What an applicant tracking system actually does

What an ATS actually is

An applicant tracking system is a database with a hiring workflow attached. That is the whole thing. It is closer to a customer records system than to anything that could be described as artificial intelligence.

When you press submit, roughly this happens: the system creates a record, stores your file, tries to read that file into structured fields, checks whether you already exist from a previous application, attaches you to a requisition, drops you into a stage called something like “New,” and makes you findable. Everything else it does is for the employer, not about you: scheduling, email templates, offer approvals, background-check integrations, compliance reporting.

The word that matters there is findable. A recruiter holding four hundred applicants does not read four hundred resumes. They type phlebotomy or forklift or Kubernetes into a search box and read what comes back. Nearly every practical suggestion on this page follows from that one fact.

The systems you will actually meet

“The ATS” is not a thing. There are dozens of products, they behave very differently, and advice that treats them as one machine with one set of rules is guessing.

• Workday. Large employers. Famous for making you create an account and then re-type your entire work history after uploading a resume.
• Taleo (Oracle). Older, still deeply embedded in large enterprises, universities and public-sector-adjacent employers.
• iCIMS. Common at large mid-market employers, including hospital systems, retailers and logistics companies.
• Greenhouse and Lever. Common at startups and tech companies. Short applications: upload, three questions, done.
• BambooHR. A small-company HR suite with hiring bolted on.
• SAP SuccessFactors, SmartRecruiters, Ashby, ADP, Paylocity and a long tail of others.

Notice what the Workday behavior tells you. If a system makes you correct every parsed field by hand, how well it read your PDF barely matters, because you fixed it. Where you never see the parse, it matters much more. Same resume, different exposure — and nobody who says “the ATS will reject this” knows which one you are about to submit to.

The 75% claim, examined

You have read some version of this sentence: 75% of resumes are rejected by an applicant tracking system before a human ever sees them.

I have not been able to find a primary source for it. It appears without a citation, or citing “studies,” or citing an article that itself cites nothing. The number drifts: 70%, 75% and 90% all circulate. Where a figure is attached to a named company, that company usually sells resume services and publishes no method, sample size, or definition of “rejected.”

Be careful with what that does and does not establish. I cannot prove that no such research exists anywhere; a source I could not find is not proof of absence. What I can say is that a statistic repeated for years without anyone naming the dataset behind it has no evidential weight, and you should not reorganize your job search around it.

It is worth asking why this number travelled so far. It is an excellent sales line. It manufactures a villain who is not you, it explains silence you had no other explanation for, and it arrives packaged with the cure: an “ATS-optimized” template, an “ATS scan,” a subscription to beat the bots. “A robot deleted it” is far easier to hear than “three hundred people applied and forty of them have already done this exact job.”

The fear is pointed at the wrong thing. Most applications to a popular posting really are never read closely — that part is true and you can observe it. The mechanism is volume and human triage, not a hidden algorithmic verdict on your bullet points.

Three distinctions worth holding separately, because collapsing them is how the myth survives:

• Software does auto-reject based on screening questions the employer configured. Real, and common.
• Software does rank or score candidates in some systems, and a recruiter chooses whether to look at that ranking. Real, and it is a sorting aid, not a gate.
• Software does not read your resume, form a judgment about its quality, and discard it. That is not a feature these products have.

What really happens after you hit submit

The realistic sequence:

1. Your file is stored and parsed into fields: name, email, phone, employers, titles, dates, education, skills.
2. The system checks for a duplicate. If you applied before, you are usually merged into your existing record, old notes included.
3. Screening questions are evaluated. If a rule is set to disqualify, your status changes immediately.
4. You land in a queue on that requisition, sorted by date, sometimes by a relevance score.
5. A recruiter searches and filters — keywords, location, license, years, availability — and opens what comes back.
6. A shortlist goes to the hiring manager. Everyone else stays in the database, frequently untouched.

Step six is where most silence comes from. Nothing rejected you. A person ran three searches, opened perhaps thirty resumes, filled the interview slots they had, and moved on to the next of the nine requisitions they are carrying. Your application is still sitting there, intact and unread.

That should change what you optimize for. You are not trying to pass a scanner. You are trying to be in the results that come back when someone searches for what they actually need.

Knockout questions: the real auto-reject

Knockout questions, disqualifying questions, screening questions — three names for the one place where software genuinely closes the door without a human involved. They have nothing to do with your resume file. Typical ones: are you legally authorized to work in this country; will you now or in future require sponsorship; do you hold an active license in this state; can you work the posted shift; how many years of experience do you have with a named skill; what are your salary expectations.

The advice is boring and correct: answer them accurately. A false answer that gets you into the process gets discovered at reference or background-check stage, when you have the most to lose. But read them precisely, because people lose roles on careless answers to questions they could have answered well.

One related point, since it worries people: in the US the voluntary demographic and veteran-status questions exist so employers can file reporting they are legally required to file. They are not supposed to be part of screening, and declining is an option offered on purpose.

What parsing is, and what breaks it

Parsing is the step where software converts your document into fields. It reads the file’s text layer and infers structure from position and wording: this line near the top with a capitalized first and last name is probably a name; this bold line above a date range is probably a job title.

