A new audit from AI recruiting platform Pin, covering more than 37,000 recruiter sourcing searches across 33,000-plus open jobs between January 2024 and May 2026, found that the biggest source of candidate invisibility in AI-driven hiring happens before any AI model ever ranks a resume. The study found 70.7% of searches applied employer-prestige filters, 51.9% of experience-gated searches required five or more years, and 96% of tenure-floor filters defaulted to exactly twelve months regardless of the role. Nearly a third of searches, 31.3%, stacked multiple bias-prone filters at once. Pin’s own framing of the finding: “Bias does not start at the ATS. Bias starts at the sourcing screen.”
The distinction matters because most AI-hiring scrutiny to date has focused on the ranking algorithm itself, including a widely cited 2024 academic study the report cites, which found language-model resume rankers preferred white-associated names 85.1% of the time versus 8.6% for Black-associated names. Pin’s data suggests that even a perfectly unbiased ranking model would still produce a skewed candidate pool, because the recruiter-set filters applied before ranking already exclude most qualified candidates who lack a “prestige” employer, round-number tenure, or arbitrary years-of-experience minimums.
The original insight for HR technology buyers: fixing AI hiring bias by auditing the model alone misses most of the problem, since the filters recruiters configure manually, often defaults left untouched inside the ATS, do the actual excluding before the AI ever sees a candidate. Talent acquisition leaders evaluating AI sourcing tools should ask vendors for filter-level usage data, not just ranking-model bias audits, and should treat a default twelve-month tenure floor or blanket prestige-employer filter as a policy decision that needs sign-off, not a harmless default setting.
Related: AI Cheating Is Turning Interviews Into a Guessing Game and Managers Are the Weak Link in AI Rollouts.
Source: Pin
