Automated Recruiting Software: 7 Checks Before You Buy

Automated recruiting software takes over seven decisions you stay answerable for, and a rule governs each one.

Automated recruiting software decides which applicants advance and which don't, and the employer buying it owns those decisions. Recruitment automation software acts alone at seven points in a pipeline, each with a rule attached. Those rules come from federal selection-procedure law and from state and city statutes, not from vendor documentation.

1. Automated resume screening and knockout filters

A knockout filter drops an application automatically when it misses a stated condition. The Uniform Guidelines on Employee Selection Procedures (UGESP) apply to all selection procedures used to make employment decisions, including review of experience or education from application forms, so the filter is one. The agencies generally treat a selection rate below four-fifths (80%) of the highest group's rate as a substantially different rate of selection, a rule of thumb that isn't a legal definition.

List every knockout before the requisition goes live, covering experience, degree, location, and work authorization, and ask whether you'd defend each as job related. Decide whether a work authorization knockout ends an application or routes it to a person, and learn the H-1B sponsorship timeline before you set it. Require an editable rule list, rates for the filter's own output rather than final hires, and a way to switch one rule off without rebuilding the requisition.

Skip the filter where the pool is small, since the rates are then too small to read. Ask the vendor for each rule's pass rate, by group, on one requisition.

2. Bias-audit obligations in the places you hire

New York City bars employers and employment agencies from using an automated employment decision tool (AEDT) unless it has had a bias audit within one year of use, the audit information is public, and notices have gone out. Under the DCWP rules, a bias audit is an independent auditor's evaluation that must at minimum calculate selection or scoring rates and impact ratios across sex, race and ethnicity, and intersectional categories.

Before switching the tool on in 2026, check which of your hiring locations set rules of their own. Require an audit dated within the last year on a public page, impact ratios across all three category sets rather than one overall figure, and annual renewal.

Skip this only if you hire nowhere with an audit rule. Ask the vendor for the date of its latest audit either way.

3. Candidate notice and consent for AI-assisted evaluation

Illinois requires an employer using artificial intelligence (AI) to analyze applicant-submitted video interviews for Illinois-based positions to notify the applicant beforehand, explain how the AI works and what general types of characteristics it uses, and obtain consent, and it bars evaluating anyone who hasn't consented. Under the Illinois Video Interview Act, the employer must delete those interviews within 30 days of a request and instruct anyone holding copies to delete them too, backups included.

New York City separately requires notice 10 business days before such a tool is used. Put that notice in the invitation email and the job posting rather than behind a click-through, because the obligation is yours and it has to land first. Require notice text you control, a consent record tied to the applicant, and a route for someone who declines.

Never skip this where it applies. Ask what the candidate sees, when, and whether you can edit it.

4. Proxy data in automated matching and ranking

Illinois makes it a civil rights violation for an employer to use artificial intelligence that has the effect of subjecting employees to discrimination on the basis of protected classes, or to use zip codes as a proxy for protected classes. The change, in Public Act 103-0804, took effect January 1, 2026. The prohibition turns on the effect of the tool, so ask what the ranking produces, not only what it was designed to do.

Interrogate what the ranking model reads, starting with zip code, school, commute distance, and named former employer. Require a list of the model's input fields, the ability to exclude one from ranking yourself, and an explanation of any geographic input that isn't a stated job requirement.

Skip this if the tool only stores applications, because the question then belongs to whoever ranks. Ask which input fields are geographic, and whether you can switch those off.

5. The record the automation leaves behind

Set the retention period for the automation's records before you switch it on. Equal Employment Opportunity Commission (EEOC) regulations require employers to keep all personnel or employment records for one year, and for one year from the termination date where an employee is involuntarily terminated, under the EEOC recordkeeping requirements.

Decide which system holds the file of record before the automation starts writing to it, and require that an automated rejection is logged there with the rule that caused it, not just in a vendor dashboard. E-Verify, the federal electronic employment eligibility system, is a separate process, and enrolling in E-Verify covers the employer side of it. Require export of a complete candidate file without a support ticket, and retention that outlives the contract.

Don't skip the record, though your existing system of record may be able to take the log. Ask what you get if you leave the vendor: the decision log, in what format, and does it name the rule?

6. Validation evidence, and who owns it

A test vendor's documentation on validity may be helpful, but the employer is still responsible for ensuring its tests are valid under the Uniform Guidelines. That's EEOC's position on tests, and New York City's DCWP FAQ takes the same position on bias audits, putting responsibility on employers and employment agencies, not the vendor that built the tool.

Read validity documentation as evidence you're collecting for yourself rather than cover you're buying. Require written validity evidence for the procedure and for job families like yours, a named internal owner who isn't the manager who chose the tool, and refreshes as the job changes.

Skip this if nothing in the tool scores, ranks, or excludes anyone, since scheduling needs no validity evidence. Ask which vendor documents you'd rely on to justify the screen, and for which job families.

7. The less-discriminatory alternative review

Where a selection procedure screens out a protected group, the employer should determine whether an equally effective alternative with less adverse impact exists, and adopt it if one does. The EEOC guidance and the Uniform Guidelines agree: where two procedures have substantially equal validity, use the one demonstrated to have the lesser adverse impact.

Put a standing review in the calendar, on a fixed schedule, comparing each screen's rates against an alternative, and treat switching as normal rather than an escalation. Require rates per screen refreshed often enough to compare two options honestly, the ability to run an alternative in parallel, and a written note of what was chosen, filed with the validity evidence.

Skip this only where the screen has no adverse impact to reduce, which you know only because you measured it. Ask whether you can run two configurations side by side on one requisition.

Check what your automated recruiting software rejects before you renew

Write down every step where the software advances or rejects a candidate without a person reading the file, and what evidence you hold for each. A second question falls out of that: is the constraint the screening, or the pool being screened? All seven checkpoints defend a filter applied to the pool you have, and none makes that pool larger.

Sourcing against job-related criteria with first-party contact details builds a shortlist out of the market rather than out of whoever applied, which is the gap these checkpoints leave open. Bring the recruiting automation question to your next renewal in those terms.