Hiring by algorithm, audited
Recruitment, job matching and hiring discrimination appear on 109 recent papers across the economics, management and AI journals, with 32 in the last quarter against about 25 per quarter before; 32 of them carry a bias, fairness, algorithm or AI keyword. Screening tools now read millions of applications, and New York requires bias audits of them. What do the audits find, and do they change outcomes?
What prompted this#
Since October 2025, 109 papers carry a hiring, recruitment, résumé, applicant, candidate-screening, job-matching or talent-acquisition keyword (54 in economics journals, 24 in management, 13 in finance, 8 in AI): "recruitment" on 8, "job matching" on 8, "hiring discrimination" on 6, "hiring decisions" on 6, "talent acquisition" on 5. Thirty-two of the 109 also carry a bias, fairness, discrimination, algorithm, language-model, AI, automated or machine-learning keyword. The quarter to September 2026 had 32 such papers against about 25 per quarter earlier.
Questions#
- In the bias audits published under New York's law, what disparities were found, and did employers change or drop the tools that showed them?
- When language models rank résumés, do they discriminate by name, gap years or address in correspondence tests, and does prompting the model to be fair fix it?
- Do algorithmic screens change who applies, as applicants learn to write for the machine?
Data that could help#
Published bias audit reports, correspondence experiments with language-model screeners, applicant tracking data from employers, and job posting text over time.
Next step#
Collect the bias audit reports published under the New York law and code the disparities found and the employers' subsequent tool use.