When parsing fails you are not rejected. Something quieter and worse happens: your record is incomplete or scrambled, so you do not come back in the searches you should have, and the recruiter who does open you sees a mess. Invisibility, not rejection.

What breaks it, roughly worst first:

• Text inside an image. A screenshot, a scan, or a resume exported as a picture has no text layer at all. This produces an empty record.
• Text in the page header or footer. Many parsers skip these regions. Contact details in a Word header is the classic version: a recruiter who wants to call you cannot.
• Multi-column layouts. Flattening a page into a single stream can interleave the columns line by line.
• Tables used for layout. Cell reading order varies, so a two-column skills table can come out shuffled.
• Data drawn as graphics. Five filled dots next to “Excel” carries no text. It says nothing to a parser and, honestly, nothing to a reader.
• Unusual or custom-subset fonts. Some exports copy out garbled or drop characters.
• Text boxes in Word, which often sit outside the main content flow.
• Invented section headings. A parser looks for words it knows.

The thirty-second test. Open your resume, select all, copy, and paste into a plain text editor. What you see is close to what the parser stores. If your phone number vanished, if your columns interleave, if your bullets became question marks, you have found a real problem. This costs nothing and tells you more than any scanner that scores you out of a hundred.

On file formats. A PDF produced by Word, Google Docs, LaTeX or a normal resume builder has a proper text layer and is accepted nearly everywhere. One produced by a design tool may be equally fine, or may be outlined text and image blocks — run the copy-paste test and you will know. The extension is not the issue; the design is. If the posting asks for a Word file, send a Word file.

What actually helps

Given that a person searches the database, here is what genuinely improves your odds of appearing in their results.

Mirror the posting’s vocabulary for skills you genuinely have. Not stuffing — translation. If you do the work but call it something else, you will not appear when they search their word for it.

Spell out an acronym once, alongside its expansion. You do not know which one the recruiter typed. Give the search both.

Use dates a parser can read. Spelled or numeric months with four-digit years, consistently, in every entry.

The rest, briefly: conventional section headings; contact details in the body; one column; the job title you are applying for somewhere it is true, such as your summary line; accurate screening answers; and the employer’s own careers site rather than a third-party aggregator when you have the choice, because the extra hop is where fields get lost. To work out which of the posting’s requirements are real in the first place, read how to read a job posting.

Hourly, licensed and trades roles

Almost everything written about applicant tracking systems assumes a salaried office job. That is not most of the market. Hospital systems, distribution centers, care homes, retail chains, dental practices and trucking companies push enormous volumes through these systems, and their filters are far more concrete than keyword relevance: license type and state, certification, endorsement, equipment, shift, availability. Those are usually the first cut. If your license details live only in a sentence buried in your third bullet point, you can be exactly qualified and never surface.

One more thing specific to high-volume hourly hiring: the form matters more than the file. You may be applying on a phone or an in-store kiosk where the resume upload is optional but the availability grid, the license dropdown and the shift questions are not. Fill in every field, even where it duplicates your resume. The fields are what the search runs against.

Folklore worth deleting

White text keyword stuffing. It does not work and it is actively dangerous. Search results usually highlight the matched term in the recruiter’s viewer, the parsed record carries no color at all, and pasting the text anywhere exposes it instantly. When it is found it does not read as a clever hack. It reads as an attempt to deceive the employer, in writing, before you have met them.

“ATS-friendly” as a product you buy. There is no certifying body and no ATS vendor approves templates. Anyone can print the words on a page and charge for it. What the phrase describes — single column, real text, conventional headings, no images — is just an ordinary well-made resume, and you can produce that in a word processor for nothing.

A score out of 100. No external tool can see which system an employer runs, how that instance is configured, which requisition you applied to, what the recruiter typed, or who else is in the pile. A score only ever describes the checks that tool ran on the text you gave it. That applies to our own resume checker: the number is a prompt to look at something, not a prediction.

“It must be one page or the system truncates it.” These systems store your whole file. Length is a question about a reader’s patience and your career stage — see how long a resume should be.

Font superstitions. No system rejects Calibri, Arial, Garamond or Helvetica. Choose something legible and stop thinking about it.

What our matcher does and does not claim

The resume and job description matcher exists because of the search behavior described at the top of this page. Here is exactly what it does, so you can judge it.

What it does: it takes the posting and your resume, both pasted by you, and compares the meaningful terms in each. It ignores common filler words, groups what it finds, and shows you which terms carry weight in the posting but appear nowhere in your resume. It runs entirely in your browser; nothing you paste is uploaded.

What it does not do: it does not know which applicant tracking system the employer uses. It does not simulate anyone’s parser. It does not know what the recruiter will type into their search box, or which listed requirements are genuinely required. It cannot predict interviews, and it has no connection to any real ATS score, because no such portable score exists.

Use it as a reading aid: a checklist of things this employer keeps mentioning that you have not mentioned back. Then decide, one term at a time, whether it is true of you. Add the true ones in your own sentences, with the specifics that make them credible. Skip the rest. A resume claiming a skill you do not have does not fail at the database. It fails in the interview, in front of a person.

Then check it reads well to a human in the resume checker, and if the evidence behind your claims is thin, work through how to quantify a bullet when you think you have no numbers.