You've heard the stat a hundred times: "48% of resumes never reach a human." But here's the problem—no one ever shows their work. The claim floats around LinkedIn think-pieces and career coaching blogs like received wisdom, but hard data? That's been conspicuously absent. Until now.

How We Ran the Experiment

Over six months spanning March through August 2026, a team ran 5,124 real resumes through six of the most widely deployed Applicant Tracking Systems in corporate recruiting: Workday, Greenhouse, Lever, iCIMS, Taleo, and LinkedIn Easy Apply. These platforms collectively handle millions of job applications annually, making them the de facto gatekeepers for anyone trying to break into new roles or pivot careers. The methodology mattered here. They weren't testing with dummy data or synthetic resumes cooked up in a lab—they used actual CVs from real candidates across industries and experience levels. Each resume was tracked through every platform to identify exactly where applications died in the pipeline, what triggered rejections, and which resume features survived parsing versus those that got silently filtered out.

What We Found

The results expose something most job seekers intuitively suspect but rarely prove: these platforms aren't just tracking candidates—they're actively making decisions about who deserves a human's attention. File format issues, keyword mismatches, formatting quirks that confuse OCR parsers, and ATS-specific requirements buried in application portals all contributed to resume death before any recruiter opened their inbox. Different platforms showed dramatically different behavior. Some were aggressive keyword matchers that filtered out qualified candidates missing specific terminology. Others had strict file size limits or required structured data fields that most resumes don't naturally include. LinkedIn Easy Apply, for instance, has its own quirks around how it maps your profile data into application forms—quirks that can quietly torpedo applications from users who think their updated LinkedIn profile is doing the heavy lifting.

The Real Problem With ATS Black Boxes

Here's where this becomes a dev-tools story and not just a career-advice piece. These ATS platforms are software systems making consequential decisions about people's livelihoods, yet they're largely opaque to both candidates and often to the recruiters using them. Most hiring managers don't know exactly what filtering their own ATS is applying—they just notice they're reviewing fewer "qualified" candidates than they expected. The irony is thick: companies spend millions on these systems hoping to find better talent faster, while simultaneously building automated barriers that may be rejecting their ideal candidates before anyone reads a single line of their work history. The data from this experiment suggests the problem isn't just algorithmic bias in AI screening—it's basic infrastructure failures baked into how these systems parse and process human-readable documents.

Key Takeaways

  • File format matters more than you think: PDF versus Word, embedded fonts, and document structure all affect how ATS parsers extract your information
  • Keyword stuffing isn't dead—it just needs to be strategic: Different platforms weight keywords differently based on job descriptions versus resume content
  • Structured data wins: Resumes that mirror job posting language in specific fields outperform those with equivalent but differently phrased experience
  • Platform behavior varies wildly: A rejection from one ATS doesn't mean you'll fail another for the same role

The Bottom Line

ATS vendors have built billion-dollar businesses on the premise that their software helps companies hire better, faster. But until someone actually stress-tests these systems with real candidate data—which is exactly what this experiment did—we're all just guessing about what's really happening in that black box. If you're job searching and wondering why your "perfect" resume keeps disappearing into void, here's your answer: the machine said no, and you'll never know exactly why. Now if you'll excuse me, I'm going to go update my own LinkedIn profile to account for whatever weird parsing quirk Greenhouse apparently has with bullet point formatting.