What an ATS resume checker measures, and what it leaves out

An ATS resume checker gives you a number. On 20 August 2026 I went looking for what sits underneath it, and built two PDFs of the same fictional resume. Same candidate, same nine years of experience, same 117 words, word for word. The only difference was the template: one plain single column, one with a sidebar for skills and education and the contact details banded across the top. Then I pulled the text back out of both, because extraction is the first thing a real applicant tracking system does to your file.

Variant A came out in the order a human reads it. Variant B led with the candidate’s email address, put his name further down, alternated the Skills and Summary headings before either section had any content, and printed Education above Experience. Same words, and a document a parser reads completely differently.

I had planned to open by saying nobody in the category shows you that extraction, and that turned out to be wrong. Of the three checker pages I read end to end that day, Resume Worded’s scanner page does show it, promising to “parse your file the way an applicant tracking system does and show you the extracted text”, and it publishes its own worked example of a two-column layout scrambling. Kickresume’s checker page and Jobscan’s homepage don’t. One in three, then, not zero in the category. What follows is an independent reproduction with a method you can rerun, not a scoop.

What an applicant tracking system actually does with your file

Workable, which sells one, defines parsing as “the process by which technology extracts data from resumes”: your name and email, the degrees you hold, your skills, your work history. Read that page end to end and you won’t find a claim that the software scores you, ranks you, or throws anything away. Extraction is the product being described.

Workday’s administrator documentation is blunter. It opens with “Resume parsing enables recruiters to create a prospect or job application when attaching a resume in Workday”, which is a data-entry convenience pointed at the recruiter rather than a filter pointed at you. A few lines down sits this: “Workday doesn’t auto-fill fields you configure as hidden or these fields”, above a two-item list: Languages and Skills.

Workday’s parser won’t auto-fill the Skills field on the application record: a recruiter types it or it stays empty. That isn’t the same as the text vanishing. It does mean the block you tuned isn’t landing where you assume. The same page states that Workday resume parsing “results can vary based on resume format and order of words”.

Order of words.

Two failures that look the same on an ATS score

When a checker hands you 62 out of 100, that number is covering two unrelated events and never says which. The first is that your file didn’t extract properly, which is mechanical, documented by the ATS vendors themselves, and fixable tonight. The second is that it extracted perfectly and the tool dislikes your phrasing, which is an editorial opinion held by a marketing company rather than by any hiring system.

Greenhouse publishes the causes of the first kind. Its support article on an unsuccessful Greenhouse resume parse, updated 2 March 2026, names three top-level causes: a file over 2.5MB, placeholder data such as “Company 1” or “employee 1” that the parser refuses to trust, and formatting. Under formatting it lists nine specific things: spaces inserted between letters, graphics or photos or word art, a resume uploaded as an image rather than a .docx or .pdf, tables and headers and footers, contact details sitting in a header, footer, or text box, a columned layout, sections without clear boundaries, company names missing an Inc. or LLC, and abbreviated job titles, their example being “Sr. Account Exec”.

The test: the same words in two layouts

Both PDFs came out of headless Chrome on 20 August 2026: same fonts, same body copy, synthetic candidate, invented employer names, an invalid email domain, so nobody’s real data went anywhere. Text was extracted with Poppler’s pdftotext at default settings, no layout flag.

Sorted and compared token by token, the two extractions are identical apart from the two pipe characters Variant A used to separate its contact line. 117 real words in, 117 real words out, both times. Only the order changed, and order is what a parser uses to decide which string is a name and which is an employer.

A is the single column, B is the sidebar with the header band. Numbers are raw pdftotext output lines, blank lines included, which is why some jump.

In Variant B the Education heading arrives on line 19 and Experience on line 21, so a degree from 2016 reaches the parser before the current job does. The first line is the email address; the name waits until line 5. A parser walking that stream has to guess, and Workday’s own sentence about order of words suggests the guess isn’t always right.

Poppler is not Greenhouse’s parser and it isn’t Workday’s. Without a recruiter seat I can’t test either engine, so I can’t tell you Variant B fails at Greenhouse. What I can tell you is that Greenhouse names columned layouts and header contact blocks as parse-breakers in its own docs, that Workday names order of words in its own docs, and that my extraction shows exactly that breakage on a file whose words are otherwise perfect.

