The Knockout Questions Nobody Audits

Most of the argument about AI bias in hiring is about the model: the resume scorer, the ranking algorithm, whatever a vendor calls its match score. That argument got more expensive in May 2025, when a federal judge let the age discrimination claim in Mobley v. Workday go forward as a nationwide collective action [1]. But in most of the pipelines I look at, the filter that rejects the most people isn't a model. It's a handful of yes/no knockout questions somebody typed into the ATS years ago, and nobody has read them since.

What a knockout question actually does

A knockout question is the blunt instrument on the application form. Do you have a Class A CDL? Are you available to work weekends? Do you have five or more years of experience in accounts payable? Answer wrong and the ATS moves you to rejected, usually with an automated email, sometimes with nothing. No recruiter ever opens the application.

They exist for decent reasons. If a role legally needs a license, there's no point screening someone who doesn't have one. The trouble is how they get written. A recruiter sets up a req in a hurry, copies the questions from a similar req, and adds one the hiring manager mentioned on the intake call. The req closes, the template lives on, and eighteen months later the same questions are filtering applicants for a role that has changed.

Harvard Business School's Hidden Workers study found that 88% of employers agreed their systems screened out qualified high-skilled candidates for not matching the exact criteria in the job description. For middle-skilled candidates it was 94% [2]. Knockout questions are the most literal version of that exact-match problem.

The questions people answer wrong

These questions are ambiguous far more often than whoever wrote them realizes. Take "Are you available to work weekends?" A nurse who can work every other weekend, which is what the unit actually schedules, has to pick yes or no. If they're honest, they pick no. Or "Do you have 3+ years of warehouse experience?" from someone with two years in a warehouse and four more on the loading dock of a grocery distributor. Whether that counts depends on the hiring manager, and the hiring manager never gets asked.

There's a reverse problem too. Plenty of candidates figured out a while ago that the safe answer is yes to everything, and they aren't a random sample. They're the people who have applied to a lot of jobs recently and know how these forms work. So the people who answer literally and truthfully get cut, and the people who learned to game it get through. I suspect most TA leaders would be uncomfortable if they saw which group their knockout rules are rewarding.

Why nobody audits them

When a team reviews its hiring for bias, the model gets the attention, because it's new and there's a vendor to call. We've written before about how bias gets back into AI hiring at the steps nobody watches. Knockout rules are the extreme case. Rejections from them show up in ATS reports as "did not meet minimum qualifications", which reads like a decision a person made. Nobody looks at those candidates, so nobody finds out whether the rule was right.

It's also cheap to check. Most ATSs can export disposition reasons. Pull every applicant rejected by an automated knockout for your five highest-volume roles last quarter, group them by the question that cut them, and read twenty applications from each group. In every audit like this I've seen, at least one question was cutting people who were obviously fine. It's the same pattern as a 27-minute application that mostly filters for patience: the form ends up deciding things nobody meant it to.

How Asendia AI handles this

The real fix for most knockout questions is to turn them back into questions, the kind a recruiter would ask on the phone and then follow up on. "Weekends?" becomes "Which weekends can you do, and is there anything that would get in the way?" Nobody could do that for 400 applicants a req, which is why the checkbox won.

Asendia is a voice-first AI recruiter. Every applicant gets a phone call, usually within hours of applying and at any hour, which matters when a lot of frontline applicants apply at 11pm after a shift. The call covers the same requirements your knockout questions were meant to check, but as a conversation. When the nurse says every other weekend, that gets written down, not rejected. When the warehouse candidate talks about the grocery distributor, their actual answer lands in your existing ATS next to the application with a structured summary, and a recruiter decides whether it counts.

Hard requirements can stay hard. If the license is a legal must, Asendia asks for it and flags anyone who doesn't have it. What changes is that the ambiguous questions stop being silent rejections. Most of the staffing agencies using Asendia push hundreds of applicants a week through a small team and had no way to talk to all of them before. Now they do, without adding recruiters, and the form can go back to the two or three things that really are yes or no.

Final Word

If you're worried about AI bias in hiring, and after Mobley you probably should be at least a little, audit the model. But check the checkboxes first. They're older and blunter, they reject more people than any model you've bought, and nobody has looked at them since the day they were typed in.

Pick one role this week. Open the application as a candidate and answer the knockout questions honestly, the way someone you'd happily hire would answer them. If the form rejects that person, you've found your first fix.

Ready to transform your hiring strategy? Schedule a Demo with our founders today!: https://asendia.ai/talk-to-founders