How ATS Software Actually Works in 2026 (It Is Not What Most Guides Tell You)

"ATS" gets used as a catchall for wildly different software, which is a big part of why the advice about it contradicts itself. An applicant tracking system, as SHRM defines the category, stores applications, lets recruiters search and filter them, and routes resumes to the right hiring manager. Job seekers usually mean something narrower when they worry about "the ATS" — either a fear that a parser mangled their file, or a hope that some number, somewhere, might unlock the process. Neither framing survives contact with how these systems actually work, and mixing them up is the root of most of the folklore below. It also wastes real effort: hours spent chasing a phantom score would be better spent fixing a file that genuinely doesn't parse, or genuinely doesn't speak to the posting in front of it.

The two questions an ATS actually asks

An applicant tracking system is not doing one job when it processes your resume. It is doing two, and they require completely different kinds of evaluation.

The first question is mechanical: can the parser extract your name, your employers, your titles, your dates, and the text of your bullets into structured fields? This is technical readability, and it is objective and countable. It has a right answer for any given file, and that answer does not depend on which job you are applying to. A resume built in a single column with standard headings either extracts cleanly or it does not — you can count exactly what came through and what did not, the same way you would count broken links on a page.

The second question is a judgment call: how closely does this resume line up with what one specific posting is asking for? This cannot be assessed from a resume sitting by itself, because "good match" is entirely defined by what that particular posting asked for. A resume that lines up extremely well against a senior data analyst posting might line up with a marketing coordinator posting not at all, and neither result says anything about the resume's quality on its own — it says something about the pairing.

Guides that blend these into one "ATS score" are the source of most of the bad advice in this space. Readability is a fact about a file. Match to a posting is an estimate about fit, and it changes with every posting you point it at.

Picture the same resume submitted to two different postings through two different systems. The readability answer doesn't move — if the file is a single column with standard headings, every system reads your last job title correctly, every time, regardless of which posting it lands against. The match answer moves constantly: strong against a posting that names your actual title and tools, weaker against one asking for a specialization you don't list. Treating that second, moving number as if it were the first, fixed one is exactly where "beat the ATS" advice goes wrong — it borrows the certainty of a countable fact and applies it to a judgment that was never certain to begin with.

Four myths everyone still believes

A few claims about ATS software repeat so often that they have started to sound like settled fact. None of them hold up once you check what these systems are actually documented and used to do.

What actually breaks parsing

Parsing engines — including the ones behind widely used platforms like Greenhouse and Lever— read a resume by extracting text in the order it sits in the file's underlying structure, not the order it appears visually on the page. A single column reads top to bottom the way a person would. Two columns can read left-line, right-line, left-line, right-line, stitching your most recent job title to the wrong employer. A table can lose its cell boundaries entirely and dump every cell into one unbroken string.

The specific things that reliably cause damage:

• Multi-column layouts and tables. Anything that depends on reading left-to-right and top-to-bottom at once.
• Text inside headers or footers. Contact information placed there is frequently dropped before it ever reaches a field.
• Text boxes and inserted graphics — including a skills section rendered as a bar chart or star rating instead of words.
• Non-standard section headings that don't match anything the system is built to recognize.
• Photos, icons used as bullet markers, and fonts that don't embed in the exported file.

None of this is about outsmarting an algorithm. It is about whether a text-extraction routine can read a file the same way a person can, and that is knowable in advance rather than guessed at — count it, and you know exactly what is true of your resume instead of trusting a global average that may not describe your file at all.

There's a rough version of this test you can run yourself: open the file, select all, copy, and paste it into a plain text editor with no formatting. If your name, your most recent employer, and your dates still read in the right order with nothing missing, a parser has a good chance too. If the paste comes out scrambled — a job title floating away from its employer, a whole column missing — that's the same failure a real parser hits, just visible to you directly instead of hidden inside someone else's software. That is the readability half of a real ATS check; the format side of it is covered in more depth in the best resume format for 2026.

Matching to a specific posting is judgment, not a score

Match strength only means something against one posting, because what counts as a strong match is entirely defined by what that posting asked for. If all you can supply is a job title, the read is necessarily broad; a fuller posting narrows it. Either way, there is no way to compress "how one hiring manager configured this one screen" into a single portable number, so it shouldn't be reported as one.

The honest way to report it is a band, not a percentage — something closer to "strong alignment," "partial alignment," or "notable gaps," alongside a plain explanation of what is driving the read. A resume is also a lossy record of a career; people leave real credentials off a resume constantly. So the absence of a keyword or a requirement on the page is evidence that it isn't on the page — not evidence that you don't have it.

That distinction matters more than it sounds like it should. If a posting lists a certification as required and your resume doesn't mention it, the accurate statement is that the posting names it and your resume doesn't show it — not that you lack it, and not that you'd be screened out. You might have let it lapse off the page by accident, the way people do with older credentials. A good match assessment tells you what's missing from the document you have, which is fixable in ten minutes, rather than implying anything about what's missing from your background.

For the keyword side of that matching problem specifically — which terms actually move the needle and which ones are filler — is worth reading before you rewrite a bullet to chase a phrase from the posting.

What to actually do about it

Once the two questions are separated, the actual work is short. Most of it is mechanical, one-time, and does not need to be repeated for every posting.

1. Fix the mechanical problem first: single column, standard headings, no tables, no text boxes, no graphics standing in for words. This is a yes/no fact about the file and it is worth getting right before anything else.
2. Get the actual posting, not a paraphrase of it, and pull the 8-10 terms that describe the real responsibilities and required skills — not filler phrases like "fast-paced environment."
3. Use those terms inside real sentences describing what you did, not as a standalone list bolted onto the bottom of the page.
4. Match your most recent title's seniority level honestly. Inflating a title to match a posting tends to create a mismatch elsewhere in the resume that a recruiter notices even when software doesn't.
5. Read it out loud before sending it. If it doesn't sound like something you'd say about yourself in an interview, a recruiter will feel that gap too — no parser needed to explain why.
6. Save your final file as a clean PDF or .docx, not a scanned image or a design tool's export with embedded graphics. A file that looks identical on your screen can still carry hidden structure a parser can't see through, so export from a template built for this rather than a general design tool repurposed for a resume.

None of this requires knowing which specific system a company runs. Millions of hires happen every month through review processes built on Workday, Greenhouse, Lever, and a dozen others that differ in the details — but the split between what can be counted and what requires judgment holds across all of them. Fix the countable half because it's free to get right. Treat the judgment half as exactly that, and stop looking for a number nobody is actually handing out.