What Applicant Tracking Systems Actually Do

If you have looked for a job recently you have been told that applicant tracking systems reject most applications before a human sees them, and that your task is to defeat the software. For a workplace-side view of how operational data is analysed, see this guide.

That premise is largely wrong, the advice built on it is mostly wasted effort, and knowing what actually happens changes what you spend your time on. For an independent public reference, consult NIST AI Risk Management Framework.

Where the "75%" came from

The figure that anchors the entire genre — that these systems reject around three quarters of resumes automatically — has no research behind it.

It traces to a 2012 sales pitch from a company that has since gone bankrupt.

One recruiter who went looking for the evidence described finding no conclusive statistical support for the premise, only conjecture, and concluded that these systems may weigh, sort or filter, but that most or all resumes are reviewed by a person.

When recruiters were asked where they first encountered the claim, 68% said it came from job seekers repeating viral social media posts, and around a fifth pointed at resume-writing services and career coaches marketing "ATS-optimised templates" through fear-based messaging.

The myth is profitable, which is why it persists. An entire industry sells the cure.

What recruiters say actually happens

A study conducted between September and October 2025 interviewed 25 United States recruiters at companies from 120 to more than 50,000 employees, covering Workday, iCIMS, Greenhouse, Bullhorn, BambooHR, SuccessFactors, Teamtailor, Phenom, Lever and LinkedIn Recruiter.

Every recruiter used knockout questions for compliance. Only 8% — two of the twenty-five — had configured content-based automatic rejection, and even then only against strict criteria such as a match below 75% or fewer than seven of ten required skills. The other 92% rejected manually or through knockout questions alone.

The authors state the limitations plainly: a small though qualitatively rich sample, reflecting consistent themes rather than national statistics, with a 90% confidence interval of 2–21% for auto-rejection.

Two notes on the source. It was produced by a resume platform, which has a commercial interest in this area — though the finding cuts against the fear-based marketing that sells "ATS optimisation" tools. And twenty-five interviews is a modest sample. Treat it as strong evidence against the 75% claim rather than as a precise measurement.

So what does the software do

It collects applications and stores them in a database.

It parses your CV into structured fields — contact details, employment history, education, skills — so recruiters can search and filter.

It lets recruiters query, by keyword, by filter, or as a ranked list.

Increasingly it ranks, with AI assistance, and presents an ordered shortlist.

The "T" stands for tracking. These are pipeline management tools, built to help recruiters handle volume.

What genuinely does stop your application

Four mechanisms, and only one of them is about your CV's content.

Knockout questions

Every recruiter uses them. Binary eligibility checks answered in the application form: work authorisation, required certification, minimum years of experience, location, education level.

These do disqualify automatically, and they are the real automated filter.

What follows from that: answer them accurately and read them carefully. An application that fails a knockout question never reaches anyone, regardless of how good the CV is.

Volume

The reason most applications go unanswered is not software. High-demand roles attract 400 to more than 2,000 applicants within days, and recruiters run keyword searches, review the top results, and cannot physically read every application.

Your CV was probably never opened. Not rejected — not reached.

This is the actual problem, and it is a different problem from the one you were told you had.

Timing

More than half of the recruiters interviewed said that applying within the first two or three days after a posting goes live substantially improves the chance of being seen, because after that they are already deep into interviewing the earliest strong applicants.

Roles are frequently paused or unposted once a strong early batch has been collected.

Which makes speed one of the highest-return things you control. See applying early.

Incomplete or blocked applications

A required field left blank, a mandatory attachment missing, or a requisition that closed minutes before you submitted. From your side these look identical to rejection.

The formatting myth, and where it came from

You have been told that a two-column layout, a table, a header or a particular font will break the parser.

This originated with a few legacy systems that used optical character recognition to generate an HTML preview of an uploaded CV. A recruiter relying on that preview instead of opening the original file could see a perfectly good CV rendered as garbage. Job seekers noticed, started searching for how to get past the ATS, and an industry formed to sell the answer.

Recruiters can always open your original uploaded file.

What is still true: parsing works better on simple structure, and a CV that parses cleanly is easier to find in a keyword search. That is a reason to keep the layout straightforward — not a reason to fear rejection over a font.

See the formatting rules that matter.

What to do instead

Given the above, the effort reallocates.

Answer the screening questions accurately, and read them before you start. They are the automated filter.

Apply early. Within days of the posting, where you can.

Make it easy to find in a search. Recruiters search by keyword against the parsed record, so the words that describe your actual work should appear in it. That is different from stuffing keywords — it means naming what you did in the terms the field uses. See tailoring a CV.

Write for the person who spends thirty seconds on it, because your competition for attention is 400 other applications rather than an algorithm.

Use referrals where you have them. A referred application enters a different queue. See referrals.

Apply to fewer, better-matched roles, and put the saved time into the ones that fit. See volume versus targeting.

What to stop doing

Stop paying for "ATS score" checks. They score you against a model of a system that does not work the way the model assumes. See why ATS optimisation tools sell you a problem.

Stop keyword stuffing, and never use white text or hidden keywords. Recruiters find them, and the response is not sympathetic.

Stop stripping your CV of anything readable in the belief that a parser demands it.

And stop attributing silence to the software. It is usually volume, timing or fit, and each of those has a different response.

The short version

The 75% claim comes from a 2012 sales pitch, and no research supports it.

Recruiters report that 92% reject manually or through knockout questions, with only 8% configuring content-based auto-rejection.

Knockout questions, volume and timing are what actually stop applications — and only the first is automated.

The formatting panic came from legacy previews, and recruiters can open your original file.

Apply early, answer the screening questions carefully, and write for a person with thirty seconds — not for a machine that mostly is not doing what you were told.