On this page12 sections
Key takeaways
- There is no universal ATS score: the number a checker shows you is that tool's invention, and employer systems never display one (ATS Verification, 2026).
- Parsing engines read the same file differently: Textkernel lifted column-layout extraction accuracy from 60% to 82% on its own engine.
- Match rates are tied to one job description: interview rates peak at 6.30% for 40–59% keyword coverage and fall to 4.95% at 80%+ (Huntr Q2 2026).
- Greenhouse auto-rejects only on rules the employer configures for Yes/No or select-style questions, different companies, different tripwires.
- Only 32 of every 1,000 applications reached a first interview in Greenhouse's Fall 2026 benchmark of 161.8 million applications.
Because “ATS compatibility” isn’t one thing. There’s no single parser, no single scoring formula, and no universal grade your resume either passes or fails. Every result you’ve ever seen was produced by a specific parsing engine, running a specific algorithm, against a specific job description, inside a specific employer’s configuration.
Different parsing tools, employer settings, and job requirements can produce different results from the same resume. Review those factors below, then check the extracted text and application fields yourself.
What ATS compatibility actually means
ATS compatibility is whether a specific parsing engine can cleanly extract your resume’s content, not a portable score you can earn once and carry around.
Here’s the part most job seekers miss: real hiring systems don’t hand out grades. “The percentage score most checker sites show you is generated by that site, not by the real ATS a recruiter uses,” as ATS Verification’s founder wrote in June 2026. Workday and Greenhouse parse your file into fields and rank you against the employer’s own criteria. There’s no 0–100 number anywhere in that pipeline.
Even the checkers half-admit it. Jobscan’s resume scanner documentation describes its match rate as a Jobscan visualization, not a score your employer’s ATS assigns. So when two tools disagree about your resume, they’re not disagreeing about a fact. They’re running different math.
The five levers that move your result
Map every verifiable cause and the variance comes from five places. Some you control fully, some partially, and some not at all.
| Source of variance | What it changes | What the data shows | How much you control it |
|---|---|---|---|
| Parsing engine | Which fields survive extraction into the ATS | One leading engine lifted column-layout accuracy from 60% to 82% (Textkernel); the same file can extract differently on different engines | Indirect, through format |
| Scoring algorithm | The number a checker shows you | Jobscan runs 30+ checks on its own scale; no employer ever sees it | Partial |
| Target job description | The keyword demands behind your match rate | Interview rates peaked at 40–59% keyword coverage (6.30%) and fell to 4.95% at 80%+ (Huntr, Q2 2026) | Full |
| File format | Whether your text even reaches the parser | Testers disagree, Jobscan’s tests favor PDF while other guides push DOCX; size limits vary by system | Full |
| Employer configuration | Knockout questions, auto-reject rules, parsing on/off | Greenhouse auto-reject fires only on employer-written rules for Yes/No or select-style questions (Greenhouse docs) | None, but knowable |
Now let’s take each lever apart.
Why the same file parses differently
Different parsing engines have different accuracy profiles, and even the best ones miss things. Before a human sees anything, software converts your file into structured fields, name, employers, titles, dates, skills.
Who builds those engines? A handful of specialists: Textkernel (which has combined its parser with Sovren’s technology), Daxtra, RChilli, and Affinda, plus in-house parsers like Workday’s and AI screening layers like HiredScore, which Workday agreed to acquire in February 2024. Textkernel says it parses more than 2 billion resumes and job postings a year and claims over 60% of the global HR-tech industry as customers. Your target employer licensed some version of one of these, and the version matters.
Even the market leader improves from year to year. Textkernel’s own engineering writeup on column-layout resumes reports that correct classification of column-related gaps jumped “from 60% to 82%,” cutting errors by more than half on a layout it estimates appears in at least 15% of all CVs. The same post logged contact-info fill-rate gains of 4–10 percentage points across 12,000+ random resumes. Same file, older engine, worse extraction.
Employers’ own documentation says the quiet part out loud. Workday’s admin guide states that “resume parsing results can vary based on resume format and order of words,” and recommends resumes without image-based styling. The vendor is telling you the parse isn’t deterministic.
Format stress-tests make the same point. Enhancv rendered one resume through 17 templates and ran 357 extraction passes across seven tools: word recovery barely moved between one and two columns, but structural integrity did not, sections stayed intact only 35% of the time with two columns versus 60% with one under the worst extraction method.
Their AI-parser test went further. Running ten real resumes through four large language models shrank the structural loss to about three points, but employer-name accuracy fell from 100% to 83% because interleaved columns separate job titles from companies. Same content, different extractor, different structured result. (For the mechanics of that first read-through, see what an applicant tracking system actually reads.)
