Where the "75% of Resumes Are Rejected by ATS" Stat Came From

Somewhere between 75% and "nearly all" resumes, depending which article you land on, get rejected by an ATS before a human ever sees them. It is repeated constantly, cited by career coaches, and used as the opening line in a lot of resume-tool marketing, including some of ours, historically. We went looking for where it actually comes from. Here is what we found, and what we could not find.

Where the number actually comes from

The honest answer is: we could not find a primary study behind it. Several independent sources trace the claim back to a 2012 vendor marketing pitch, not a peer-reviewed study, an academic paper, or even a large-scale industry survey with a published methodology. That is a meaningfully different thing than "a study proved this." A vendor pitch is a sales document. It exists to sell ATS software or resume-optimization tools, and a scary number is a good way to do that.

We want to be precise about what we're claiming and what we're not. We are not saying the statistic has been definitively debunked. Our own attempt to re-verify its exact origin story got mixed results, some sourcing lines up cleanly with the 2012-pitch explanation, some doesn't fully close the loop. What we can say confidently is that no one has produced a real primary study that supports the number, and the most commonly cited origin is a piece of marketing collateral, not research. The right label for a claim like that is contested, not confirmed and not disproven. Anyone who tells you it's settled in either direction is overselling their certainty.

What an ATS actually does

An applicant tracking system is, functionally, a searchable database with a workflow attached. Recruiters use it to store every application that comes in, search across them by keyword, and move candidates through stages, screened, interviewing, offer, rejected. That last transition is normally a human clicking a button, not software autonomously deciding. Most platforms are built to help a person triage a large pile of applications faster, not to replace the person making the call.

That doesn't mean the software is irrelevant to your odds. If an ATS can't extract clean text from your resume, a recruiter searching for a keyword you actually have on your resume won't find you, and depending on the volume of applications, that alone can mean nobody manually reads the file at all. The mechanism of harm is a bad parse making you invisible in a search, not a robot reading your resume and rejecting it on the merits. The end result can look identical from the outside. The cause is different, and the fix is different too: fixing your formatting, not "beating the algorithm."

What actually causes an instant, algorithmic rejection

If there is a real auto-reject mechanism in modern hiring software, it is knockout questions, not resume parsing. Many applications now include a short screening form alongside the resume upload: are you authorized to work in this country without sponsorship, are you willing to relocate, do you hold this required license or certification. Answer wrong, and some systems are configured to filter the application out before anyone reviews it, completely independent of what the resume itself says.

This shows up constantly in hiring discussions in a specific way: visa and sponsorship status gets treated as an override on everything else. When someone posting for feedback mentions they need visa sponsorship, the response is routinely some version of "this is almost definitely the problem," ahead of any critique of the resume's actual content. One widely-discussed, if unverifiable, account describes a candidate changing only their name and visa status on an otherwise-identical resume and getting hired after 18 months of no callbacks under the previous identity. We can't verify that specific story, and we're not presenting it as confirmed fact, but the pattern it illustrates, that one checkbox answer can matter more than resume quality, lines up with how recruiters consistently describe knockout questions working.

Is AI actually screening applicants now?

This is where the discourse gets genuinely unresolved, not just contested in the way the 75% stat is. On one side, a recruiter with decades of hiring experience will flatly call the whole "AI is screening you out" narrative a myth, kept alive by resume-coaching businesses because fear sells consultations. On the other side, real, named employers now put a literal "do you consent to AI screening?" checkbox in front of candidates during the application itself. Both of those things are true at the same time, for different companies. There is no single honest answer to "does AI screen resumes now" that applies across every employer, because the honest answer is that it depends entirely on which employer, and that information usually isn't visible to the applicant.

What we won't do is resolve that disagreement for you with a made-up number, because nobody currently has one worth trusting. What we will say is that the practical advice barely changes either way: write for a human first (clear structure, real evidence, no padding), keep the file parseable (so whichever kind of software is involved reads it correctly), and stop treating "the algorithm" as a monolith you can reverse-engineer. There isn't one algorithm. There are thousands of different configurations across thousands of different employers, and optimizing for a specific mental model of "the ATS" is optimizing for something that mostly doesn't exist in the form people imagine.

What still actually matters for parsing

None of this means formatting is a myth alongside the 75% stat. It means the reasoning behind good formatting advice is usually stated backwards. Tables, text boxes, columns, and icons are a real risk, not because a robot penalizes you for using them, but because some parsers genuinely garble the text extraction when content sits in those structures, out-of-order paragraphs, missing words, headers that vanish. A resume the parser mangles is a resume the human reviewer never gets to read cleanly either, since most reviewers work off the parsed/searchable version, not a rendered image of your PDF.

The PDF-versus-Word debate is a good example of why this space rewards skepticism over folk wisdom. "PDF is always safest" is repeated as settled fact almost everywhere, but it isn't universal: real parsing-accuracy comparisons exist showing at least one major platform reading DOCX files more reliably than PDFs for certain resumes. The honest position is that format safety depends on the specific platform an employer uses, which you usually can't know in advance, so the more reliable move is keeping the underlying structure simple (single column, standard section headers, real selectable text, not text embedded in an image) regardless of which file format you end up exporting.

It's also worth separating ATS anxiety from a different, larger problem: a lot of "my applications go nowhere" experiences trace back to postings that were never real openings in the first place, filed for compliance reasons, kept open to build a talent pipeline, or already filled internally. Surveys of recruiters themselves report a sizeable share admitting their own employer posts listings that aren't real. That's a labor-market problem, not a formatting problem, and no amount of resume optimization fixes it. Knowing which problem you actually have changes what's worth your time.

Three different problems people lump together as "ATS rejection"

• A bad parse: your formatting garbled the extracted text, so a keyword search doesn't find you even though the words are on the page. Fix: simplify the structure, not the content.
• A knockout question: a screening-form answer (sponsorship, location, a required license) filtered you out before the resume mattered at all. Fix: nothing about your resume changes this, it's a fit question, not a quality one.
• A ghost or already-filled listing: the role was never really open, or was filled before you applied. Fix: also nothing about your resume, and no amount of rewriting closes this gap.

Only the first one is actually about your resume. Conflating all three into "the algorithm rejected me" is exactly how a contested statistic turns into a load-bearing belief that shapes how people write, when the real fix, in most cases, has nothing to do with writing at all.

The version of this we actually believe

Both readers matter: the parser deciding whether your text extracts cleanly and shows up in a keyword search, and the human who spends about eight seconds on the initial scan once it does. Optimizing for one at the expense of the other doesn't make sense, since both have to say yes. That's the entire premise behind SupaCV's ATS checker: a real score across six dimensions (keywords, formatting, completeness, quantification, readability, and language quality), not a single mystery number, and not a scare stat to get you to sign up.