Hype Index: 'AI Rejects 75% of Resumes Before a Human Sees Them'

TL;DR: No. I selected 30 currently-live pages asserting that 75% of resumes are rejected before a human sees them, or its 2026 restatement that AI does the rejecting, and followed every citation to its terminus. 29 resolved. Zero terminated in a primary study. Fifteen cited nothing at all, which puts the median chain depth at 0 hops. The longest chain in the sample was two hops and it ended at a page that returns HTTP 500. The oldest use of "75%" I could retrieve from Preptel, the vendor the figure is usually credited to, is a keyword-match threshold, not a rejection rate. Automatic rejection is real, but it comes from recruiter-configured knockout questions, which is a different mechanism from the one the claim describes.

Several sites have already called this figure a myth, and repeating them earns nothing. What nobody has published is the shape of the chain. So I measured it.

How did I run the count?

On August 15, 2026 I assembled 30 currently-live pages that assert the claim, from Fortune and Forbes down to resume-tool blogs, a Minnesota state government careers page, and two Medium posts. I fetched each with curl, located the sentence carrying the figure, extracted every hyperlink and named attribution within one sentence of it, then fetched that target and repeated until the chain stopped.

Chain depth is the number of hops from the page to the last document in its chain. Naming no source is depth 0. Linking a primary study is depth 1. Linking a blog that links a study is depth 2.

29 of the 30 resolved. intelligentcv.app returned HTTP 500 on two attempts with different user agents, so it is out of the tally. It matters anyway, and I will come back to it.

Where does the chain actually end?

Terminus	Pages
A primary study containing the figure	0
The Preptel vendor claim	1
Another page that does not carry the figure further	13
Nothing at all	15
Resolved	29 of 30

Median chain depth: 0. Mean: 0.59. Longest: 2.

Fifteen of 29 pages state the number with no source of any kind, or with an attribution that names nobody: "according to multiple industry reports," "according to 2026 recruitment benchmarks," "you have heard the statistic." TopResume tells readers "75% of job applications don't get seen by human eyes" and cites nothing. So does the Minnesota CareerForce page, which puts it at "about 75 percent."

What did the citations turn out to be?

The 14 pages that do cite something are the interesting half, because most of them cite a document that does not contain the claim.

Forbes, March 11, 2026, headlined "Why AI Is Rejecting Your Resume Before Humans See It," links the words "over 75% of resumes are rejected by ATS software" to a staffing agency's blog post. That post's own opening line reads: "Despite the often-cited '75% of resumes are rejected by ATS' statistic, there's no strong empirical evidence that ATS systems flatly reject that large a share before a human sees them." Forbes sourced the number to a page debunking it. The same article's second use of the figure links to an Economic Times story about a worker rewriting a resume with AI, which does not contain the figure either.

Two pages credit Jobscan. I read Jobscan's ATS usage report and its ATS landing page. The only "75%" on either is a match-score readout in the product interface. Jobscan does not make this claim.

One page credits SHRM and links to SHRM's talent acquisition topic hub. One credits Klaxos and links to a page with no such figure. One credits a 2014 Forbes Next Avenue column that does not contain it. One, an AI-in-hiring statistics roundup, attributes "75 percent of resumes are discarded without human review" to HiringThing, whose article is titled "Applicant Tracking Systems Aren't Excluding Job Applicants, People Are" and exists to dismantle the number.

The single page that reaches Preptel gets there wrong. ExpertResumePros writes: "A 2019 study from job search services firm Preptel found that a whopping 75% of resumes failed to pass ATS systems." Preptel shut down in 2013. Its supporting link is a Yahoo Finance URL that now 404s.

Did the AI restatements bring any new evidence?

Eight of the 29 are AI-era restatements: 2026 pages that move the agent of the sentence from software to AI. Three of the eight cite any source dated after 2023, and all three of those sources are other 2026 blog posts in this same sample that themselves cite nothing primary. The restatement refreshes the date on the citation without adding a single new observation.

The flagship is Fortune, March 15, 2026: "75% of resumes never reach a human: the new rules of job searching in the AI era," bylined by Alex Chepovoi, co-founder and CEO of Global Work AI. Its sentence reads: "Not because of people but Applicant Tracking Systems, declining 75% of resumes." The only link attached to it points at intelligentcv.app, a resume tool's blog post, which is the one page in my set that would not load. A business magazine's AI-era version of the claim, written by a vendor selling AI job-search software, is sourced to a resume tool's marketing page that is currently returning a server error.

What was the 75% originally counting?

Something else. I pulled Preptel's own archived site. Its October 2011 post on what HR managers do when job hunting says that "more than 90% of today's employers trust résumé scanning software to select 1st tier candidates, and those 1st tier résumés have to match at least 75% of the keywords in the job description to get forwarded to HR."

That is a match threshold, a score a resume needs to clear, and the mirror image of the claim built on top of it. The January 2012 snapshot of Preptel's homepage carries no 75% figure at all; its numbers are "82% get interviews" and "61% get an offer" for customers rated Very Strong. The vendor's number described how good you had to be. The industry rewrote it as how many people get killed.

Jobscan, incidentally, still uses 75% as a match-score target. The threshold survived. Only the sentence changed.

So does anything reject you automatically?

Yes, and this is the part the debunks undersell. Enhancv's structured interviews with 25 recruiters across 10-plus ATS platforms found that 23 of 25 (92%) do not configure content-based auto-rejection, and only 8% use an AI match score as a definitive filter. But the same 25 report knockout rules used 100% of the time when present: yes or no eligibility checks on work authorization, required licence, location. Those reject you instantly, without a human, by design.

So the honest version is not that nothing is automated. The automation you actually face is a recruiter's checkbox about whether you can legally do the job, not an algorithm grading your font. Reformatting does not get you past a work-authorization knockout. Twenty-five recruiters is a small sample and I am not treating 92% as a population parameter, but it is 25 more recruiters than the 75% figure has ever had behind it.

What would change my mind?

A published dataset. Any ATS vendor with a customer base could report the share of applications auto-rejected before human view, and none has. Employers are not filling the gap either, even where a statute tells them to: zero of 201 Illinois postings mention the disclosure law that has applied since January. If Workday or iCIMS published that number and it landed near 75%, this post is wrong and I would say so.

What this count does not prove is that your resume gets read. Enhancv's recruiters describe the real bottleneck as volume, and postings drawing hundreds of applicants in days mean plenty of resumes are never opened by anyone. That is a true and depressing thing. It is just not a robot rejecting you, and it is not measured by this number. Same failure mode as the AI SDR churn stat that traces to an article which does not contain it, and the same as the outreach industry's favourite personalization stats. A number detaches from its referent, the hedge falls off, and the sentence outlives the measurement. For the version of this where the rules are written down in public and you can actually check them, see what open-source projects now require of AI-written code.