Short answer: no. The number is a garbled 2012 sales pitch, and the company that first floated it was out of business by 2013. Does an applicant tracking system ever reject you on its own? It does, but in one narrow way that an employer switches on deliberately, and it has almost nothing to do with the story you have heard, the one where a robot reads your resume, catches a missing keyword, and trashes it before a person ever sees your name.
That version is folklore. What actually happens after you hit submit is both more boring and more worth knowing.
Where the 75% number came from
It traces back to a single source: one company, quoted in one article. Preptel, a service that sold resume optimization, was quoted in Computerworld in March 2012 saying that applicant tracking systems “kill 75% of candidates’ chances of landing an interview.”
The wording is where the whole thing turns. The claim was about chances, not resumes. It was an odds claim floated by a vendor who happened to sell the cure, and nobody in that sentence said three of four resumes get auto-deleted, or that a human never lays eyes on yours. A soft claim about odds is a dull thing to repeat, though, so somewhere across a decade of blog posts and LinkedIn threads it firmed up into something crisp and frightening: the machine rejects 75% of resumes. The stat got scarier every time it changed hands.
There is no study under any of it. No sample size, no methodology, no dataset you could go and check. An HR professional named Christine Assaf went hunting for the source in a 2020 investigation that Ask a Manager later reposted, and what she found was a hall of mirrors: outlets citing “Preptel” with no link, a Forbes mention that credits no one, and not a single hit on Google Scholar. Preptel itself folded in 2013, so the number has now outlived, by more than a decade, the company that put it into the world.
The tell is that the figure will not sit still. Each retelling rounds it off and drops its one weak citation, until it reads as settled fact with nothing behind it. Real statistics do not drift like that, because a real statistic is bolted to a specific study with a specific method, and that anchor holds the number in place. A laundered one drifts, because there was never anything underneath it to hold. A decent rule of thumb: a scary hiring stat with no link to click is one you should assume somebody invented.
What an applicant tracking system actually does
Strip off the branding and an applicant tracking system is a database, with search bolted on top and a little workflow around the edges. It takes your resume, parses it into fields, and files it. A recruiter opens that database, searches it, sorts what comes back, and drags candidates from one stage to the next. That is the whole job, closer to a specialized inbox than to a bouncer with opinions about your career.
You do not have to take our word for it. Greenhouse is one of the bigger vendors, and here is how it describes its own product to the people applying: “AI doesn’t score or rank applications, nor does it make any decisions about whether or not you move forward in the process.” Applications, the company says, “are reviewed by real people, one by one, in the order received.”
That is one vendor, and some systems do sell AI scoring as a bolt-on, so do not walk away thinking no software anywhere ever ranks a resume. That is not the claim. The claim is narrower, and it holds: out of the box, a mainstream ATS files and sorts, and the deciding goes to a person.
When the software really does reject you on its own
There is exactly one common situation where the software rejects you with no human in the loop, and it is both real and easy to sidestep: an employer’s knockout questions. Most systems let an employer bolt these onto the application. Greenhouse’s own support documentation spells it out: an organization “can set up application rules so that, based on an applicant’s answer to a question, they will automatically be rejected.” Workable does the same thing, and adds a detail that matters, that these rules are “supported only for Yes/No types of questions.”
So the automatic gate is not a machine squinting at your resume and disliking your verbs. It is an answer rule the employer set. Are you authorized to work in the United States? Do you have an active nursing license? Are you able to work on-site in Denver? Answer one of those the wrong way and yes, you are out, with no human required. Workable caps these to Yes/No questions; Greenhouse also lets an employer key the rule to a single-select or multiple-choice answer. Either way the footprint is the same: a handful of screening questions an employer chose to ask, not a verdict on your resume.
There is a larger and more serious version of automated screening, and this one has real research behind it. In 2021 Harvard Business School published Hidden Workers: Untapped Talent, built on a survey of more than 2,250 executives. It found that most employers do use these systems to filter or rank applicants, and that capable people get screened out at scale. This is where that “88%” you have probably run into actually comes from: 88% of the employers surveyed agreed that qualified, high-skills candidates get “vetted out of the process because they do not match the exact criteria established by the job description.”
Look hard at what the 88% is counting, though. It is employers admitting that their own hiring process throws away good people, not a rejection rate a keyword robot applies to your resume. The report gives a concrete case: set the system to flag any resume with an employment gap over six months, and it will do exactly that, automatically, on that basis alone. That is real and it is automatic, but it is still a switch a human chose to flip. So when someone recites the 88% as “88% of resumes are auto-rejected,” they have made the exact error that turned a 2012 sales line into gospel.
So what actually filters you out?
Four things do, and not one of them is a keyword robot reading your resume:
Knockout questions
The one genuinely automatic gate, the one from a few paragraphs up. Work authorization, a license you do not hold, a city you cannot get to. Answer them straight. And if you keep tripping the same knockout over and over, the problem is not a robot, it is that you are applying to jobs you are not eligible for.
A recruiter searching and skimming
This is the real bottleneck, and it is a person. A popular corporate req can pull far more applicants than any human could read end to end, and nobody sits down and reads all of them front to back. The recruiter searches the pile for the few resumes that hit the terms they care about, pulls a shortlist, and reads those. Most rejections are this undramatic: a human glanced, decided you were not among the closest matches, and moved on.
Employer hard filters
The six-month-gap and must-have-this-exact-credential rules from the HBS report. Automatic, yes, but employer-configured, and nowhere near universal.
Parsing failures
The one nobody warns you about. Build your resume out of tables, columns, text boxes, header regions, or section labels the parser has never seen, and it can garble the import, dropping your dates or titles or degree into the wrong field. The irony is a quiet one: that original 2012 article was about a resume that scored badly precisely because the system misread its formatting, not its keywords. A broken layout is a more real threat to you than keyword density ever was.
Should you still tailor your resume?
Yes, though not for the reason you were sold. Tailoring works for two plain, mechanical reasons. A human recruiter searches and ranks by the words in the posting, so matching those words gets you found and gets you ranked higher. And your resume has to survive parsing in the first place, or none of your words are legible to anyone. What tailoring does not do is outwit a keyword robot that was about to auto-reject you, because that robot is not there. Jan Tegze, a recruiter who has written about this from inside the field, puts it plainly: the large majority of applications are seen by a human, and tailoring buys you rank and attention, not a secret override.
In practice it comes down to a few unglamorous habits. If the posting says “stakeholder management” and your resume says “worked with clients,” and that is honestly the same work, switch to their phrase, because “stakeholder management” is the string the recruiter will type into the search box. Keep the layout boring enough for a parser to read: one column, section headings a machine will recognize, nothing load-bearing buried inside a graphic. And slow down on the screening questions. Answer them carefully, answer them honestly, because on the whole application they are the only thing with the power to reject you by itself.
The version worth keeping is simple. The software files and searches, and a person decides. There is no machine quietly binning three of every four resumes over a missing word, because that machine was a sales pitch, and the company behind it has been gone since 2013. In almost every case a human is reading you. That is worse news than the robot story in one way, since you cannot blame an algorithm for a no, and much better news in every other way, since humans can be persuaded and robots cannot.
How we know this: every figure here is traced to a primary source. The “75%” origin comes from the original 2012 Computerworld article; the description of what applicant tracking systems do comes from the vendors’ own product and support documentation (Greenhouse, Workable); the scale of automated screening comes from Harvard Business School’s Hidden Workers report. Where a claim is one practitioner’s estimate rather than a study, we have said so.