<![CDATA[The rapid deployment of generative AI tutors in multicultural classrooms promises personalized learning but risks exacerbating inequity. These tools, often trained on culturally-biased, “WEIRD” (Western, Educated, Industrialized, Rich, Democratic) data, may not account for the diverse linguistic and contextual needs of all students, potentially reinforcing a dominant “algorithmic monoculturalism.” This study investigates the impact of culturally-misaligned AI tutors on educational equity. It aims to (1) audit the cultural responsiveness of commercial AI tutors, (2) quantitatively measure their differential impact on student belonging and engagement, and (3) qualitatively explore the lived experiences of marginalized students. A sequential explanatory mixed-methods design was employed. Phase 1 involved a computational content audit (AICR Rubric). Phase 2 was a quasi-experiment (N=180) with pre/post-tests measuring belonging and engagement. Phase 3 used phenomenological interviews (N=30) with marginalized students. The audit confirmed significant cultural misalignment in AI tutors (Tutor A M=1.5/5.0). The quasi-experiment revealed a statistically significant decline in academic belonging (p < .001) and engagement for the marginalized group, with no negative effect on the dominant group. Qualitative themes of “Perceived Algorithmic Judgment” and “Cognitive Friction” explained this iatrogenic effect. Standard “one-size-fits-all” AI tutors can actively cause harm, creating new equity gaps by failing to address cultural context. The study provides a novel framework for equity-focused AI assessment and calls for a design paradigm shift towards culturally sustaining technology.]]>