Why the scores differ even when parsing is fine
Because every checker is its own scoring engine with its own weights; same inputs, different math. Jobscan’s documentation says its match rate is built from your hard skills, education level, job title, soft skills, and other keywords, across 30+ checks in five categories, and recommends aiming for at least 75%.
That 75% is Jobscan’s benchmark, not the labor market’s. Enhancv’s testing showed that simulated “smarter” readers (large language models) produce measurably different results from classic extraction tools on the very same file. Neither one is the pipeline your employer runs.
The deeper problem: these tools don’t measure anything the employer does. They generate an opinion, sometimes a useful one. The score’s real value is as a checklist generator; it surfaces keyword gaps and formatting flags you’d otherwise miss. The number itself is a rounded guess wearing a suit. (There’s a deeper look in why ATS scores can be misleading.)
Why the job description changes your match rate
Because the match rate measures overlap between your resume and one specific posting. Swap the posting and you change the keyword set, the denominator of the whole calculation. Jobscan is explicit that its rate is computed against the job description you enter.
Huntr’s Q2 2026 report (built on 179,363 saved jobs, 102,777 applications, and 32,082 tailored resumes, with data pulled August 7, 2026) shows what that means in outcomes. The median resume covered 33.3% of a posting’s keywords, and interview rates peaked at 6.30% for resumes covering 40–59% of them. Coverage of 60–79% dropped to 4.40%, and even 80%+ recovered only to 4.95%.
Read that again: more keywords did not mean more interviews. Maximal matching underperformed moderate matching, which tells you the match rate is a targeting tool, not a scoreboard. It also explains why two scores for the same resume can both be “correct”, they’re answers to different questions.
Why file-format advice contradicts itself
Because the answer is “it depends on the employer’s stack,” and the big testers genuinely disagree. Jobscan’s own tests, laid out in a post it updated in May 2025, conclude that “most applicant tracking systems read and parse PDF resumes more accurately” and recommend PDF unless a posting asks for Word. Plenty of other guides still push DOCX as the safe default. Neither deserves gospel status; follow the posting and keep both formats ready.
What’s not contested is what breaks parsing on any format:
- A file with no text layer; scans and screenshots are images, so there’s nothing to extract.
- Contact details buried in headers, footers, or text boxes, where many parsers don’t look.
- Tables, multi-column layouts, and spaced-out lettering, which older engines read out of order.
- Oversized files: a third-party Greenhouse guide published in July 2026 reports that files over 2.5MB simply aren’t parsed.
Greenhouse itself accepts .doc, .docx, .pdf, .rtf, and .txt files, and the stakes are real, Jobscan reports 97.8% of Fortune 500 companies use an ATS. When a parse fails, details don’t vanish politely; they go missing in ways a recruiter may never notice. That failure mode deserves its own read: resume parsing errors: why your details go missing.
What the employer configures without telling you
More than you’d think, the same resume walks into different tripwires at every company. Screening is configured per employer, per job post.
The cleanest evidence is Greenhouse’s own support documentation. Auto-reject is opt-in: an employer writes a rule against a custom application question of type Yes/No, single-select, or multi-select, picks the answer that triggers rejection, and can attach a rejection reason and email.
Those rules live only on Greenhouse’s Plus and Pro tiers, and auto-rejected candidates don’t even trigger the usual new-application notification. Auto-reject isn’t a parser feature; it’s a rule a human at the employer wrote against form answers.
Keyword screening? Mostly human. A third-party guide to Greenhouse published in July 2026 states plainly that the system “does no automatic content filtering on keyword density.” Workday hands employers similar switches: they can configure application templates to hide parsing from external candidates entirely, and its parser never auto-fills the Languages and Skills fields.
This layer explains a lot of “same resume, different results” stories that have nothing to do with formatting. One employer’s knockout question on work authorization rejects you; the next company doesn’t ask it. No checker can see any of it.
What candidate-side checkers can and can’t copy
Checkers can approximate your keyword fit and flag format risks. They cannot see the employer’s side at all.
What they do well:
- Keyword and skill gaps against a job description you supply.
- Formatting red flags: columns, tables, headers and footers, file type.
- Readability basics: sections, dates, contact fields, word count.
What they can’t replicate:
- Which parsing engine your target employer licensed, and which version.
- The employer’s ranking weights, knockout questions, and auto-reject rules.
- Recruiter behavior, who actually opens the file, and for how long.
So treat the output as a pre-flight check, not a simulation. It’s the difference between checking your tire pressure and knowing the road conditions ahead.
What actually predicts interviews
Tailoring to the actual posting beats score-chasing, and the funnel is brutal no matter what. Greenhouse’s Fall 2026 Hire Standard benchmarks, covering 161.8 million applications across 9,473 companies from January through August 2026, put it in one line:
Of every 1,000 applications, only 32 ever reach a first interview.
