningful facial expressions. Eighteen adult participants were to interpret an equivocal sentence which could be disambiguated by examining the emotional expressions of the character speaking to them face-to-face. To examine self-control and self-awareness of gaze in this context, visual feedback is provided to the participant by a real-time gaze-contingent viewing window centered on the focal point, while the rest of the display is blurred. Eye-tracking data showed that the viewing window induced changes in gaze behaviour, notably longer visual fixations. Notwithstanding, only half the participants ascribed the window displacements to their eye movements. These results highlight the dissociation between non volitional gaze adaptation and self-ascription of agency. Such dissociation provides support for a two-step account of the sense of agency composed of pre-noetic monitoring mechanisms and reflexive processes. We comment upon these results, which illustrate the relevance of our method for studying online social cognition, especially concerning Autism Spectrum Disorders (ASD) where poor pragmatic understanding of oral speech are considered linked to visual peculiarities that impede face exploration.