Gaze in the Machine: How Mechanical and AI-Based Eye Contact Correction Shape Communication Experience and Gaze Behaviour in Video Calls

Video calls prevent mutual eye contact because the camera and the partner’s image on screen are spatially separated. Teleprompters and AI-based gaze correction promise to restore it, but their effect on users is largely unknown. In a within-subjects study (N = 32), participants played a conversational guessing game with a partner in four conditions: face-to-face, a standard video call, a video call in which the partner used a teleprompter, and a video call with AI-based gaze correction of the partner’s video. We measured social presence , gaze-specific experience and camera awareness by questionnaire, and gaze behaviour by mobile eye tracking. All video conditions showed a large and robust loss of co-presence compared with face-to-face conversation (Cohen’s d z ≈ −1.5 to −1.8), together with heightened camera awareness. Neither intervention narrowed this gap; equivalence tests confirmed comparable co-presence and camera awareness across video conditions. AI-based correction made eye contact feel less natural than the standard video call (d z = −0.54), yet participants’ gaze behaviour remained equivalent. Gaze differences in the teleprompter condition 1 coincided with a smaller on-screen face and are attributable to framing. Restoring the appearance of eye contact thus does not restore what video calls lack. We discuss implications for evaluating and designing gaze technologies.