AI Hiring Tools Are Under Investigation. Here's What That Means for You.

Here's what's actually happening behind the scenes. About 97.8% of large companies use automated screening software to filter resumes before a human ever sees them. The average corporate job posting gets 250 applications. Seventy-five percent of those resumes get filtered out by software that does something incredibly simple: keyword matching. That's it. It's basically ctrl+F. If the exact words from the job description appear on your resume, you pass. If they don't, you're out. It doesn't understand context. It doesn't evaluate experience. It doesn't look at what you actually did. It searches for strings of characters and either finds them or doesn't.

The laws exist because this process is completely opaque. You submit your resume into a black box and get silence back. No explanation. No visibility into what matched or what didn't. Just nothing. Forty-four percent of job seekers report being ghosted entirely. Seventy-two percent say the process damages their mental health. And about 30% of job postings aren't even real jobs, they're ghost postings companies leave up to look like they're growing. So you're not just fighting a vocabulary test, you're fighting a system designed to be inscrutable.

The bias investigations aren't about malicious intent. They're about the mechanical reality of how these tools work. If a job description says 'project management' and your resume says 'program management,' the software doesn't know those might be the same thing. If the posting asks for 'customer success' experience and you wrote 'account management,' you're filtered out. Not because you're unqualified. Because you used different words. That's not intelligence. That's a vocabulary matching game where nobody told you the answer key.

When I started mapping this problem, I didn't expect the scale. I built a skill graph to understand what screening software actually looks for versus what candidates actually write. The data was uncomfortable. There are roughly 287,000 distinct professional skills across industries. But here's the thing: those skills have about 891,000 different ways people refer to them. Different companies use different terms. Different industries have different conventions. Different decades have different vocabulary for the same work.

The average job posting contains about 53 distinct skills. Which means every time you apply, you're playing a matching game with 53 targets, each of which might be written a dozen different ways, and the software is doing exact string matching. If you say 'Solutions Engineer' and they wrote 'Sales Engineer,' you're out, even though those are the same role at different companies. If you say 'led cross-functional initiatives' and they want 'project coordination,' the software doesn't connect those dots. It just sees two different phrases and moves on.

This isn't about being bad at resumes. It's about a system that treats language like a database query instead of communication. The problem is diagnosable. It's not vibes, it's not luck, it's not some mysterious quality you're missing. It's measurable vocabulary mismatch in a system that can't understand synonyms.

The audits happening in NYC, Illinois, and Colorado exist because regulators finally realized these tools make consequential decisions with zero transparency. But enforcement is hard when the tools themselves are proprietary black boxes. Even when companies submit bias audits, there's no standard for what 'fair' looks like when the underlying mechanism is just word matching.