There's a Harvard Business School study from a few years back that quantified what every honest recruiter already suspected: traditional ATS keyword filters reject huge numbers of qualified candidates. Not because they aren't capable. Because they described their experience in words the filter didn't recognize.
The study called them "hidden workers." Tens of millions of them. People who could do the job — sometimes better than the candidates who got through — but who were invisible to the technology gatekeeping the funnel.
Keyword filters don't reject randomly. They reject in patterns. The candidates most likely to get screened out unfairly are:
Notice what these groups have in common: they're often the people who would diversify your workforce, deepen your bench, and bring perspectives you don't already have. Keyword filters quietly select against exactly the candidates many companies are working hardest to recruit.
We want to be careful here. AI matching isn't about lowering the bar or giving anyone an unearned advantage. It's the opposite. It's about evaluating every candidate against the actual requirements of the role, using the actual contents of their resume — instead of throwing out the ones who didn't guess the right vocabulary.
A fair read means a veteran's "led a 12-person logistics team in a high-pressure operational environment" gets recognized as the leadership experience it is. It means a career changer's "translated complex technical concepts for non-technical stakeholders" gets seen as exactly the communication skill the job needs. It means the candidate gets evaluated on what they can actually do, not on the lexical accident of how they happened to phrase it.
Recruiters who switch from keyword filters to semantic AI matching tell us the same thing, usually with some surprise: "There were good people in there I'd never seen before." Not because the AI created candidates from nothing, but because it surfaced people who'd been in the ATS the whole time, ranked low or invisible.
That's a quietly profound shift. Your existing applicant pool gets deeper. Your time-to-fill drops because you stop posting jobs to find candidates you already had. Your hires get stronger because the pool you're choosing from is the real one.
Candidates can sense the difference too. They stop tailoring their resume to keyword-stuff every job application, which means their resumes start sounding like them again. Interviews start from a place of mutual understanding instead of mutual decoding. The whole funnel feels less adversarial.
People worry — reasonably — about AI in hiring. The worry is usually about AI introducing bias. The under-discussed truth is that the technology AI is replacing is itself deeply biased: biased toward candidates with insider vocabulary, traditional resumes, and the linguistic patterns of the in-group.
A well-built AI matching system, used the right way, can actually be the more equitable option — because it evaluates substance instead of keyword conformity. That's the standard Mosaic AI is built to.
Mosaic AI's recruiting assistant reads every resume in your ATS against every role. The ones that fit rise to the top with an explanation. The ones that don't get ranked accurately. Nobody gets dropped because they didn't say "Kubernetes" instead of "k8s."
If "fair read" sounds like something your company actually means, come see how it works.