Do ATS Reject 75% of Resumes? The Oldest Sources We Found

The oldest sources we found for the claim that applicant tracking systems (ATS) reject 75% of resumes before a human sees them are two articles from March 2012. Both credit a 75% figure to one company, Preptel, which sold a resume-optimization tool, and neither says how the number was produced. Hiring software does reject some applicants, through knockout questions and auto-reject rules that employers set, and several products sort or rate the applicants who remain, which can affect who is read first. We found no published count of how many applications are rejected this way.

We checked the sources on October 9, 2026. Offermic, which publishes this page, sells a resume builder, so we have our own commercial interest in what you believe about resumes and hiring software. We are not affiliated with any company named here.

The oldest sources for the 75% figure

We went looking for two things: the earliest dated place the number appears, and anything that shows how it was measured.

The earliest appearance we found is an article by Meridith Levinson of CIO.com. The CIO.com address we tried now leads to a different, later article, so we read the copy on Computerworld, which dates it March 4, 2012. It opens by saying that applicant tracking systems

kill 75% of candidates’ chances of landing an interview as soon as they submit their resumes, according to job search services provider Preptel.

The article names no study behind the number and gives no sample or method. It introduces Preptel’s CEO, Jon Ciampi, as a former general manager at SumTotal Systems, “a maker of applicant tracking systems”, and says his company “aims to help job seekers penetrate these systems”.

The second, a Scripps Howard News Service column by Marvin Walberg dated March 22, 2012, lists four things that it says happen “At most companies”. One of them:

At least 75 percent of resumes are discarded for using the wrong words.

The line after the list reads: “These statistics are offered by Jonathan Ciampi, president and founder of Preptel Corp. in Danville, Calif.” Preptel reposted the start of the column on its own site on April 2, 2012.

So within one month the figure appeared in two wordings, one credited to the company and one to its founder. The first can be read two ways, as three in four candidates losing their chance or as each candidate’s chance falling by three quarters, and the article doesn’t say which. Both are close to the claim repeated today, and neither is its exact sentence: the words about a human never seeing the resume aren’t attached to the figure in either article. The Scripps column comes nearest. Its list begins with “Computers read your resume first, not humans.”, and its fourth tip says “a human will only read it if it is a good match to the job description.”

Both articles point to Preptel, so we went to its own site as the Internet Archive kept it, and read or searched 18 pages from captures made between 2010 and 2013.

Preptel’s product, Resumeter, scored a resume against a job description. The product page, in a capture from September 2010, says: “People don’t weed-out resumes, computers do.” Preptel had a commercial interest in that message, as we have in ours.

The home page, as captured on March 9, 2012, linked to one report, which made it the obvious place to look for the number. It is a Bersin & Associates research bulletin by Josh Bersin, dated March 18, 2011, about Preptel’s product, and the 75% figure is not in it. What it reports is a “little test”: one resume, written to be a perfect match for a clinical scientist job, submitted to “a typical Taleo applicant tracking system”. The resume “only scored 43 percent “relevance” in the ATS” after the parser lost one job, a location and two degrees. Then there is the figure that shows the score. It is a screenshot of a candidate search-results list sorted by a “Relevance” column, credited to a “Bersin & Associates and Preptel Live Test”, and the test resume is the first row in it, at 43%, with the next candidate at 42%. The CIO.com article retells the test without that detail. A parser misreading one resume in 2011 isn’t a rejection rate.

Preptel did publish claims about its own data. Blog posts by Jon Ciampi dated December 29, 2011 report average “job fit” ratings “After analyzing thousands of job seeker résumés”. Those posts don’t contain the 75% figure, and they give no exact sample or method.

The latest of the 18 pages is the home page as captured on October 7, 2013. It carries a notice that says “we have had to shut down the service” and asks customers to download their resumes by August 30, 2013.

Almost a year after that date, a Fortune advice column of August 22, 2014 told a reader that the “75% rejection rate” “probably came from a study by a job search services firm called Preptel”. The column gives no title, date or sample for that study. The reader’s letter, as printed, gives no number: it says a colleague read that the systems “reject most resumes before a human being even gets involved in the process”.

Where the trail stops

We found no Preptel study, data set, sample size or method behind the number. We also did not find a 2012 source that says 75% of resumes are rejected before a human sees them.

Our reading has limits. The Internet Archive holds several hundred more addresses from Preptel’s site that we did not open, and we did not read paywalled coverage from that period.

“Attributed to Preptel” is therefore as far as the evidence goes. Two writers credited the company or its founder, the company reposted one of them, and we found no place where Preptel explained the number.

What the Harvard and Accenture report measured

The strongest published evidence we found that software filters people out is “Hidden Workers: Untapped Talent” by Joseph Fuller, Manjari Raman, Eva Sage-Gavin and Kristen Hines, published in September 2021 by Harvard Business School’s Project on Managing the Future of Work and Accenture. It rests on two surveys. One covered 2,275 executives in the United States, the United Kingdom and Germany and was fielded online by a third-party firm in January and February 2020. The other covered 8,720 workers in the same three countries.

The authors separate the ATS, which they call “a workflow-oriented tool”, from a Recruiting Management or Marketing System (RMS), which supports recruiters in tasks that include “automated candidate scoring”. About two-thirds of the employers surveyed (63%) said they use an RMS, and the report says that “more than 90% of employers in our survey use their RMS to initially filter or rank potential middle-skills (94%) and high-skills (92%) candidates.”

On whether qualified people get filtered out, it reports: “A large majority (88%) of employers agree, telling us that qualified high-skills candidates are vetted out of the process because they do not match the exact criteria established by the job description. That number rose to 94% in the case of middle-skills workers.” The 88% and 94% come from a question answered only by employers whose organization uses an RMS, and they add together the answers “Yes, sometimes”, “Yes, often” and “Yes, always”.

The report is specific about one filter: “48% of employers filtered middle-skills candidates based on employment gaps of more than six months.” Of a resume with such a gap, the authors write: “A recruiter will never see that candidate’s application, even though it might fill all of the employer’s requirements.” The report’s examples are criteria the employer set, such as a degree, a credential, years of experience or a gap in work history, and its own heading for the problem is “Inflexibly configured automated recruiting systems”.

One sentence puts a size on the filtering, as an estimate: “As many as 78% of the business leaders we interviewed estimated that half or more of middle-skills candidates were eliminated by filtering, and 80% said that more than half of candidates for high-skilled positions were similarly disqualified.” It is printed without the question or the number of leaders who answered it.

All of these are executives’ answers, collected in early 2020: beliefs and estimates about their own organizations. The report did not examine any employer’s applications or software, and it gives no measured share of resumes that software rejected. It supports the statement that many employers filter with software and believe they lose qualified people by doing so. It does not support 75% or any other measured rate.

What 25 recruiters told Enhancv

The main published evidence we found on the other side comes from Enhancv, which sells a resume builder. Its page, published on November 3, 2025 and updated on September 30, 2026, reports “25 in-depth interviews with U.S.-based recruiters and talent acquisition professionals conducted by Enhancv between September and October 2025”.

Its main number: “Of those recruiters, 23 (92%) said their systems do not auto-reject resumes for formatting, content, or design.” The remainder: “Only 2 of 25 recruiters (8%) said their ATS automatically rejects applicants for reasons other than knockout questions.” The page puts the filtering elsewhere: “Knockout questions are the real filters: 84% of recruiters rely on them to screen for work authorization, certifications, or location”. It also reports that “44% of recruiters said their ATS includes an AI or “fit” score”.

The same page describes how an application can go unread without being rejected. A score can bury it: “Low score ≠ auto-reject, but in high-volume roles, it can push you to the bottom of the pile.” And it says in its own voice that many resumes are never read: “The reality is that even in companies that promise “every resume reviewed,” many never get seen.” One interviewee, a vice president of human resources, estimated that a single question about residency and visa sponsorship knocks out “probably 30%” of applicants for the technical roles the company was then posting. That is one person’s estimate about one employer.

Enhancv states the first limit itself: “The sample size is intentionally small but rich in qualitative depth. These findings represent consistent themes across recruiters rather than nationally representative statistics.” The figures are what recruiters said about their systems. Nobody inspected the systems, and the page does not say how the 25 were chosen. The article also recommends Enhancv’s templates in its own text. One thing we could not square: the page says the two recruiters with auto-rejection use Bullhorn and BambooHR, and elsewhere it describes a recruiter on Phenom who applies a score threshold to reject low matches. It does not reconcile the two.

Enhancv’s interviews support the statement that most of these 25 recruiters said a person does the rejecting. They cannot tell you what happens at the employer you applied to.

What ATS vendors’ own documents say

The third kind of evidence is what the vendors publish themselves, which we read on the dates in the source list. These pages show what an employer can switch on, and none of them says how many employers do or how many applicants a rule removes.

We have also written a separate page on each of the five vendors, built from its public documentation: Workday, Greenhouse, Lever, iCIMS and Ashby.

Rules an employer sets can reject you automatically

Greenhouse documents such a rule under the name auto-reject: “With auto-reject, your organization can set up application rules so that, based on an applicant’s answer to a question, they will automatically be rejected as a potential candidate.” The same support article adds: “Anyone set up to receive notifications about new candidate applications will not be notified of auto-rejected applicants.”

Ashby’s knowledge base has a feature of the same name: “Auto-Reject allows you to reject candidates whose application submissions match certain conditions automatically.” The employer picks a question on the application form and the answers that trigger the rule, and the rules “can only be configured to run at the point of the candidate’s application submission”. The page also explains how a recruiter can filter for auto-rejected candidates to review them.

Lever, in a blog post from December 2021, put it in terms of knockout questions: “Knockout questions defined by recruiters and hiring managers auto-eliminate prospects who don’t meet the bare-minimum criteria for roles (e.g., a certain number of years of experience or certifications).”

The iCIMS line is of a different kind. It comes from a glossary entry on applicant tracking systems, which describes the category and is not a specification of iCIMS’s own product. The entry lists three things such software lets a company do, and the third is “Automatically filter out candidates who are not viable”.

In the Greenhouse and Ashby documents, the rule reads an answer on the application form. Neither page describes a rule that reads the wording or layout of the resume.

Features that sort or rate applicants

Greenhouse’s Talent Matching FAQ says the feature uses algorithms to “sort the candidate pool into match categories (such as Strong, Good, Partial, Limited, or Needs manual review)” against criteria the employer sets. It adds: “Talent Matching does not auto-reject or auto-advance any candidate, including candidates in lower match categories or those labeled “Needs manual review.” All disposition and hiring decisions are made by the hiring team.” A resume with “rare formatting or technical issues” that the feature can’t analyze is labeled “Needs manual review”, and those candidates “must be reviewed by your team”.

Workday wrote in a post from June 2025: “Workday’s AI recruiting tools provide insights on how well a candidate’s qualifications match the requirements for a posted job.” And: “Workday AI does not make hiring decisions.”

Ashby’s AI page says: “The AI never ‘ranks’ or gives numerical ratings to applicants, a human must always be involved in decision-making.” Its knowledge base describes more than that sentence suggests. The article on AI-Assisted Application Review tells reviewers they can add an “AI Job Criteria Met Percentage” column and “sort by this column to rank candidates from best fit (highest percentage) to least fit (lowest)”.

Lever’s 2021 post argues that “artificial intelligence isn’t an ATS feature that can or should make important decisions regarding who talent teams should or shouldn’t move forward with”. The same post names “auto-archiving unqualified individuals” among the ways AI can help.

What one US law says about automated screening

New York City sets conditions on the use of some automated hiring tools. Its Department of Consumer and Worker Protection (DCWP) says Local Law 144 of 2021 “prohibits employers and employment agencies from using an automated employment decision tool unless the tool has been subject to a bias audit within one year of the use of the tool, information about the bias audit is publicly available, and certain notices have been provided to employees or job candidates.” DCWP set July 5, 2023 as the start of enforcement.

DCWP’s FAQ defines an automated employment decision tool as a computer-based tool that uses machine learning, statistical modeling, data analytics or artificial intelligence, helps employers make employment decisions, and “Substantially assists or replaces discretionary decision-making.” Screening counts: “The Law defines employment decision more broadly to include screening.” The law applies only to an employer or employment agency that uses such a tool “in the city”. There, job candidates who are New York City residents must be told that a tool is used and which “job qualifications or characteristics” it will assess, 10 business days before use. The notice can be in the job posting, sent by mail or email, or placed on the employment section of the employer’s website.

Neither DCWP page gives a figure for how often such tools reject anyone. The FAQ does say that the published summary of a bias audit must include “the selection or scoring rates” and impact ratios for all categories, so rates for individual tools exist in those summaries. We did not collect them. Both pages define the law’s reach by what a tool does, not by the name applicant tracking system.

What this means for your application

This section is our view.

The automatic rejections that vendors document run on answers to questions the employer wrote, so treat the application questions as the gate. Read each one and answer it accurately. If a truthful answer fails a requirement, rewording your resume will not change the outcome of that rule.

The sorting that vendors and the Harvard report describe compares a resume with criteria the employer set. Where a requirement in the posting is true of you, say so in plain words that are easy to find. Don’t add what you can’t back up: an interviewer can ask about any line, and our guide to resume-based interview questions covers how those questions are built.

Keep the file simple enough to parse. The 2011 Bersin test and Greenhouse’s note on “formatting or technical issues” both show that software can fail to read a resume, and a plain layout with standard headings lowers that risk.

Don’t chase a number. A checker’s score belongs to the checker. Ours does too: the resume score in our resume builder is our own measure, not a score from any applicant tracking system, and no employer sees it. Its rules are published on that page, which also collects vendor statements in its ATS section.

Then prepare for the interview that the resume earns. Our free resume to interview questions tool lists the questions your resume is most likely to raise.

The limits of this page

Neither side of this argument has a measured rate. The 75% figure has no method that we could find. The best evidence we found that software filters people out is a survey of what executives believe, and the best evidence we found against blanket auto-rejection is 25 interviews published by a company that sells a resume builder. We found no vendor page that reports how many applicants its customers’ rules reject.

Vendor documentation changes without notice, and so do the products. If an employer’s own application page or notice tells you something different from this page, go by what the employer says.