Can I use Ai to screen job applicants?

This is HR guidance, not legal advice. Hiring rules vary by state and city and they keep changing, so run any screening change past your employment counsel before you use it on live applicants.

What do screening tools actually do?

Most screening tools do three simple things. They sort resumes by keywords, rank applicants against a job description, or flag applications that meet rules you set, like a required certification or years of experience.

That is useful. It is not judgment.

A tool can match a resume to a posting. It cannot judge teamwork, learning, or how someone handles pressure. Treat it as a first pass, not a final answer. The moment a score becomes the decision, you have handed a people decision to a pattern matcher.

It helps most when volume is high and the first criteria are clear. If you get 200 applications for one role, a tool can help you find the 40 that meet the stated requirements, so your team spends its time on real conversations instead of sorting.

It also helps with consistency, but only if the starting criteria are fair and job related. Every applicant gets measured against the same first bar. If that bar is too narrow, the tool will repeat that narrowness for every single applicant, faster and more confidently than a human would.

Where does screening go wrong?

Here is the part vendors skip. The most common problem is overtrust.

A team turns a tool on and starts treating a score as a decision. Good candidates get cut for reasons no one can explain, and no one notices until a strong hire slips away or an applicant asks a fair question the team cannot answer.

The second problem is hidden criteria. Some tools weigh factors that have little to do with the job, or they repeat patterns from past hires that reflect past bias rather than future performance. If you cannot explain, in plain language, why an applicant was ranked low, you should not use that ranking to cut people.

The third problem is silence. Applicants left guessing after an automated screen cannot raise mistakes. That hurts your reputation, and it tells good people exactly how your company treats people when no one is watching.

What is the one rule that keeps screening fair?

Put a person between the tool and the rejection. That is the whole rule.

First, a human reviews the criteria before the tool runs and asks whether each one is truly required for the job. Second, a human reviews a sample of applications the tool ranked low, not just the ones it ranked high, to catch people wrongly cut. Third, a human makes the decision to move someone forward or close out an application, and can override the tool.

Keep a short record of the tool recommendation, the human decision, and why. If you cannot show your work, you do not have a process. You have a guess with software on top of it.

What five questions should I answer first?

Before you use any screening tool, answer these five questions in writing:

1. What does the tool measure, and is each factor related to this job?
2. Can the vendor explain, in plain language, how scores are produced?
3. Who reviews the results, and who can override them?
4. How does an applicant ask for a human review or raise a concern?
5. What do you tell applicants about how Ai is used in your process?

If you cannot answer one of those questions, pause. That gap is the risk. Fix the process first, then turn the tool on. Audit work on hiring tools runs alongside your counsel, not instead of it.

What do the state rules say?

There is no single federal Ai hiring law. Title VII, the ADA, and the ADEA apply whether a person or an algorithm decides. The EEOC removed its Ai employment guidance from its website in early 2025.

State rules: California, automated decision system rules under FEHA, effective October 1, 2025. Illinois, HB 3773, effective January 1, 2026. Texas, TRAIGA, effective January 1, 2026. New York City, Local Law 144, enforced since July 5, 2023. Colorado, replacement law aimed at January 1, 2027. The detail lives on the state pages: California, Illinois, Texas, New York City, Colorado.

What to do next

Start small. Pick one role with clear requirements and high volume. Use Ai only for the first sort, against criteria you already use by hand, and review the low ranked group yourself for the first few rounds.

If it saves time without cutting qualified people you would have kept, it is useful. If it keeps cutting people your team wants to talk to, fix the criteria or the tool before expanding. Do not scale a broken filter just because it is fast.

Read the Why Ai page for the thinking behind the sequence. Getting hiring tools reviewed is part of an Ai risk and compliance audit.