OptiResume

Methodology · Evidence

How ATS actually works, and how we score for it

Most resume tools sell you a scary number and a myth. We took the opposite bet: every one of our 27 rules maps to published evidence, and the things the evidence does not support, we refuse to score. Here is exactly what we know, and what we deliberately do not claim.

No myths. No flattery. Twenty-seven checks you can audit.

How an ATS actually works in 2025

An Applicant Tracking System is the software companies use to parse, store, and search every application. Almost every mid-to-large employer runs one. The folklore says it is a robot gatekeeper that shreds three of every four resumes on sight. That is not what the evidence shows.

In a 2025 recruiter survey by Enhancv, 92% of recruiters said their ATS does not auto-reject resumes for formatting, content, or design. The small share of automated rejection that does exist is almost always driven by knockout questions, work authorization, location, a required license, not by a parser scoring your bullet points. The scary “75% get auto-rejected” stat has no credible source behind it. It is a sales tactic, and we will not repeat it.

So if the ATS is not the executioner, what is the real risk? Two quieter failures. First, your resume can parse badly, the machine misreads it, so the text a recruiter searches never contains what you actually wrote. Second, your resume can parse perfectly and still miss the words the recruiter searches for, so a human never sees it. Those two failures, findability and readability, are where applications quietly die. They are exactly what a good score should measure.

What the evidence says

  • A skills list is cheap talk until you prove it

    In a randomized field experiment published through the NBER, simply listing more technical skills, with nothing in the work history to back them, moved employer interest by zero. Recruiters discount unproven claims. That single finding is why our flagship rule checks whether each skill you list actually shows up in a project or a bullet, and flags the ones that do not. Prove it, or drop it.

  • Parsing breaks at text extraction, not at meaning

    Textkernel, one of the largest commercial resume parsers, is blunt about where parsing fails: the document-to-text conversion, long before any semantic step. Graphics used as text, vertical or diagonal text, and multi-column layouts that scramble reading order are the usual culprits. Parsing benchmarks find roughly one in five real resumes uses a layout that trips naive extraction. Modern parsers often cope; legacy ones genuinely do not. So we warn about a risky layout, we never pretend it is an instant death sentence.

  • Length follows experience, not a one-page rule

    The universal “one page or you are done” rule is not real. A ResumeGo simulation with 482 recruiters found two-page resumes were preferred 2.3x for experienced candidates. The honest version is a band, not a commandment: one page for a fresher with limited history, room to breathe once you have the experience to fill it. We score length against your career stage, not a myth.

  • Keyword matching is semantic, and callbacks are not ours to promise

    LinkedIn's own engineering describes recruiter search as learned, semantic ranking, not literal string matching, which is why our match layer credits “ML” as “machine learning”. And on the biggest question, does a resume score predict a callback, the audit literature in PNAS is clear that identity effects dominate that outcome. So we never claim our score predicts callbacks. It measures what your resume does, not what a biased world does back.

Myths we refuse to score

A rule that punishes you for a myth makes your score dishonest. These are the popular beliefs the evidence does not support, so no rule of ours touches them.

  • The ATS auto-rejects your resume

    92% of recruiters say it does not. We score readability and findability, and let no rule quietly fail you for a rejection that is not happening.

  • A two-column layout is instant death

    It is not. It can make a Skills section parse wrong, which we flag as a warning so you can fix it, but we never auto-fail a resume for having two columns.

  • Every resume must be exactly one page

    False past the fresher stage. We score length against your experience, so a strong two-page mid-career resume is not penalized for existing.

  • 75% of resumes are killed by the robots

    A scare stat with no credible source. You will never see it in our product, our marketing, or our score.

Our method

The score is 27 deterministic local rules with fixed weights. No language model sits anywhere in the formula, so the same resume always produces the same number, and every point has a name, a fix, and a reason you can check. AI is used only where it belongs, to rewrite a weak bullet or draft a cover letter, never to grade you behind a curtain.

Those 27 rules only measure three things the evidence actually supports: machine readability, whether a parser can extract your resume cleanly; recruiter findability, whether the words a recruiter searches for are present and match semantically; and human-skim quality, whether your bullets are quantified, active, and scannable in the few seconds a real person spends. Everything else, the folklore, is left out on purpose.

And we do it without keeping your resume. We parse your file in memory, extract the text, and discard the original bytes immediately, no PDF, no DOCX stored. Personal information is redacted before any AI ever sees the text. An honest score and a privacy guarantee are not a trade-off here; they are the whole point.

What we build on

Named sources, in plain text. Every rule in the engine cites the evidence behind it right next to the fix, inside the product.

  • Enhancv recruiter survey (2025): 92% of recruiters say their ATS does not auto-reject resumes.
  • NBER field experiment: listing extra skills, with nothing to back them, moved employer interest by zero.
  • Textkernel parsing guidance: most parse failures happen in document-to-text conversion, not semantics.
  • RealResume parsing benchmark: roughly one in five real resumes uses a layout that breaks naive reading order.
  • LinkedIn engineering: recruiter search ranks on learned, semantic models, not literal keyword match.
  • ResumeGo recruiter simulation: 482 recruiters preferred two-page resumes 2.3x for experienced candidates.
  • PNAS callback audit studies: identity effects dominate callbacks, so no honest tool can promise them.
  • EEOC and gov.uk National Careers Service: the legal basis for what personal data to leave off a resume.

See your resume scored the honest way

27 rules you can audit, a keyword match you can trust, and not a single myth. Free, no credit card, and your file is never stored.

Score my resume, freeBrowse placement drives