Every few months I see the same line in a 'how to beat the ATS' article: 75% of resumes are rejected by an Applicant Tracking System before a human ever reads them. I have used a version of that number myself in earlier guides on this site, because it is repeated so widely it stopped sounding like a claim and started sounding like a fact. Then a reader emailed me asking where the number actually came from. I did not have a good answer. So I spent a weekend finding one.
Where the 75% number actually came from
Several people before me have done the same digging, and the trail is consistent. The 75% figure shows up first around 2012, in promotional material from a small resume-optimization company called Preptel. Preptel closed in 2013. No survey, dataset, or methodology was ever attached to the claim, before or after. In 2014, a Forbes contributor who ran a competing resume service repeated the number in an article that read more like an advertisement than reporting. From there it followed a familiar pattern: a tech publication cited the Forbes piece a few years later, a business news outlet cited the tech publication, and within a decade the number had been copied into hundreds of blog posts, YouTube scripts, and LinkedIn posts, almost none of which link back to an original source. It is a tidy case study in how a made-up statistic becomes 'common knowledge' simply by being repeated enough times by people who sound credible.
What the real research actually says
The most credible large-scale research on this topic is not about a flat rejection percentage at all. In September 2021, Harvard Business School's Project on Managing the Future of Work, working with Accenture, published a report called 'Hidden Workers: Untapped Talent,' based on surveys of more than 8,000 workers and over 2,250 executives across the US, UK, and Germany. The headline finding: 88% of the executives surveyed admitted that their own hiring software, mainly resume screening configured around exact-match keywords and credentials, was filtering out candidates who were genuinely qualified for the role. The report estimates that around 27 million people in the US alone are 'hidden workers,' capable candidates excluded by how a system is set up, not by a lack of skill.
That is a meaningfully different claim than '75% of all resumes are auto-rejected.' It is not a fixed percentage applied to every application that comes in. It is a finding that companies frequently set their filtering criteria too narrow, for example requiring an exact job title match or a precise number of years, and the software does exactly what it was configured to do.
More recent data points the same direction, even if the exact number keeps moving. A 2025 SHRM survey of HR professionals found that 43% of organizations now use AI somewhere in their HR function, up from 26% the year before, and 51% use AI specifically in recruiting. Of the organizations using AI in recruiting, 44% use it to screen resumes. Tellingly, 19% of organizations using AI or automation in hiring admitted, in that same survey, that their own tools had overlooked or screened out qualified applicants.
So does an ATS actually reject most resumes?
Mostly, no, not in the way the viral stat implies. As I wrote in how an ATS actually works, most platforms are built to parse, store, filter, and rank, not to issue an automatic, irreversible rejection on most applications. A resume that matches the job description's language ranks near the top of a recruiter's search results. A resume that does not match ranks low enough that a busy recruiter, working through 200-plus applications for one opening, may never scroll down far enough to see it. The practical outcome can look identical to rejection. The mechanism is not a robot hitting reject, it is a sorting algorithm combined with a recruiter's limited time.
The exception
Some applications use hard knockout questions, things like work authorization, required licenses, or a minimum years-of-experience cutoff, that really can filter a candidate out before a human ever opens the resume. That is a genuine, configurable form of automatic rejection. It is just a narrower mechanism than 'the ATS read my resume and rejected it.'
Why the inflated number still causes real damage
Statistics like this are not harmless just because they are wrong. I have talked with job seekers who, believing a fixed 75% of resumes get tossed by a robot no matter what they do, either give up on tailoring their resume at all ('why bother, it's random') or swing the other way and try to game the system with invisible white-text keyword stuffing, which most modern parsers catch and which can actively hurt your ranking. Both reactions come from treating the ATS as an unbeatable black box instead of what it actually is: a search engine that rewards a resume written in the job description's own language. See what counts as a good ATS score and how to tailor a resume to a job description for the version of this advice that is actually backed by something real.
What the numbers that do hold up actually tell you to do
- ▸Stop trying to beat a fictional reject button and start matching the job description's exact language, since ranking, not rejection, is the real mechanism.
- ▸Take narrow filters seriously: if a posting says '5+ years' and you have 4, a tailored summary that reframes your experience matters more than a single stat suggests.
- ▸Assume formatting still matters. Parsing failures from tables, columns, and text boxes are a documented, mechanical cause of lost data, unlike the 75% figure.
- ▸Skip keyword stuffing. Modern parsers, and the SHRM data above, both point to screening systems getting more sophisticated, not less.
Key takeaway
The 75% number is not real, at least not as a verified statistic. What is real, and better documented, is that a meaningful share of qualified candidates lose ground to overly narrow filters and language mismatches they could have fixed. Treat your resume as something to make legible and well-matched to the job description, not something to trick. That part of the myth was true all along.