Even Better than the Real Thing: How Imperfection Shapes Trust and Engagement with Digital Humans

This study challenges the assumption that more realism in digital humans always leads to greater trust and engagement. Using eye-tracking and post-exposure surveys, we compared viewer responses to three video presenters: a highly realistic digital human, a real human, and an imperfect altered human, represented by a real presenter altered to have unblinking eye contact. While participants rarely noticed visual imperfections consciously, the human with subtle flaws led to significantly greater trust and willingness to pay. The imperfect video outperformed the fully realistic, unaltered human video, suggesting that perfect realism may not always be best. These findings offer important implications for the design of AI-driven digital humans, highlighting that strategic imperfection can enhance authenticity, trust, and engagement in customer interactions. Moreover, the results contribute new empirical insights into the Uncanny Valley theory, suggesting that user affinity and trust may peak not at perfect realism, but can peak at a point just prior to the full realism.