White Applicants: Generally served as the most-favored group (the baseline benchmark). The AI system typically recommended candidates from this group at the highest rates.
Black Applicants: Faced the highest level of systemic bias. 26% of Black applicants applied to roles where the AI tool actively discriminated against their demographic. The study noted job-specific pigeonholing, where Black candidates might be frequently recommended for labor-focused roles (like warehouse jobs) but rarely for corporate roles (like finance).
Asian Applicants: Also faced notable discrimination, with 15% of Asian applicants applying to positions where the AI system flagged adverse impact against their group.
Gemini TLDR by ethnic group: