Step-by-step guide AI Job-Search Fast-Track: experienced professionals (3–15 years) →
No. There is no study behind the claim that applicant tracking systems (ATS) reject 75% of CVs, or resumes, before a person sees them. The figure goes back to a 2012 magazine article that credited it to a company selling resume optimisation, with no sample and no method. What the screening vendors publish about their own software describes grading and sorting, with a person making the decision. Automatic rejection is documented in one place: your answers to the questions on the application form. Employers also set filters of their own, which is the part most myth-busting articles leave out. Here is what's on the record, as of November 2026.
Where the 75% figure comes from
We followed the number back as far as it goes. On 1 March 2012, CIO.com published an article about getting past applicant tracking systems. It said the systems "kill 75 percent of candidates' chances of landing an interview as soon as they submit their resumes, according to job search services provider Preptel" (CIO.com, archived copy). The article names no study, no sample and no method. Much of the rest is advice from Preptel's chief executive on how to format a resume for these systems.
Preptel was not a neutral observer. Its homepage from May 2012 sold expert resume analysis and promised to show job seekers how a hiring company would evaluate their resume. So the best-known statistic about screening software began as a claim by a company that sold the cure. We could not find an original study behind it, and none of the articles we saw repeating the number links to one.
The counter-claim deserves the same care. Many debunking articles cite a survey by Enhancv, which sells a CV builder. Of the 25 US recruiters it interviewed in September and October 2025, 23 said their system does not reject applications automatically for formatting, content or design, and 2 said theirs was set up to reject on content (Enhancv, updated 30 September 2026). Enhancv says itself that the sample is not nationally representative. Twenty-five interviews show what some recruiters see. They don't give a rejection rate, and the 92% that gets quoted from them is no more a law of nature than the 75% was.
What the vendors say their software does
A better source is the documentation vendors write for the employers who pay them. We read three sets on 10 November 2026: a Workday HiredScore release note from August 2026, Greenhouse's Talent Matching article, updated on 5 October 2026, and Ashby's page on its AI features.
| System | What its documentation describes | Who rejects |
|---|---|---|
| Workday (HiredScore) | Grades each applicant A to D for one job. Basic qualifications decide between A or B and C or D. Since August 2026, employers can opt in to an AI model that checks each qualification separately against the parsed CV and shows the recruiter which were met. | The release note describes grading. It does not mention rejecting anyone. |
| Greenhouse (Talent Matching) | Sorts applications into strong, good, partial and limited matches against criteria the employer defines and weights. Where the employer allows it, a candidate can ask for a manual review instead. | Greenhouse states that the feature does not advance or reject candidates automatically. |
| Ashby | Checks each application against criteria the recruiter writes. Ashby says its AI never ranks applicants or gives them numerical ratings. | Ashby says a person must always be involved in the decision. |
Two cautions. These are three vendors among many, and every employer configures its own system. And being sorted is not the same as being read. A recruiter with several hundred applications is likely to start with the top grades and may never reach the rest. Greenhouse wrote in February 2026 that recruiters handle nearly three times as many applications per role as in 2021. In its October 2026 survey of 1,200 US job seekers, 81% had been rejected in the past year without speaking to anyone at the company. From the outside, the bottom of a sorted pile feels exactly like a robot saying no. The cause is different, and so is the fix.
Where automatic rejection is real: the form
Most systems let an employer attach a rule to a question on the application form. These are usually called knockout questions, and they cover things like right to work, location, a required licence or notice period. Two vendors spell it out:
- Greenhouse documents an Auto-Reject rule that screens and rejects candidates on their answers to custom application questions. The employer chooses whether a rejection email goes out (Application Rules Overview, updated 2 March 2026).
- Workable documents knockout rules for yes-or-no questions. A "No" to any one of them disqualifies the application. The candidate still gets the usual confirmation email and is not told. Recruiters see disqualified candidates in a separate tab and can restore them (Workable Help).
Notice what these rules read: the box you tick, not the CV you upload. A rejection that arrives minutes after you apply most likely starts here. Three habits follow.
- Answer every form question truthfully and completely. If you would need a visa and the ad says the employer sponsors, say exactly that.
- Make your CV say the same thing. If you would relocate, put it on the form and in your CV header.
- If you answered wrongly by mistake, write to the recruiter. Workable's own page shows that a person can reverse a disqualification.
Never tick a false "yes" to get through. It puts a false statement on record with the employer, and answers get checked later.
The filters employers set themselves
Here is the evidence the debunkers tend to skip. In September 2021, Harvard Business School and Accenture published Hidden Workers: Untapped Talent, based on a survey of 2,275 executives in the United States, the United Kingdom and Germany. More than 90% of the employers said they used their recruiting management system to filter or rank candidates at the start. And 88% agreed that qualified high-skills candidates were screened out because they did not match the exact criteria in the job description. For middle-skills candidates, 94% agreed. Almost half filtered out middle-skills candidates whose work history had a gap of more than six months.
The employer survey dates from early 2020, and practice may have moved since. But it settles the argument in a more useful way than either statistic does. Software does remove qualified people, by rules the employer chose: a required credential, an exact match with the job description, an unexplained gap. Not a hidden score for your choice of font.
The same goes for search. In a 2025 survey of 384 recruiters by Jobscan, which sells CV tools, 76.4% said they filter candidates by skill keywords and 55.3% by job title (Jobscan). The words on your CV matter because people search with them.
What the law asks of employers
Rules on automated screening are arriving, and they point the same way: tell candidates, and keep a person involved. This is a summary, not legal advice.
- New York City. Local Law 144 bars employers from using an automated employment decision tool unless it has had a bias audit within the past year, information about the audit is public, and candidates have been given notice. The city has enforced it since 5 July 2023 (NYC Department of Consumer and Worker Protection).
- The EU AI Act. Annex III lists AI used to analyse and filter job applications and to evaluate candidates as high-risk. The European Commission's own summary, updated on 19 October 2026, says the rules for these areas apply from 2 December 2027. Under Article 26, employers who use such a system must put a competent person in charge of oversight and tell people that it is being used on them.
- The GDPR. Article 22 gives you the right not to be subject to a decision based solely on automated processing that significantly affects you, with exceptions. Where an exception applies, you keep the right to have a person step in, to give your view and to contest the decision. Whether a particular screening step falls under it is not settled, but you can always ask an employer whether a person reviewed your application.
What to do with this
If no robot is binning three CVs in four, the work moves from tricks to three things you control.
- The form. Treat it as the gate it is. Read each question against the ad's requirements before you answer.
- A file that reads cleanly. One column, standard headings, real text. When we tested this for our guide, standard text extractors found five problems in a two-column template, including the name being read after the sidebar and a date drifting away from its job. The one-column version of the same CV had none.
- Proof for each requirement, in the employer's words. That is what every system in the table is built to look for. In our test on a fictional CV, the ad's terms that the candidate could truthfully claim and that appeared in the CV went from 5 of 18 to 14 of 18 after tailoring, with nothing untrue added. Our article on tailoring a CV with ChatGPT or Claude has the routine.
What to ignore: a universal "ATS score". Each system compares your CV with one employer's list for one job, so a checker's score is a rough guide at best. And skip hidden keywords in white text. That trick is detectable, as we explain in what employers can and can't tell about AI on your CV.
If your CV lists duties and you're not sure where the proof would come from, start with how to quantify achievements when you have no numbers. The full method, from the evidence bank to the form answers and interview stories, is in The AI Job-Search Fast-Track, experienced professionals edition: 23 pages, eleven tested prompts and a checklist, built on a worked example we ran in ChatGPT and Claude.
No method can promise an interview. What you can do is make sure the gate you actually face, a form, a sorted list and a busy reader, sees what you've really done.
Frequently asked questions
Do applicant tracking systems reject CVs automatically? Mostly not on their own. The documentation we read from Workday, Greenhouse and Ashby describes software that grades or sorts applications against one employer's criteria and leaves the decision to a person. The automatic rejections that are documented come from knockout questions on the application form, such as right to work, location or a required licence.
Where does the claim that 75% of resumes are rejected by ATS come from? The earliest source we could find is a CIO.com article of 1 March 2012, which credited the figure to Preptel, a company that sold resume optimisation. The article gives no study, sample or method, and we could not find an original study behind the number.
What is a knockout question? A question on the application form with a rule attached, for example about right to work, location, a licence or notice period. Greenhouse documents an Auto-Reject rule based on answers to such questions, and Workable documents that a No to a knockout question disqualifies the application. The rule reads the answer you tick, not your CV.
Is there an ATS score I should aim for? No single one. Each system compares your CV with the criteria one employer set for one job, so a score from a CV checker is a rough guide at best. Visible proof for each must-have in the ad, in the employer's own words where they are true for you, is what the matching tools are built to look for.
Can I ask whether a person reviewed my application? Yes. In the EU, Article 22 of the GDPR limits decisions based solely on automated processing that significantly affect you, and where an exception applies you can ask for a person to step in. In New York City, employers must give candidates notice before using an automated employment decision tool. Wherever you live, it is reasonable to ask a recruiter how applications were reviewed.