AI tutors are becoming the new faces of digital classrooms, but these faces may determine who is trusted and who is ignored. Anthropomorphic avatars designed to increase engagement, may also activate social heuristics that reproduce offline patterns of bias. Yet little empirical research has examined how these identity cues shape epistemic trust and enacted reliance in AI-mediated learning. This study addresses this gap by systematically isolating their effects on learners’ evaluative judgments and behavioral uptake in diverse learning contexts. Two experiments examined how avatar race, gender, and age shape trust in AI-mediated education. Study 1 (N = 102) used a within-subjects laboratory design with tightly controlled avatars; Study 2 (N = 294) adopted a between-subjects online design across varied instructional domains. Four outcomes, credibility, warmth, competence, and willingness to act, were measured alongside a behavioral index of epistemic uptake, defined as incorporating AI guidance into learners’ own work. Across both studies, social identity cues strongly influenced evaluations and behavior. White avatars, and in STEM contexts Asian male avatars, were rated more credible and competent. Older Black female avatars faced compounded penalties across all measures. These perceptual hierarchies shaped behavior: participants were more likely to adopt guidance from avatars aligned with their racial ingroup or stereotypical expectations of expertise. Domain moderated these effects, with STEM and procedural tasks amplifying bias, while reflective and interpersonal tasks attenuated it. Our findings show that trust in AI tutors is as much about the messenger as the message. Avatar identity guides not only learners’ perceptions but also their actions, underscoring the need for equity-aware avatar design and trust calibration mechanisms to ensure AI instruction empowers all learners. Building on these insights, we outline practical design recommendations to ensure that AI instruction amplifies learning rather than entrenches bias. • Avatar cues reshape how students trust and act on AI-driven guidance. • Social identity systematically shapes perceived expertise and reliance on AI tutors. • Bias in digital classrooms intensifies in STEM and procedural learning contexts. • Intersecting race, gender, and age cues produce compounded credibility penalties. • Equitable AI tutoring requires mechanisms that support appropriate trust beyond surface cues.
Anthis et al. (Fri,) studied this question.