How an ATS actually reads your resume

By UnchartedCareer. Last updated: July 2026.

An applicant tracking system reads your resume by running it through a parser that pulls the raw text off the page and maps it into structured fields like your name, email, phone, job titles, and dates. It does not reject you on that read. When the parse fails, most systems still create your candidate record, keep your resume attached, and let a recruiter fill in the missing fields by hand, so a failed parse costs you clean searchability, not your place in the pile.

Most of what people fear about the ATS is a sales story from over a decade ago. The real risk is smaller and more fixable. A handful of layouts break the machine read before any human sees you, and almost nothing else on the page decides whether the parser can read you.

We tested that reading step ourselves instead of repeating what vendors claim. In a June 2026 test of 48 synthetic resumes read by three local open-source text extractors (pdfplumber, pdfminer.six, python-docx), a clean single-column resume lost 0 percent of its fields, while a resume whose text was saved as a picture lost 37.5 percent and dropped the candidate's name, email, and phone on every read. Read those numbers as a floor on what can go wrong in the reading step, not a verdict on any named ATS. Layout is the lever, and you can see the full method and data in our ATS Autopsy study.

What is an ATS, and what does it do with your resume?#

An ATS is a database with a recruiter workflow bolted on top, per Workday's own product documentation. It collects your application, files it under the job requisition, and lets a recruiter search the pile, sort it, and move people through stages.

The parser can fail, and a failed parse is not a rejection. When a resume fails to parse in Greenhouse, the candidate record still gets created, your file stays attached, and the recruiter enters the missing details by hand (Greenhouse, 2026). You lose clean searchability, not your spot in the pile. That gap, between a parse that failed and a decision that rejected you, is where most of the fear lives and where most of it dissolves.

Where did the "75 percent of resumes get auto-rejected" number come from?#

You have heard the number. Some version of "75 percent of resumes are rejected by an ATS before a human ever sees them," sometimes cited as 70 percent, sometimes inflated to 88 percent. Career sites, recruiters, and resume tools repeat it as settled fact. It was invented.

The figure traces to a 2012 sales pitch by Preptel, a resume-software vendor with an obvious incentive to make the machine sound scary. Preptel went out of business in August 2013 and never published a study or a method behind the number, as documented by Ask a Manager (2020). The chain runs through an uncited Forbes mention and a wall of resume blogs, and every link dead-ends at the same defunct vendor with no source data. The number drifts between 70, 75, and 88 percent because there was never an original to anchor it.

When people defend the stat, they reach for real research: the Harvard Business School and Accenture report "Hidden Workers: Untapped Talent," from September 2021. It does say software filters at scale, with more than 90 percent of surveyed employers using a recruiting system to make a first cut or rank applicants (Harvard Business School and Accenture, 2021). But read what it blames. It found that 88 percent of employers agree qualified high-skills candidates get screened out because they do not match the job description's exact criteria, and it pins that on rigid, recruiter-configured filters like employment gaps and hard degree requirements, not on a parser auto-rejecting resumes sight unseen (Harvard Business School and Accenture, 2021). The report's own fix is to pick six to eight minimum skills that filter applicants in. That is a story about how recruiters set must-haves, not a robot deleting three quarters of humanity.

What actually breaks a machine read of your resume?#

Here is what the June 2026 test showed, layout by layout, all of it from three local open-source text extractors (pdfplumber, pdfminer.six, python-docx) and read as a floor on the reading step, not a verdict on any named ATS.

Start with what is safe. Single-column resumes lost 0 percent of their field reads. Nonstandard section headings lost nothing either, so calling a section "Where I've worked" instead of "Experience" cost zero, because the parser reads the content underneath the label, not the label itself. Two-column resumes came through nearly clean at 1.4 percent loss. If your design is one of these, the reading step is not your problem.

Now the layouts that actually break. Text saved as an image lost 37.5 percent of its field reads, the worst result in the test, and on those resumes the extractors dropped the candidate's name, email, and phone on 100 percent of reads, because a picture of text leaves nothing for a text extractor to read. These extractors run no OCR, and some commercial systems do, so treat the image figure as a worst case rather than a certainty. Contact details parked in a page header lost 25 percent of fields, and the same details in a footer lost 25 percent, because the extractor often reads the body and skips those bands. An unusual or embedded display font lost 20.8 percent when its glyphs came out as garbage characters. Three side-by-side columns lost 14.6 percent as the reading order interleaved and the employer got separated from the job title. Work history crammed into a bordered table lost 12.5 percent when the cells collapsed into the wrong fields. Greenhouse's own documentation names the same culprits: multi-column layouts, tables, headers and footers, contact info in a text box, graphics, and image-based files (Greenhouse, 2026).

Now the rule everyone repeats, and the way the test breaks it. Across every layout, PDF files lost 19.9 percent of field reads and Word files lost 0 percent. Read fast, that looks like proof you should never send a PDF. Read honestly, it says the opposite. Almost all of the PDF loss came from the hard layouts above, not from the file type, and a clean single-column text-based PDF lost nothing. The old "always PDF, never Word" rule and its mirror image both miss the real failure mode, a resume saved as an image instead of as text. Greenhouse recommends PDF for best results, treats Word as equally acceptable, and flags the image upload as the thing that breaks (Greenhouse, 2026). The format war is a distraction. The file being a picture is the catastrophe.

