Every hiring manager we talk to is drowning in applications and tempted by the same promise: let software read the pile and hand you the top ten. Some of what AI does in hiring is genuinely useful and low risk. But hiring is one of the few places where an efficiency gain can turn into a discrimination claim, and the tool will not warn you when you cross the line.
So let us be clear before we go further. This article is general information, not legal advice. We are an IT services company, not a law firm, and nothing here substitutes for an employment attorney reviewing what you do, in the states where you hire. Have that conversation first. There is still a useful way to think about this: is the tool helping a human work faster, or judging a person?
What The EEOC Already Says About Screening Tools
Many owners assume AI hiring is an unregulated frontier. In fact the rules about screening people were written broadly, and screening is screening regardless of what does it.
The EEOC’s published Questions and Answers on the Uniform Guidelines on Employee Selection Procedures describes selection procedures as including “application forms, interviews, performance tests, paper and pencil tests, performance in training programs or probationary periods, and any other procedures used to make an employment decision.” That last phrase is the one to sit with. It is deliberately open ended.
The EEOC’s fact sheet on Employment Tests and Selection Procedures names three federal statutes that apply to employment testing: Title VII of the Civil Rights Act of 1964, the Americans with Disabilities Act, and the Age Discrimination in Employment Act. It distinguishes disparate treatment, meaning intentional differences in how people are treated, from disparate impact, where a neutral procedure disproportionately excludes people on a protected basis without job relatedness and business necessity.
One more point matters a great deal. The Uniform Guidelines Q&A states that where a selection procedure is administered on an employer’s behalf by an outside party, “the employer remains responsible.”
Where AI In Hiring Is Genuinely Low Risk
Plenty of hiring work has nothing to do with evaluating a candidate. This is where AI earns its keep.
- Writing the job post. Turning a manager’s rough bullets into a clear posting. Ask it to flag jargon and unnecessary requirements that narrow your pool without a business reason. A human approves the final requirements.
- Scheduling and logistics. Interview times, confirmations, chasing no responses. Nobody’s candidacy is being judged. This is calendar work.
- Drafting interview questions. A consistent set of job related questions tied to actual duties, plus a scoring rubric. Asking everyone the same questions is better practice anyway.
- Taking notes. Capturing what was said so interviewers can pay attention instead of typing. Tell candidates it is happening, ask permission, keep the notes factual.
- Answering candidate questions. Benefits, location, schedule, next steps. Grounded in your own published material, this is a service improvement, not a screen.
Notice the pattern. The AI produces material for a human. It never ranks, scores, or eliminates anybody.
Where The Risk Climbs Fast
The moment a tool produces an opinion about a candidate, the EEOC’s guidance on selection procedures becomes directly relevant, and you need your attorney involved.
- Ranking or scoring applicants. A tool that sorts your pool is making the first cut, whatever the marketing calls it. Most people never look past the top of a ranked list, so the ranking is the decision.
- Automatic rejection. Knockout rules are what the concept of a selection procedure was written to cover. If nobody reviews the rejections, nobody notices a pattern in them.
- Video and audio analysis. Anything inferring traits from a face, a voice, or speech patterns. Consider how it interacts with disability, age, accent, and national origin. The EEOC’s fact sheet says the ADA requires reasonable accommodation “including in the administration of tests,” absent undue hardship.
- Games and personality assessments. The EEOC’s fact sheet already lists personality tests among employment tests and selection procedures. An AI layer does not change that.
- Proxies you did not intend. A model built on your past hires can learn patterns you never wrote down: a zip code, a school, a gap in employment. It will not tell you. You have to look.
The EEOC’s fact sheet advises employers to ensure selection procedures “are properly validated for the positions and purposes for which they are used,” and warns that tests should not be “adopted casually by managers who know little about these processes.” That last line describes how most AI hiring tools get bought.
You Have To Be Able To Explain The Decision
Here is the practical test we give clients. If a rejected candidate asks why, or someone official asks a year from now, can a human explain the decision in plain language tied to the job?
“She interviewed better on the technical questions and had directly relevant experience with the equipment we run” is an explanation. “The system scored him a 62” is not. It is a number with a story you cannot tell. If your process produces the second kind of answer, the process is the problem, regardless of whether the outcome was fair.
This is where Harrison’s framing about treating AI like a junior staff member earns its keep. You would not let a new hire in their first week make final hiring calls, and you would not accept “I just had a feeling” as their reasoning. You would let them organize the pile and flag things. Then you would decide, and be able to say why. The addition here: write the reasoning down at the time.
Records, Disclosure, And The Vendor Question
Three practical matters, all of which your attorney should confirm.
- Records. The EEOC’s published recordkeeping guidance states that employers keep all personnel or employment records for one year, and for terminated employees “for one year from the date of termination.” It adds that once a charge of discrimination is filed, relevant records must be kept “until the final disposition of the charge or any lawsuit based on the charge.” If a tool scored candidates, those scores are part of the hiring record.
- Disclosure. As a practical matter and not a legal one, tell candidates plainly when AI is involved and what it does. Rules at the state and local level vary and change, and we have not tried to summarize them here. Ask your attorney which apply where you hire, which may not be where you are headquartered.
- Vendors. Ask what the tool measures, how it was validated for roles like yours, what testing exists for adverse impact, what accommodations exist for candidates with disabilities, and what data you keep. Get it in writing. A vendor’s confidence is not your defense.
The Bottom Line
Use AI to write the posting, book the interviews, build a consistent question set, and take notes. Be deliberate about anything that ranks, scores, rejects, or infers, because the moment a tool forms an opinion about a person, you own that opinion. A human decides, explains that decision in job related terms, and writes it down at the time.
One more time, because it matters: this article is general information, not legal advice. Employment law is fact specific and varies by state and locality, and an employment attorney must confirm what applies to your business, your roles, and the places you hire. For the broader groundwork, we wrote how to actually prepare your team for AI without the hype.
On the technology side, we can help you work out what a hiring tool is actually doing under the hood, what data it keeps, and what to put to the vendor in writing before you sign. We serve small and mid sized businesses across Denton County and North Texas. Contact us today
Sources:
Comments are closed