What ATS Software Actually Does, and What Resume Tools Get Wrong About It

By Mustafa Tarabya, founder of CVBooster · Published 2026-08-01 · Updated 2026-09-03

10 min read

A resume checker gives you a 62 out of 100 and a red banner warning that your resume "may not pass ATS screening." So you spend an hour cramming phrases from the job posting into a Core Competencies block until the number turns green. Then you apply and hear nothing, exactly like before. The number moved. Nothing else did. That loop is the product most resume tools are actually selling, and it rests on a claim that is not true: that applicant tracking systems read your resume, assign it a score, and reject you automatically below a threshold. They do not. Knowing what ATS software genuinely does is worth more than any score, because it tells you which formatting rules are real and which are folklore you can safely ignore.

What an applicant tracking system is actually for

An ATS is a database with a hiring workflow bolted onto it. Workday, Greenhouse, Lever, iCIMS, SmartRecruiters, Ashby, Taleo and the rest all do roughly the same core set of jobs. They take an application, extract structured fields from the attached file (name, email, phone, employers, job titles, dates, degrees, skills), store that as a candidate record, keep an audit trail so the company can answer questions about who applied and how they were treated, move candidates through stages like applied, screen, interview and offer, send templated emails, and let recruiters search and filter the pile.

Notice what is not on that list: judging you. The parsing step is a data-entry step. It exists so a coordinator does not have to retype your phone number into a form, and so the record is searchable later. Record-keeping obligations are a genuine driver of ATS adoption in the first place. Companies need a defensible log of applicants, not an algorithm with opinions about your font.

Where the auto-reject myth comes from

You have read the line a hundred times: something like 75 percent of resumes are rejected by the ATS before a human sees them. Trace it and you land on resume-tool marketing citing other resume-tool marketing. No major ATS vendor documents a feature that discards applications scoring below a cutoff, because that is not a feature companies would buy. A system that silently deletes qualified applicants is a legal and commercial liability, not a selling point.

There is one kind of genuine automated filtering, and it is worth being precise about it: knockout questions. Are you legally authorised to work in this country. Do you have an active nursing licence. Do you have at least three years of experience with X. Will you relocate. Those are answered by you, in form fields, before you ever hit submit. Some employers configure them to auto-disposition applicants who answer wrong. That is real, and no amount of resume formatting changes it. It also has nothing to do with parsing your PDF.

The other real thing is much more boring. Recruiters do not read all 400 applications. They search, filter, sort, and work the first page or two of results. Never being read is completely real. Being deleted by a robot that scored your bullet points is not. The distinction matters because the two problems have entirely different fixes.

The "ATS score" is a number your resume tool invented

No applicant tracking system produces a compatibility score and hands it to a candidate. Some platforms do rank or match applicants against a requisition internally, and those signals are recruiter-facing, advisory, and routinely ignored. None of that reaches you, and none of it is the percentage sitting in a checker's dashboard.

What a checker actually computes is its own rubric: string matching between your text and the job description, plus a set of formatting heuristics, weighted however the person who built it decided to weight them. That is a legitimate product. It is just not a measurement of anything external.

Here is the tell. Run one unmodified resume through three different free checkers. You will frequently get three meaningfully different scores, sometimes 20 or 30 points apart, with contradictory advice attached. If they were all measuring the same real-world system, they would converge. They do not converge, because there is nothing out there to converge on.

Tip: Score the same file on two different tools before you trust either one. If the numbers disagree, you have already learned the important thing: you are being graded by a rubric, not by an employer's software.

What genuinely breaks parsing

Parsing failures are real. They are just narrower and more specific than the fear marketing suggests. These are the ones worth your attention.

Text that is not text. A PDF exported as an image, a scan, or a screenshot has no extractable characters. If you cannot select and copy the words in a PDF reader, no parser can read them either. Same problem with contact details rendered as icons only: the little envelope glyph is a picture, and your email address next to it may be one too if the export flattened it.
Multi-column layouts. Some parsers reconstruct reading order from the PDF content stream rather than from what your eye sees. A left sidebar can interleave line by line into your experience section, turning "Senior Analyst" and "Python, SQL, Tableau" into one scrambled blob. Modern parsers handle columns far better than they did five years ago. The problem is you have no way to know which vintage sits behind any given portal.
Headers and footers. In a DOCX these live in a structurally separate part of the file, and some parsers skip them entirely. Put your name and phone number only in the page header and there is a real chance the record arrives with no contact details at all.
Tables used for layout. A table where the job title is in one cell, the dates in another and the bullets in a third can flatten into an order that no longer associates them. Tables are also where dates most often detach from the role they belong to.
Unusual or badly embedded fonts. If a font is subset without a proper character map, extraction can return garbage or drop ligatures. Rare with mainstream typefaces, common with novelty ones.
Non-standard section headings. Parsers segment your document by looking for recognisable heading text. Call your experience section "Where I Have Made a Dent" and the block underneath may end up unclassified, misfiled, or dumped into a generic notes field that nobody filters on.
Inconsistent or informal dates. "Summer '22 to now" is human-readable and machine-hostile. Tenure calculations and employment-length filters run on parsed dates.
Text boxes, SmartArt, charts and skill bars. All of it sits outside the main text flow, and a skill bar filled to 80 percent carries no meaning even when it is extracted.

