What Your Resume Looks Like to the Robot Reading It First

Last month a candidate sent me a resume that looked sharp. Two columns, a shaded skills sidebar, a little icon next to each phone and email line. Clean in Word. Clean as a PDF. She’d applied to 61 jobs and heard back from one.

That’s not bad luck. It’s a parsing failure, and it’s the single most common reason a qualified person never gets a callback. Nobody at the company read that resume and passed on her. Nobody read it at all. Here’s how ats parses your resume before any human does, and where that specific layout falls apart.

How ATS Parses Your Resume, Step by Step

An applicant tracking system doesn’t read your resume the way a person does. It runs the file through a pipeline, and each stage hands off a slightly degraded version of your document to the next one. First comes text extraction: the system pulls raw characters out of your file, using Optical Character Recognition when it has to and a direct text layer when it can. Jobscan’s breakdown of ATS mechanics describes this stage plainly: the parser extracts a text layer, then tokenizes it and runs Named Entity Recognition to label chunks as a company, a date range, a job title, or a skill.

That NER step is the part almost nobody talks about. The parser isn’t reading your resume for meaning. It’s guessing which words belong to which category based on position and pattern, the same way spellcheck guesses whether a word is a typo. Put your dates in an odd order, or bury your job title inside a paragraph instead of a heading, and the guess goes wrong. Your ten years at a company can get logged as zero. I’ve watched it happen to a resume that read perfectly well to a human.

The Fields It’s Actually Trying to Fill

Once the text is tokenized, most systems try to slot it into a fixed set of database fields: name, contact info, most recent title, employer, dates, education, and a skills list. This isn’t a stylistic preference. It’s how the ATS lets a recruiter later search 400 resumes for candidates with 3+ years and a specific certification, in under a second, without opening a single file by hand.

If your resume doesn’t map cleanly onto those fields, you don’t get flagged as ineligible. You get logged as incomplete. A recruiter running a filter for people with a listed job title never sees you, because as far as the database is concerned, that field is blank. I’ve seen this happen to genuinely strong candidates more times than I can count, and it’s almost always invisible to them until someone points it out. They assume the silence means rejection. Often it means the system never filed them anywhere at all.

Standard section headings matter here more than people expect. A parser is far more confident labeling a block of text as your work history when it sits under a heading that literally reads Work Experience or Professional Experience than when it sits under something creative like Where I’ve Been or My Journey. Save the personality for the interview. The heading’s only job is to be recognized.

What Silently Breaks the Parse

Here’s the part that actually matters if you’re editing a resume this week. A handful of formatting choices look completely normal to a human and scramble or erase content for a parser.

Multi-column layouts are the biggest offender. A parser generally reads left to right, top to bottom, across the full width of the page. Put your work history in a left column and your skills in a right column, and the parser can read straight across both, interleaving your job titles with your software list into a paragraph that means nothing. Tables cause a related mess: some parsers skip table content entirely, so a resume built as a table can render as almost blank text underneath the polish.

Headers and footers are the second trap. Contact information placed in a Word header often never reaches the parsed text at all, because many parsers don’t scan headers or footers by default. That one habit, popular because it looks tidy, can mean the phone number and email never make it into the applicant record. A recruiter who wants to call you has no way to.

Then there’s the smaller stuff that adds up: icons standing in for section labels instead of the words Skills or Experience, text boxes (frequently skipped outright), unusual fonts that OCR misreads character by character, and images of text, which have no text layer for anything to extract. None of these are dealbreakers on their own. Stack three or four together, which is exactly what a trendy resume template does, and you’ve built something that looks great and reads as almost empty.

File Format Still Matters, Even in 2026

PDF versus Word remains an open question, and it shouldn’t be. Word documents parse the most predictably because their internal structure is simpler for a parser to walk. Most current systems can read PDFs too, but PDF parsing quality varies a lot by vendor, and a PDF exported from a design tool rather than Word’s native save-as-PDF is far more likely to carry layout artifacts that confuse extraction. If a job posting doesn’t specify a format and gives you a choice, .docx is still the safer default. Not because PDF is inherently broken, but because the failure modes are less common and easier to predict.

I don’t think this settles the debate for everyone. Some large employers have genuinely good PDF parsers, and in those cases the format choice barely matters. The trouble is you rarely know which kind of parser sits on the other end of a job board’s upload button, so defaulting to the safer format costs you nothing and occasionally saves you.

Why This Filter Exists at All

It’s tempting to treat ATS software as an obstacle invented to keep good candidates out. That’s not quite fair, even if the effect can feel that way. Applicant tracking systems exist because the volume math stopped working for humans years ago. A single corporate opening can pull 250 applications, and a recruiter covering a dozen open roles simply can’t read all of them by hand. According to Jobscan’s research, 97.8 percent of Fortune 500 companies now run some form of ATS, a figure that’s held steady across multiple years of tracking. The filter isn’t going anywhere. The real question is whether your resume’s built to survive it.

I’ll admit I’m not fully sold that keyword-matching is a good proxy for fit, even where it’s technically working as designed. A parser can confirm you used the phrase project management without confirming you were any good at it. That’s a real limitation of the system, not a reason to ignore it. We covered the keyword side of this in more depth in our ATS resume tips post, if formatting alone doesn’t explain your callback rate.

A Fast Way to See What the Parser Sees

Copy the text of your resume out of Word or your PDF viewer and paste it into a plain text editor, no formatting, nothing fancy. Notepad works. TextEdit works. What lands on the page is roughly what a parser sees: no columns, no color, no icons, just a linear stream of text in the order your file actually stores it.

If that stream reads as a jumbled mix of your skills list interrupted by job titles interrupted by dates, that’s the exact confusion a real ATS will produce, and it’s worth fixing before you submit anywhere. If your phone number vanished because it lived in a header, put it back in the body. Five minutes with a text editor catches problems that would otherwise cost you weeks of silence. It’s a small test, but I haven’t found a faster way to catch the big mistakes.

Where This Fits Into Actually Applying

Fixing your resume’s structure solves the parsing problem, but it doesn’t solve the volume problem, which is a separate headache once you’re applying broadly. If you’re sending out dozens of applications a week and tailoring each one by hand, LastRoundAI’s Auto-Apply reads each job description and each form field, then fills them from your real résumé, drafting a tailored cover letter and screening answers grounded in what you’ve actually done. It never invents experience, and every application sits in a review queue for your approval before it goes out. We wrote up the full mechanics in how Auto-Apply actually works, and the first-impression side of resumes gets its own treatment in what recruiters see in the first six seconds. It’s a different problem than parsing, but the two show up in the same week for most job seekers, so it’s worth knowing both exist.

None of this makes the process feel less mechanical, and I don’t think pretending otherwise helps anyone. A resume shouldn’t have to survive an NER model’s best guess about which line is a job title. But it does, right now, for the vast majority of roles you’ll apply to this year. Build for the parser first. Then write for the human who eventually gets to read it.