Where the 75% ATS rejection number came from

You have almost certainly read some version of it. Around three quarters of resumes are thrown out by software before a person ever looks at them. It appears in blog posts, in LinkedIn advice, in the sales copy of products that exist to save you from it, and until recently it appeared in eight articles on this site, which is part of why this page exists.

It is a myth. Not an exaggeration or a figure that has aged badly, but a number that was never measured in the first place.

Following the citations backwards is unusually easy, because they run out fast. Almost every article quoting the figure cites another article quoting the figure. Keep going and the trail ends in the same place: marketing material published in 2012 by Preptel, a company selling resume optimisation.

Preptel went out of business in 2013. It published no methodology, no sample size and no survey. Career consultant Christine Assaf documented this in an investigation titled Your Job Application Was Rejected by a Human, Not a Computer, concluding that the statistic was created without any study, survey or context behind it.

So the most repeated number in resume advice originates with a defunct vendor that had every commercial reason to make screening software sound frightening, because it was selling the remedy.

A statistic does not need to be true to travel. This one had three things going for it.

It explains a painful experience. Applying for work and hearing nothing is demoralising, and a number that blames a machine is kinder than the alternatives.

It is unfalsifiable from where you sit. You cannot see what happened to your application, so nothing you experience will ever contradict it.

And it sells things. Every product that promises to get you past the robot is easier to sell once you believe the robot is there.

The honest answer is that there is no good public figure for how many resumes are discarded without human review, because the data sits inside employers rather than in published research.

What does exist is narrower and more useful. Enhancv put the question directly to 25 US recruiters in 2025, and 23 said their systems do not auto-reject on formatting, content or design. Twenty-five people is a small sample and we are not going to present it as more than that, but it is 25 more people than the original claim was based on.

Harvard Business School and Accenture found in 2021 that 88% of employers say their own systems filter out qualified candidates. Read that precisely: it is employers describing what they believe their software does, which is a different measurement from a rejection rate, and it says the problem is real without saying how large it is.

The mechanism people attribute to parsing is usually something simpler. Employers configure knockout questions, and those genuinely do filter automatically: work authorisation, a required licence, a minimum years-of-experience gate answered yes or no. That is a rule someone deliberately set, not a machine judging your formatting, and it is worth understanding properly: how knockout questions work.

Parsing still matters, just not as an execution. A resume the parser reads badly is harder to search and harder to shortlist, which costs you opportunities without ever producing a rejection.

Which formatting choices genuinely cause a parser trouble is covered in what an ATS actually does with your resume, and what an employer is required to disclose is in AI in hiring: what you are entitled to know.

We sell a resume score, so it is worth being direct about what ours is. It is an AI estimate of how well a resume is likely to parse and match a posting. It is not a test against any employer's applicant tracking system. Nobody outside Workday, Greenhouse or Taleo can run one, and anyone who claims to is describing something they cannot do.

We would rather tell you that than repeat a number that would make our own product sound more necessary.