Are ATS systems filtering out your best candidates?

The hiring process is supposed to connect employers with the right people, but increasingly and infuriatingly, it's doing the opposite. 

Job seekers feel like their applications vanish into a void or get a rejection email that comes fast enough to give them whiplash. Recruiters feel like they're drowning in applications that all look the same. 

And sitting between them is a layer of automation that was built for efficiency but is systematically excluding the candidates both sides need.

To help unpack what's actually happening, we spoke with Douglas Clayton and Calida Jones, Co-Founders at Creative Evolutions, a human-centered consultancy that (among other tasks) facilitates hiring processes for organizations across the country. Their approach is built on a simple premise, that there's no single "best practice" for hiring, and the systems most companies rely on are filtering for the wrong things.

The 75% stat is wrong, but the problem it describes is real

Let's get this out of the way. The widely repeated claim that "75% of resumes are rejected by ATS before a human ever sees them" has no study behind it. It traces to promotional material from Preptel, a resume optimization company that published the number in 2012 and went out of business in 2013. No methodology was ever released, but the stat spread through years of uncritical repetition across career blogs, LinkedIn posts, and resume tool marketing.

When Enhancv surveyed 25 U.S. recruiters across 10+ ATS platforms in late 2025, 92% said their systems do not auto-reject resumes based on content at all. Most ATS platforms parse, rank, and sort. They don't throw resumes away.

But here's why debunking the stat doesn't make the problem go away. A resume that parses poorly, ranks low, or lands on page four of a recruiter's search results might as well have been rejected. That's not a myth, and it is a very real design problem. 

The most rigorous study on this topic is Harvard Business School and Accenture's Hidden Workers: Untapped Talent. It surveyed 8,720 workers and 2,275 executives across the U.S., U.K., and Germany, and estimated that 27 million Americans are "hidden workers" who are able and willing to work but systematically excluded by several factors, including automated screening. 

The most striking finding to keep in mind is that another study found 88% of employers acknowledged that qualified candidates are being screened out simply because their applications don't precisely match the language in the job description. 

That means if your job posting says "client relations" and a candidate's resume says "account management," they may never surface in your results even if they've been doing the exact same work for a decade. 

AI is making both sides worse

AI can be a helpful tool. We all know this. But we also all know it’s capable of making things a lot harder when it’s not yielded correctly. 

On the candidate side, AI tools have made it trivially easy to generate polished, keyword-optimized resumes at scale. Candidates can paste a job description into ChatGPT and get a tailored resume in seconds. 

Some go further. "Whitepaging," where candidates paste the full text of a job posting in white font on a second page so the ATS reads every keyword as a match, has been widely documented as a job seeker tactic. Modern parsers can catch the white-text trick, but the broader pattern of AI-inflated applications is harder to filter.

Jones has seen the impact firsthand. 

"There was one search we did where literally 10 people had the same answer," she said. "I had basically memorized it because I'm reading application after application and I was like, yeah, I just read this." 

She stressed that the problem isn't that candidates are using AI. It's that they're submitting AI output without adding their own thinking, which makes it impossible to distinguish between them or feel confident that they’re even fit for the position. 

On the employer side, AI-powered ATS features are layering additional algorithmic screening on top of keyword matching. These tools can introduce their own biases, filtering out candidates with non-linear career paths, employment gaps, or credentials that don't pattern-match to the training data. 

Clayton points to the root issue: "AI operates on best practice. It operates on the most common way, because that's what it knows, unless you tell it specifically not to." 

The result is that AI-driven screening can miss qualified candidates whose experience doesn't fit conventional patterns. Veterans, caregivers, and career changers are disproportionately affected.

And when all the resumes that make it through look identical, the system creates a new problem. 

"Now how are we distinguishing between 15 resumes that all have been crafted to use exactly the same language?" Clayton asked. "Are we just distinguishing based on the ranking of the university they went to? Or because this person has seven years of experience instead of six, which likely isn’t actually helpful in determining who the person is and what they can do?"