If layout is the part that worries you, the structure section of our engineering resume guide covers section order rather than word choice.

Does a low ATS resume checker score mean you get auto-rejected?

No, at least not according to any vendor documentation I could find. Greenhouse says that when a parse fails you “manually input the candidate’s details into the fields”, so the resume stays attached to the candidate record and a human types the rest. The application isn’t discarded. No ATS doc I read describes automatic rejection from a parse failure or a keyword count, and a third-party score has no wiring into the employer’s system anyway.

A badly built file costs you a recruiter’s patience, some fidelity in the search index, and possibly a mis-typed job title. It doesn’t trigger a robot rejection, because nothing in the documented workflow is wired that way. The detection side of this is in whether an ATS can tell your application was written by AI.

New York City puts real legal weight on automated screening. Under Local Law 144 of 2021, employers can’t use an automated employment decision tool unless it has had an independent bias audit within the previous year, the results are posted publicly, and candidates get notice. DCWP began enforcing it on 5 July 2023. That is what a real automated decision gets, and it is a long way from a $9 browser widget grading your bullet points.

Where the 99 percent number comes from

Two of the three checker pages I read on 20 August 2026 open with the same statistic. Kickresume’s checker page, fetched 20 August 2026, says applicant tracking systems are “used by 99% of Fortune 500 companies and even 20% of smaller businesses”, with no source attached. Jobscan’s homepage the same day: “Ninety-nine percent of Fortune 500 companies use an ATS as part of their recruiting strategy”, also unsourced.

I went looking for the primary source and couldn’t find one. It might exist, but if the two companies using it to sell you something can’t be bothered to source it, treat the number as folklore that hardened into a sales line.

Thresholds get the same treatment. Kickresume says “Anything from 75% up is generally considered good”; Resume Worded says “a good score is 85 or above, and ideally 90 or above”. Neither explains what those numbers correspond to inside a real hiring system, and neither can, because the scale is theirs.

What I could verify about prices on 20 August 2026

Here is what I read off each company’s own pages on one day.

Resume Worded, Jobscan, Teal and MyPerfectResume carry no price above because I couldn’t read one. Jobscan’s pricing URL returns a 301 to app.jobscan.co/plan, which draws its numbers in the browser; the other three blocked, timed out or 404’d on me. That’s my method’s limit, not a vendor decision, and I won’t quote prices I didn’t see.

Disclosure: LastRound AI ships a free resume checker, which I deliberately kept out of that table. Grading my own homework next to three competitors tells you nothing, and the argument here is that scores from interested parties deserve suspicion. Ours included.

The 30-second version you can run yourself

Open your resume PDF in any viewer, select all, copy, paste into a plain text editor. What you get is roughly what a parser gets, minus the vendor’s cleverness. Is your name the first line? Are your job titles still attached to the right employers and dates? Did the columns interleave, so a skill lands between two bullets from 2019? (Watch for anything that vanishes outright, which happens when your skills are drawn as an image or your contact block lives in a true document header.)

If that all reads correctly, stop optimizing: your file parses, and the rest of the feedback from any ATS resume checker is a stranger’s taste in verbs. If something is broken, fix the layout and ignore the score, which was never measuring that anyway. Our piece on what a recruiter reads in the first six seconds is a better use of the next hour than chasing 85.

Once the file is clean, volume becomes the bottleneck instead of formatting, which is where Auto-Apply earns its place: it ranks openings by fit against your resume, drafts a tailored cover letter and the screening answers for each role from that same resume, and queues every submission for your approval. The free tier also gives you 15 credits a month, and those reset monthly rather than rolling over.

The thing I actually want and can’t buy is a seat on the other side of the upload. Every claim here about what happens after you hit submit is reconstructed from vendor docs plus a desktop extraction, because Greenhouse and Workday don’t sell candidates a window into their own parse. Of the three checker pages I read on 20 August 2026, Resume Worded’s simulation came closest, and it’s still a simulation of somebody else’s software. If you work in recruiting and would run these two files through your real stack, I’d rather see that output than be right.