Tip: The cheapest parse test in existence is not a checker. Open your PDF, select all, copy, and paste into a plain text editor. What you see is close to what a parser sees. If the order is scrambled, your phone number vanished, or your job titles ran into your skills list, you have a real problem. If it reads cleanly top to bottom, you are done with formatting and can go work on the content.

Why keyword scoring rewards the wrong behaviour

This is the strongest criticism of score-selling tools, and it holds up. Because the number rises whenever a string appears, the fastest route to a green score is to paste the posting's skill list into your resume verbatim. The metric rewards repetition and punishes legitimate paraphrase.

Consider two candidates. One writes "stakeholder management" three times because the posting used the phrase. The other writes "ran a weekly steering group with three business owners and the finance lead, and cut approval turnaround from two weeks to three days." The scoring tool prefers the first. Every recruiter alive prefers the second. The tool cannot tell the difference between evidence and echo. A human tells the difference instantly.

The stuffing endgame is the white-text trick, invisible keywords hidden behind a matching background. It does get extracted. It also gets you binned the moment a human opens the file and sees a suspiciously empty region, and it reads as dishonesty rather than cleverness. Do not do it.

The honest version is narrower: use the posting's exact vocabulary for things that are actually true of you, inside the bullet where you describe doing the thing. That satisfies search and survives human reading at the same time.

What actually matters, part one: being findable

Recruiters use the ATS the way you use a search box. They type "Kubernetes" or "CPA" or "Salesforce administrator" or a boolean string, and they look at what comes back. This is where keywords genuinely matter, and it works nothing like a percentage.

Search is binary per term. Your record either contains the string the recruiter typed or it does not. That single fact reorganises most keyword advice.

Exact strings beat clever synonyms. Write the tool the way the market writes it. "Power BI", not "Microsoft's BI tooling". "Kubernetes", not "container orchestration" on its own.
Include the acronym and the expansion once each. "Search engine optimization (SEO)". "Certified Public Accountant (CPA)". You do not know which one the recruiter typed.
Total match percentage is close to meaningless. Hitting 92 percent across 40 terms does not help if the one term the recruiter searched is among the three you left out. Coverage of the few decisive terms beats coverage of everything.
Your record persists. Being in the database for a role you did not get is how people get contacted about a different opening six months later, which is an underrated reason to apply properly even to long shots.

What actually matters, part two: being readable in six seconds

The six-second figure gets thrown around loosely, but the underlying behaviour is not in dispute. The first pass on a resume is a skim, and that skim decides whether a real read ever happens. In it, a recruiter is looking for your current title, current employer, tenure, apparent scope, any decisive credential, and whether the top third of page one plausibly matches the role they are filling.

So design for the skim, not for the parser.

Put a target title near the top that matches the role's vocabulary, where it is true of you.
Most recent role first, with dates in a consistent position the eye can track straight down the page.
Put scope numbers in the first two bullets of your most recent job: team size, budget, transaction volume, user count, ticket throughput. Anything countable.
Two pages maximum. One page if you have under roughly eight years of relevant work.
Accept that the top third of page one carries your entire case. Everything below it is corroboration for a reader you have already convinced.

The checklist, and all of it is testable

Export as PDF unless the posting or the portal explicitly asks for DOCX, in which case send DOCX. Never .pages, never a scan, never an image.
Keep one clean single-column version as your submission file. Keep the designed two-column version for email attachments, networking and your own site, where a human opens it directly and no parser is involved.
Use conventional section names: Summary, Experience, Education, Skills, Certifications, Projects. Spend your creativity on the bullets, not the labels.
Keep contact details in the body of the document, never only in the page header or footer.
Use one date format with four-digit years throughout. Mar 2023 or 03/2023, chosen once and applied everywhere.
No text inside images, no icon-only contact lines, no skill bars or rating dots on the submission version.
Mirror the posting's actual vocabulary where it is genuinely true, inside the bullet that proves it.
Give both the acronym and the full term once for anything a recruiter might search either way.
Name the file Firstname-Lastname-Role.pdf. It lands in the recruiter's downloads folder, and "resume(3).pdf" is a bad first impression.
Run the copy-paste test before every submission. Two minutes, and it catches almost every real parsing failure.

What you can stop worrying about

The exact score any tool gives you, in either direction.
Whether Arial beats Calibri beats Garamond. Any common, properly embedded, widely available font parses fine.
Whether PDF is "banned". It has not been broadly true for years, and portals that want something else say so on the upload screen.
Reaching a 100 percent keyword match. Nobody is measuring that except the tool charging you for it.
Standard bullet characters, bold text and horizontal rules. All of it extracts cleanly.
One-page dogma when you genuinely have 15 years of directly relevant experience.

The honest version of this advice

The pitch that sells is fear: an invisible machine is shredding your application, so buy the score to survive it. The truer version is less sellable. Your resume was almost certainly parsed correctly. It is sitting in a database next to 400 others, and either it is not surfacing in the searches recruiters actually run, or it surfaces and fails to make its case in the first skim. Both problems are fixable. Neither is fixed by a score.

Use tools for the part they do well. A checker that tells you your PDF has no extractable text, or that your dates failed to parse, or that your experience section was not recognised, is reporting something real and useful. A checker that tells you 68 out of 100 is reporting on its own rubric. Take the first, ignore the second, and put the reclaimed hour into the top third of page one, which is the only part of the document a human is guaranteed to look at.

Looking for a worked example for your job title? Browse the resume examples by job title.

Get your free ATS score · All articles