Clayton's broader point is worth noting, which comes down to the fact that AI screening shouldn't be a cost-cutting measure. It can help handle the high-touch work (skimming a thousand resumes), but you need the human element to consider unconventional candidates, lean into skills assessments, and be more intentional about what they're actually screening for.

How AI hiring discrimination is playing out in court

A federal class action lawsuit against Workday is testing whether AI hiring tools can be held liable for discriminating against Black, female, over-40, and disabled applicants. The lead plaintiff was rejected from over 100 jobs with companies that used Workday's AI screening. 

In 2024, a federal judge allowed the case to proceed, writing that "drawing an artificial distinction between software decision-makers and human decision-makers would potentially gut anti-discrimination laws in the modern era." 

With 356 million applications processed through Workday Recruiting in 2024 alone, the outcome could reshape how every employer thinks about automated screening.

What's being tested has nothing to do with the job

Clayton made the observation that most hiring processes evaluate candidates on skills that have nothing to do with the actual work. 

"How well you write a cover letter, how well you can tell your story on a resume, how well you can navigate an interview," he said. "There are some jobs where that is the skill. But most jobs, it isn't. And often you're way into a process before they go, 'all right, please do a skills test.'"

The entire ATS filtering layer operates on the same flawed premise. It evaluates how well a candidate can present themselves on paper, and not whether they can do the work in question. 

A software engineer who doesn't happen to list the exact framework you mentioned in the job description gets ranked below someone whose AI-generated resume included every keyword, and an experienced professional without a four-year degree gets filtered out before anyone reads the rest of their application. The system is testing resume-writing ability and keyword awareness instead of true performance. 

Jones described the gap between paper and reality as something she encounters constantly. 

"We see this all the time. Literally every day," she said. "AI will have you out here looking like somebody's CEO, and then they hire you, but really you should not be CEOing. Because there are all these different skill sets and muscles that you have to be able to flex."

What recruiters can do about it

The problem in question isn’t one you can solve by ditching your ATS. If only it was so easy, right?

In reality, most companies need applicant management software when you're processing hundreds or thousands of applications. But what you can control is what you layer on top of it and where you put the human checkpoints.

Here’s what you can do: 

Audit your knockout criteria

Many ATS configurations include screening questions or hard filters that were set up once and never revisited. Degree requirements in particular often exclude qualified candidates for most roles. 

Review what's actually filtering people out and ask whether those filters reflect the real requirements of the job.

Add an objective skills checkpoint

The most direct way to cut through the noise on both sides is to verify what candidates can actually do. A skills assessment administered early in the process gives you a signal that isn't affected by how well someone's resume was optimized for your parsing algorithm. It catches the qualified candidate whose resume parsed poorly. 

It also catches the unqualified candidate whose AI-polished resume parsed perfectly. As Clayton put it, "I'm a big believer in skills testing because you can test something specific and get actual data."

Don't automate the parts that require judgment

ATS is good at organizing, tracking, and surfacing information. It's not so great (aka bad) at evaluating nuance, reading between the lines of a non-linear career, or identifying potential in someone whose experience doesn't fit a template. 

Jones made the point simply: "Are we really giving people a real chance to show how they're qualified (or not) for a role if we aren't taking the time to read their materials, speak with them, and give them an opportunity to showcase and highlight what they can or can't do."

The system is broken in both directions

The irony of the current hiring landscape is that candidates and employers are both frustrated for legitimate reasons. Candidates feel like they're shouting into a void. Employers feel like they can't find qualified people. And a meaningful part of the explanation is the same, boiling down to the fact that the automated systems sitting between them are optimized for efficiency. 

If you want to expand your candidate pool and increase the quality of your hires, that’s not going to cut it. 

The fix is actually pretty simple, even if it goes against what top tech is saying. Stop using ATS output alone as a substitute for evaluation alone. It can help you manage volume and even move candidates effectively through the process when used ethically, but you need to use skills data to evaluate candidates and human judgment to find the right hire. And most importantly — build and keep enough human checkpoints into the process.