AI in hiring — which efficiency costs you fairness
The stage you apply it to decides whether this is an efficiency win or a liability.
AI Tools Hub Editorial · published · updated
Safe to automate
- Structuring applications: listing experience, extracting stated qualifications
- First-line answers to candidate questions
- Interview scheduling
- Interview notes, with consent
- Drafting the job description
None of these decide an outcome. Organising information so a person can judge better does not raise the bias problem.
Do not automate
- Pass/fail decisions
- Aptitude scoring learned from past hires
- Assessment from facial expression or voice
- Filters that correlate with protected attributes
The second is the dangerous one: it reproduces the pattern of whoever you hired before, and a machine-produced number looks objective, which makes the bias harder to see than it was when humans did it.
Stay able to explain
"The system scored you low" is not an explanation. Record the human reason for each rejection. That requirement alone usually rules out automated decisions.
Tell candidates
Decide whether to disclose AI use, and obtain explicit consent for recording interviews. Several jurisdictions are regulating AI in hiring, and a candidate's location may bring those rules into play.
The best use is the job description
Turning requirements into prose and removing phrasing that discourages particular groups is safe, fairness-neutral, and affects applicant volume — the highest-return use in the process.
FAQ
What if volume is unmanageable?
Mechanical filtering on stated requirements — three years of experience, a specific certification — is fine. Filtering on inferred aptitude is not.