Should You Use AI to Screen Candidates? A Small Business Owner's Guide

A guy I worked with last year had 90 applications for one warehouse role. He runs a small distribution business, 22 employees. He told me he ran every resume through an AI screener his payroll software was pushing, and it handed him a "top 8." He hired one of them. Six weeks later that hire was gone, and the candidate he'd really wanted, the one with the gap on her resume because she'd been caring for a sick parent, never made the cut. The software ranked her 61st.

That story is the whole article. AI can save you real time. It can also throw away the person you should have hired and put your name on a discrimination claim you didn't see coming.

In my 25 years training small-business owners to hire, I've watched a lot of "miracle" tools come and go. AI screening is the newest one, and it's the first one I'd actually tell you to consider. Not embrace. Consider. There's a difference, and getting that difference right is the point.

What these tools actually do today

Let's be clear about what you're buying, because the marketing makes it sound smarter than it is.

Most AI hiring tools you'll run into as a small business fall into three buckets. The first is resume screening and ranking. You dump in applications, the tool scores them against the job description and hands you a sorted list. The second is scheduling and chat. A bot answers basic questions, collects info, and books interviews so you're not playing email tag with 40 people. The third, and the one I want you to be most careful with, is AI video and voice analysis. These tools record a candidate answering questions, then "analyze" their word choice, tone, even facial expressions, and spit out a score on things like "conscientiousness" or "culture fit."

The first two are tools. The third is a tool pretending to be a judgment, and that's where owners get burned.

Where AI genuinely helps a small shop

I'm not anti-AI. When you're running the business and hiring on the side, the top of the funnel is brutal. Here's where these tools earn their keep.

Speed and sorting. If you've got 90 applicants and 90 minutes, a screener that flags the people who clearly don't meet your hard requirements (the license, the years, the legal must-haves) is genuinely useful. It's doing what you'd do with a highlighter, just faster.

Scheduling. The back-and-forth of booking interviews is pure time you'll never get back. A scheduling bot that lets candidates pick a slot is the single lowest-risk way to use AI in hiring. Nobody gets discriminated against by a calendar.

Drafting the job post. This one's underrated. A good job description does half the screening for you, because the right people apply and the wrong people don't. AI is a solid first-draft writer here. I still want you editing it hard, but it beats a blank page. If you want to see what a strong post looks like, I walk through it in how to write a job description.

Organizing applicants. Keeping track of who applied, who you called, who you're waiting on. That's administrative work, and handing it to software is fine.

Notice the pattern. AI is great at the stuff that's about logistics and volume. It gets dangerous the second it starts making the call about who's worth your time as a human being.

Where AI hurts you

Here's the part the sales rep won't lead with.

It filters out your best non-traditional candidate. AI ranks on patterns from the past. The career-changer, the person with a gap, the one whose resume doesn't use the exact words the model likes, they get buried. Some of the best hires I've ever seen looked wrong on paper. A machine trained on "what good resumes look like" is, by design, going to penalize anyone who doesn't look like the last batch.

It gets gamed. Candidates know these tools exist. The savvy ones stuff their resume with the right keywords, sometimes in white text you can't even see, and they rocket to the top of your list. So the tool doesn't surface your best candidate. It surfaces your best resume-optimizer. Those are not the same person, and the optimizer is often the weaker hire.

It gives you false confidence. When software tells you someone scored 94, your brain relaxes. You stop doing the real work. That number feels like proof, and it isn't.

You lose the human read. Hiring well is about catching the things a resume can't tell you. How someone talks about their last boss. Whether their story holds up when you press on it. The pause before they answer a hard question. No AI catches that, and the more you lean on the score, the more you stop listening for it. That read is the entire reason the 5-minute phone screen works, and it's the part you can't outsource.

The legal exposure nobody mentions

This is where I get serious, because it's the part that can actually cost you money.

When you use an AI tool to screen candidates, you are legally on the hook for what it does. Not the vendor. You. The U.S. Equal Employment Opportunity Commission published technical assistance in May 2023 spelling this out. If your screening tool produces an "adverse impact," meaning it selects people of one race, sex, or other protected group at a meaningfully lower rate than another, you can be liable under Title VII even though a vendor built the tool. The EEOC's own document specifically names resume-screening software and chatbots as examples of tools this applies to. Relying on a vendor's claim that their tool is "bias-free" does not get you off the hook (see the EEOC's guidance).

There's an Americans with Disabilities Act angle too. The EEOC has separately warned that an AI tool can illegally screen out a qualified person with a disability, for example a video tool that penalizes a speech difference, or one that doesn't account for someone who needs a reasonable accommodation.

And if you hire anyone who lives in New York City, even for a remote role, you've got a specific law to know about. New York City's Local Law 144 took effect January 1, 2023, with enforcement starting July 5, 2023. It requires that any "automated employment decision tool" used to substantially help decide who gets hired be put through an independent bias audit within the prior year, that a summary of that audit be posted publicly, and that candidates get advance notice the tool is being used. Penalties run from $500 to $1,500 per violation, and each day can count. You can read the city's framework through the NYC Department of Consumer and Worker Protection rules.

Here's the plain-English version. The fancier and more "decision-making" the tool, the more legal weight lands on you. A scheduling bot is low risk. An AI that scores candidates and ranks them is the kind of thing that gets audited and litigated. Know which one you're using.

David's rule: AI is an assistant, never the decider

After all that, here's how I'd actually use AI to screen candidates if I were running your shop. The rule is simple. AI handles volume and logistics. You handle judgment.

1. Use AI to draft, not to decide. Let it write the first version of your job post and your screening questions. Then you edit. You know your business and your customers; the model doesn't.

2. Let it sort on hard requirements only. Use a screener to filter for the non-negotiables: the required license, the legal eligibility, the minimum years. Do not let it score "fit," "personality," or "potential." Those are your job.

3. Never let it shrink your pool to a final list. Have it flag the obvious nos, then you personally eyeball everyone who cleared the bar. That's where you catch the great candidate with the ugly resume.

4. Keep the human steps human. The phone screen, the interview, the reference checks. No bot does these for you. The questions that actually predict whether someone will perform are the ones you ask live and follow up on, which is the whole point of interview questions that predict performance and a real reference check conversation.

5. If you use a scoring or video tool, do your homework. Ask the vendor for their bias audit. If you hire anyone in NYC, confirm you're meeting Local Law 144. And if a tool can't show you how it scores people, don't use it.

6. Stay skeptical of the number. When the software loves someone, ask why. When it buries someone, ask why. Treat every score as an opinion to check, not a verdict to follow.

The thread that ties it together

I teach a 10-phase hiring system, and people sometimes ask which phases AI can take over. The honest answer is none of them completely. AI can assist the early phases, the writing, the sorting, the scheduling, the organizing. It can give you back hours you don't have. What it can't do is the part that actually separates a good hire from an expensive mistake, which is your judgment applied to a real human being.

This limitation won't be fixed by next year's tools. It comes from the nature of the thing. Hiring is a bet on a person, and a person is exactly what the machine can't see.

Use AI for what it's good at. Keep your hands on the wheel for everything that matters. If you want the full method for the parts that matter, it's all in The Naked Interview.

AI can sort the pile, but the real decision lives in the conversations. Grab our phone screen script for the first human step no bot can do for you.