What auto-rejects you, and what only ranks you?#

If a rejection email lands within minutes of your application, the parser did not send it. A recruiter-configured knockout question did. Do you have work authorization? The required license? Do you meet the minimum years of experience? Answer in a way that fails the rule and the system fires the automated rejection, per Built In's reporting (2025). That is a recruiter drawing a hard line in software, fair to call automated, but it is an opt-in setting per job post, not a default, and it has nothing to do with the parser reading your layout. In most ATS workflows a human still has to act to reject you at all.

The newer AI scoring layer is the one people confuse with a gatekeeper. Some systems rate your resume against the job description and hand back a match number or a band, and recruiters often work the list top-down. But these layers rank, they do not delete. Greenhouse states that its Talent Matching is assistive AI that does not auto-advance or auto-reject anyone, and that recruiters and hiring managers stay responsible for every hiring decision (Greenhouse, 2026). A low score buries you in the sort order. It does not remove you from the pile. The fix for a low score is relevance, not panic.

Do white-font keywords and free ATS checkers work?#

Stuffing keywords does not work, and the white-font trick actively hurts you. When an ATS parses your resume it strips the formatting and reads only the underlying text, so keywords you hid in white font are not invisible. They are ingested as plain text and show up in the parsed output the recruiter reads, per Built In (2025). Recruiters catch it with a select-all highlight, and some systems auto-flag the manipulation, so the trick meant to sneak you in gets you cut instead. A wall of repeated terms, hidden or not, reads as exactly what it is.

Free "ATS checker" sites deserve the same skepticism. The popular open-source ats-screener project states plainly that its scoring is an approximation from public documentation and community reports and does not reflect the actual proprietary algorithm of any platform (ats-screener, 2026). Run one for a rough smoke test, but do not treat its score as the verdict a real recruiter's system would return.

How do you tailor a resume without rewriting your life?#

Tailoring is not inventing a new persona for every posting. Pull the must-have terms from the job description, find where each one is genuinely true of your experience, and write that line in their phrasing. If the posting says "incident response" and you ran incident response, write "incident response," not your team's internal nickname for it. You are removing the translation step between what you did and what the recruiter typed into the search box.

But findable comes before relevant. Candidates invert this constantly, fussing over keywords while their contact details sit trapped in an image the parser cannot read. Being in the database beats being a blank record nobody can search.

How can you check your own resume in ten minutes?#

Ten minutes, a few checks, no tool required.

The parse test. Copy your whole resume and paste it into the plainest text editor you own, the one with no formatting. Read what lands. If your name, email, phone, and every job title come through in the right order, a parser will manage the same. If the contact line vanishes, or two columns weld into a word soup where "Senior nurse" sits three unrelated words away from "Mercy General," you have a layout problem a human never gets to see past.

The picture test. Try to select your name and contact details with your cursor. If they highlight, that is real text and a parser can read it. If your name is part of an image and will not highlight, it is invisible to a text extractor, the single worst thing you can do to a resume.

The six-second test. Hand your resume to someone, let them look for six seconds, then take it away. Ask two things: what role am I going for, and what makes me plausible for it? If they cannot answer both, the recruiter working a packed queue will not either.

Then the worksheet, top to bottom. Findable: one column, standard section order, contact details in the body of the page and never in a header, footer, or image, saved as a text-based PDF or Word file and never as a scan. Relevant: the posting's top five must-haves each appear once, in the recruiter's words, only where they are true, with a summary line that names the target role. Defensible: every bullet could become an interview question and you have the number or the story ready to answer it. Three passes, in that order, and do not move on until the one before it is clean.

You can run every check above by hand, and you should at least once, because it teaches you exactly what the reader is blind to. When you want the two-minute version that flags which fields survived the read and scores your text against a specific posting, run your file through UnchartedCareer's free ATS resume scan. Same logic as these tests, faster.

How do we know this?#

The ATS Autopsy is our own original test: 48 synthetic resumes from three fabricated personas, with no real people, no personal data, and reserved example.com emails, across nine layouts, each saved as PDF and Word, read by three local open-source extractors (pdfplumber 0.11.4, pdfminer.six 20231228, python-docx 1.1.2). That produced 600 scored field reads on 2026-06-24, reproducible from a clean checkout because the generator and scorer are deterministic. We did not test any named commercial ATS and make no claim that any product rejects or discriminates against a candidate. These extractors run the same text step those systems run first, so the numbers are a floor on the reading step, not a verdict on any named ATS.

The vendor-behavior claims come from primary documentation, mainly Greenhouse's support articles on parse failure and Talent Matching (Greenhouse, 2026) and Workday's description of what an ATS is. The myth's origin is traced through the Preptel record via Ask a Manager (2020) and the Harvard Business School and Accenture "Hidden Workers" report (2021). Nothing here draws on data from UnchartedCareer users.