Against that backdrop, the candidate-side data is clear. Huntr tracked 101,501 tailored and 1,893,054 untailored applications from October 2024 through June 2026: tailored resumes earned a 5.71% interview rate versus 3.09% untailored, 1.85x higher. And the tailoring that paid off was moderate, not maximal, as the keyword-coverage curve above shows.
Greenhouse’s own numbers back strategy over polish. Applications per hire ranged from 171 at enterprise companies to 474 at small businesses, and the report calls company size the biggest predictor of hiring performance. Check that the resume parses correctly and clearly shows your fit for the posting. Application strategy and employer demand also affect the outcome.
What to do instead: a five-step testing routine
Stop hunting for one universal number; it doesn’t exist. Run this instead:
- Check the parse before the score. Paste your resume into a plain-text editor, or use a checker that shows the extracted text. Confirm your name, phone, email, every employer, title, and date survived. Fix gaps before you glance at any percentage.
- Test against the real posting. Run keyword tools against the actual job description you’re applying to, never a generic benchmark. Target moderate coverage: the outcome data peaks at 40–59% keyword coverage and drops when you push past 80%.
- Follow the posting’s file instructions exactly. Keep both a text-based PDF and a DOCX, single column, standard section headings, contact info in the body. No tables, no headers or footers holding your phone number, no scanned pages, and stay under the size limit.
- Re-run after every edit, same tool, same job description. Track your deltas, not your absolutes. A 10-point gain on the same tool against the same posting means something; the raw number doesn’t transfer anywhere.
- Read the application form before you answer it. Knockout questions are the one automated rejection trigger most systems have. Answer them accurately, and don’t assume a rejection at one company predicts anything at the next; the rules were written by different humans.
Do that and you’ve covered the two levers you fully control (the job description and the file) while de-risking the one you don’t: the employer’s configuration. That’s real ATS compatibility work, as opposed to worshipping a number that was never yours.
FAQ: same resume, different ATS results
Why did the same resume get different scores on Jobscan and another checker? Because each checker runs its own proprietary algorithm with different weights and checks. There’s no standardized scoring, so the same inputs legitimately produce different numbers, neither tool is “wrong.”
Do ATS systems automatically reject my resume? Rarely, and only when configured. In Greenhouse, auto-rejection happens only when the employer writes a rule against a Yes/No or select-style application question; there’s no automatic keyword-density filtering.
Is PDF or DOCX better for ATS compatibility? It depends on the employer’s system: Jobscan’s tests favor PDF, while other guides favor DOCX. Follow the posting’s instructions, and whichever you send, make sure it has a real text layer.
Do two-column resumes fail ATS parsing? Usually not for raw text, modern extraction recovers words from two columns nearly as well as one. The real risk is scrambled reading order: sections stayed intact 35% of the time with two columns versus 60% with one under the worst method Enhancv tested.
What match rate should I aim for? Jobscan recommends 75% or higher, but the outcome data points elsewhere: Huntr’s Q2 2026 report found interview rates peaked at 40–59% keyword coverage and fell at 80%+. Match moderately, then stop.
Can recruiters see my Jobscan score? No. The score exists only inside the checker’s site. Recruiters see their own ATS’s parsed fields, the original file, and their own screening notes.
Why does my score change when I apply to a different job? Because match rate is calculated per job description. A new posting brings a new keyword set, so the same resume produces a different, and equally “correct”, number.
Is there such a thing as one universal ATS score? No. Employer systems like Workday and Greenhouse don’t produce a 0–100 grade, and every checker generates its own. Any “universal score” is marketing, not measurement.
Sources
- ATS Verification: “Best ATS Resume Checker 2026: The Parse, Not the Score” (June 25, 2026; updated September 29, 2026)
- Enhancv: “What Happens to a Two-Column Resume When a Machine Reads It” (2026)
- Fastapply: “How to Beat the Greenhouse ATS (2026)” (July 10, 2026)
- Greenhouse: “The Hire Standard” recruiting benchmarks, Fall 2026
- Greenhouse Support: “Auto-reject” documentation
- Huntr: “Q2 2026 Job Search Trends Report” (data pulled August 7, 2026)
- Jobscan: “Resume PDF vs Word: Which Should You Choose?” (updated May 27, 2025)
- Jobscan: Resume Scanner & Match Rate documentation
- Textkernel: “Improving Information Extraction from Column Resumes” (engineering blog)
- Textkernel: Resume Parser product page
- Textkernel: Sovren & Textkernel combined parsing
- Workday Admin Guide: “Concept: Resume Parsing”
- Workday Newsroom: “Workday Announces Intent to Acquire HiredScore” (February 26, 2024)
Written by ResumeFlex. Questions or corrections? Contact ResumeFlex. Read our source and correction standards.
Filed under
ATS & Parsing