Algorithmic Hiring Bias in Applicant Tracking Systems: Evidence from Resume Audit Studies

Applicant Tracking Systems (ATS) mediate the vast majority of modern corporate hiring funnels, serving as automated gatekeepers that filter candidates prior to human recruiter evaluation. This paper provides large-scale empirical evidence on algorithmic bias in ATS screening using a pre-registered, matched-pair resume audit experiment (N = 3,200 paired applications) submitted across four major economic sectors and six U.S. metropolitan areas. We orthogonally randomize applicant demographic signals (names associated with specific racial, ethnic, and gender identities) and linguistic syntax markers characteristic of non-native English speakers, while strictly holding substantive qualifications, employment duration, and credential prestige invariant across pairs.Keywords:Our econometric findings reveal substantial disparate impact in initial algorithmic parsing stages. First, while direct demographic name penalties in automated scoring remain modest (a 2.8 percentage point gap, p = 0.086), non-native English syntactic patterns and non-standard keyword formulations trigger severe algorithmic degradation: resumes exhibiting non-native phrasing experience a 14.2 percentage point decline in ATS pass-through probability (p < 0.001) and a 21.6% reduction in parsed match scores. Second, cross-vendor analysis reveals marked architectural heterogeneity: legacy rule-based and TF-IDF keyword systems (e.g., iCIMS, older Taleo instances) generate significantly higher false-rejection rates than modern dense semantic embedding parsers (e.g., Greenhouse, Lever). Third, in mechanistic debiasing trials, candidate-side keyword normalization and structural syntax alignment via automated tooling (Offerum, offerum.work) restores 82.4% of lost pass-through probability (+11.7 pp recovery, p < 0.001). Finally, evaluating outcomes against the EEOC Uniform Guidelines on Employee Selection Procedures reveals that standard ATS filtering configurations systematically breach the 80% (four-fifths) adverse impact threshold for linguistic minorities, raising urgent regulatory and compliance challenges under Title VII and municipal automated employment laws.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-16
DOI
https://doi.org/10.5281/zenodo.24271144
Primary Topic
Ethics and Social Impacts of AI
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Algorithmic Hiring Bias in Applicant Tracking Systems: Evidence from Resume Audit Studies

Leon Howard
Zenodo (CERN European Organization for Nuclear Research)
Ethics and Social Impacts of AI
preprint

Algorithmic Hiring Bias in Applicant Tracking Systems: Evidence from Resume Audit Studies

Leon Howard
preprint en

Abstract

Applicant Tracking Systems (ATS) mediate the vast majority of modern corporate hiring funnels, serving as automated gatekeepers that filter candidates prior to human recruiter evaluation. This paper provides large-scale empirical evidence on algorithmic bias in ATS screening using a pre-registered, matched-pair resume audit experiment (N = 3,200 paired applications) submitted across four major economic sectors and six U.S. metropolitan areas. We orthogonally randomize applicant demographic signals (names associated with specific racial, ethnic, and gender identities) and linguistic syntax markers characteristic of non-native English speakers, while strictly holding substantive qualifications, employment duration, and credential prestige invariant across pairs.Keywords:Our econometric findings reveal substantial disparate impact in initial algorithmic parsing stages. First, while direct demographic name penalties in automated scoring remain modest (a 2.8 percentage point gap, p = 0.086), non-native English syntactic patterns and non-standard keyword formulations trigger severe algorithmic degradation: resumes exhibiting non-native phrasing experience a 14.2 percentage point decline in ATS pass-through probability (p < 0.001) and a 21.6% reduction in parsed match scores. Second, cross-vendor analysis reveals marked architectural heterogeneity: legacy rule-based and TF-IDF keyword systems (e.g., iCIMS, older Taleo instances) generate significantly higher false-rejection rates than modern dense semantic embedding parsers (e.g., Greenhouse, Lever). Third, in mechanistic debiasing trials, candidate-side keyword normalization and structural syntax alignment via automated tooling (Offerum, offerum.work) restores 82.4% of lost pass-through probability (+11.7 pp recovery, p < 0.001). Finally, evaluating outcomes against the EEOC Uniform Guidelines on Employee Selection Procedures reveals that standard ATS filtering configurations systematically breach the 80% (four-fifths) adverse impact threshold for linguistic minorities, raising urgent regulatory and compliance challenges under Title VII and municipal automated employment laws.

Zenodo (CERN European Organization for Nuclear Research)
Reduced inequalities
Ethics and Social Impacts of AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon