ResumeAtlas study · June 2026 – September 2026
What breaks ATS resume parsing? A 2026 ResumeAtlas benchmark
We ran ResumeAtlas's ATS-parse analyzer — the same one behind our ATS resume parser — over 218 real resumes people submitted to the tool, and read the blended ATS score of 1,085 resume analyses. This is what a text-extraction parser loses, factor by factor.
A ResumeAtlas study, not an industry benchmark. Small sample, tech-skewed, one analyzer. Free to cite with attribution. Last updated September 20, 2026.
Executive summary
- 16.5% of the 218 resumes did not produce a usable candidate record in our analyzer — meaning it lacked a cleanly extracted identity, recognized fewer than two of the three core sections (Experience, Education, Skills), or had under 40 words of readable text.
- 40.8% were flagged on “Tables & columns”, the most common parse problem in the set (18.8% at a severity our analyzer treats as likely to fail).
- 7.8% had no cleanly extractable name, 5% no email and 6.9% no phone number — the fields every ATS writes to your applicant profile first.
- 40.8% showed multi-column or table artifacts in the extracted text — the signature of a layout read out of order.
- Separately, across 1,085 analyses, the blended ATS score was below 70 for 12.5% of resumes — the band where a parser is likely dropping or scrambling whole sections.
Key numbers
16.5%
Did not yield a usable candidate record
n = 218 resumes
60.6%
Had all three core section headings recognized
n = 218
7.8%
Missing at least one core contact field (name/email/phone)
n = 218
12.5%
Blended ATS score below 70
n = 1,085 analyses
How to read this: These are resumes real people submitted to ResumeAtlas between the dates above, run through ResumeAtlas's own ATS-parse analyzer. The sample is small, skews toward software, data and engineering roles, and covers signed-in users only. Treat it as a directional ResumeAtlas study, not a definitive measure of every resume or every ATS.
Methodology
Two datasets, generated together on September 20, 2026.
1. The parse analyzer, over 218 real resumes
Every distinct resume whose text a signed-in user submitted to ResumeAtlas between June 2026 – September 2026 (322 submissions, 218 unique after removing re-submissions of the same file) was run through analyzeResumeForAts — the deterministic analyzer that powers our ATS resume parser. It runs 12 checks that model the text-extraction and field-mapping stage most applicant tracking systems share before keyword matching: contact fields, section headings, one work-history row per role, date consistency, a readable skills list, single-column reading order, and clean characters. Each check returns pass, warn, or fail. “Flagged” below means warn or fail; “likely to fail” is the fail level only.
2. The blended ATS score, from 1,085 analyses
ResumeAtlas's resume-vs-job-description analysis assigns a single 0–100 ATS score to every run. We read that score from 1,085 analyses between March 2026 – June 2026. It is a broader, higher-level signal than the 12 checks — one number covering layout, structure and readability together — and it is also published on our resume statistics page.
Definitions
- Usable candidate record: a cleanly extracted name or email, and at least two of Experience / Education / Skills recognized as headings, and at least 40 words of readable text.
- Multi-column / table artifact: runs of tab characters or wide interior whitespace in the extracted text — what side-by-side layout blocks and table cells leave behind when a parser flattens them to one stream.
- Contact field missing: the analyzer found no value for that field in readable text — not that the resume necessarily lacked it.
Exclusions and privacy
Aggregate counts and shares only. No resume text, name, or any identifying field is published or stored for this page — the analyzer runs, the pass/warn/fail tallies are kept, the text is not. The sample is signed-in users who used a paid or free analysis between the dates above, and skews toward software, data and engineering roles.
What broke, in order
Share of the 218 resumes flagged on each check (warn in amber, likely-to-fail in red). Ordered by how often each was flagged.
Tables & columns
40.8% flagged · 18.8% likely to fail
Flagged when: two-column layouts and tables for core content.
Section detection
39.4% flagged · 16.1% likely to fail
Flagged when: creative headings like “my journey” or “toolbox”.
Icons & graphics
28.4% flagged · 19.3% likely to fail
Flagged when: contact icons, emoji, and decorative glyphs.
Skills
27.1% flagged · 22.9% likely to fail
Flagged when: skill grids, rating bars, or column layouts.
Work experience
25.7% flagged · 11% likely to fail
Flagged when: roles merged or missing title/date pairing.
Education
25.7% flagged
Flagged when: degrees buried in prose or unlabeled sections.
Dates
9.2% flagged · 6.9% likely to fail
Flagged when: mixed styles, “’21-’23”, or dates in graphics.
Name
7.8% flagged · 7.8% likely to fail
Flagged when: name inside a header band, logo, or image.
Headers & footers
7.8% flagged
Flagged when: name/email/phone trapped in header or footer bands.
Phone
6.9% flagged
Flagged when: number split across table cells or stylized.
5% flagged · 5% likely to fail
Flagged when: email shown only as an icon or in a graphic.
The analyzer's bullet-structure check is excluded here: text extraction normalises or strips bullet glyphs, so its fail rate reflects extraction, not the resume. See limitations below.
Can an ATS read a two-column resume, tables, headers, icons — and PDF?
The same 218 resumes, answered question by question. Every rate below is the share of submitted resumes the analyzer flagged for that problem — not a vendor demo, not a rule of thumb.
Can an ATS read a two-column resume?
Often not cleanly. 40.8% of the 218 submitted resumes were flagged for two-column layouts or tables holding core content. Read as plain text, the two columns interleave line by line — a skills sidebar ends up spliced into the experience section, and dates drift away from the jobs they belong to. The parser does not crash; it files your content in the wrong place, which is worse because nobody tells you.
Flagged on 40.8% of 218 resumes, June 2026 – September 2026. Fix: run the file through the ATS format converter (free, no signup) and it comes back single-column with the same content.
Can an ATS read tables in a resume?
Same failure, same rate: tables and columns are one check here, flagged on 40.8% of resumes. Cell boundaries vanish on extraction, so a table of jobs becomes a run of text where the title, company and dates are no longer on the same line. A phone number split across cells (6.9% had an unreadable phone) is the small version of the same problem.
Flagged on 40.8% of 218 resumes, June 2026 – September 2026. Fix: run the file through the ATS format converter (free, no signup) and it comes back single-column with the same content.
Can an ATS read text boxes, icons and graphics?
No — text inside a shape or an image is not text to a parser. 28.4% of resumes were flagged for contact icons, emoji or decorative glyphs, and an email shown only as an envelope icon reads as no email (5% had none the parser could find). 62.8% had no readable LinkedIn URL for the same reason. Spell it out: the word "Email:" followed by the address.
Flagged on 28.4% of 218 resumes, June 2026 – September 2026. Fix: run the file through the ATS format converter (free, no signup) and it comes back single-column with the same content.
Can an ATS read a skills grid or rating bars?
Poorly. 27.1% of resumes had a skills section the parser could not use — grids, star ratings, progress bars or a two-column skills block. Keyword search runs on the skills field, so a skills section that does not extract is the single most expensive layout choice. A comma-separated list under a "Skills" heading passes.
Flagged on 27.1% of 218 resumes, June 2026 – September 2026. Fix: run the file through the ATS format converter (free, no signup) and it comes back single-column with the same content.
Does an ATS understand creative section headings?
It understands the standard ones. Only 60.6% of resumes had all three core sections (experience, education, skills) recognised; 39.4% did not, and the usual cause is a heading like "My Journey", "Toolbox" or "Where I've Been". The parser maps headings to fields by name; an unrecognised heading means the content underneath has no field to land in.
Flagged on 39.4% of 218 resumes, June 2026 – September 2026. See it: the ATS resume parser (free, no signup) shows exactly which fields a parser extracted from your file.
Does an ATS read PDF or Word better?
Neither, when the layout is the same. We took 35 single-column templates, produced each as a PDF and as a Word file from the same text, and parsed both: 35 of 35 pairs reached the same verdict, 35 of 35 found the same sections, 35 of 35 recovered the same contact fields, and the median parse score was 100 for PDF against 100 for Word. The file type is not the variable; what you put on the page is. Recruiters differ, with some asking for a PDF and some for Word, so send the one the recruiter or posting asks for.
35 template pairs, ResumeAtlas parser, data generated 2026-10-25. Single-column templates only; a designed template can export differently as PDF and Word, which is a layout question, not a file-type one. Fix: the ATS format converter returns both formats from one upload.
PDF vs Word on real uploaded files
The same comparison on files people actually checked (September 2026 – October 2026): 577 PDFs scored a median 85 with 81.8% parsing into a usable record; 176 Word files scored a median 85 with 84.7% parsing. Real files differ in layout as well as type, so a gap here is mostly the layouts people choose in each program, not the file format.
Section headings and contact fields
Section headings are how a parser decides where your content belongs. In the 218 resumes:
60.6%
All three core headings recognized (Experience, Education, Skills)
23.4%
Only two of the three recognized
16.1%
One or none recognized — content routed as unstructured text
Per section, that is 79.8% with a mapped Experience heading, 77.1% Skills, 79.8% Education — the wording people use, the order they put sections in, and the headings no parser has a field for are on resume sections, headings and order.
Contact fields not extracted
These fields populate your applicant profile before a recruiter opens the file.
| Field | Not extracted |
|---|---|
| Name | 7.8% |
| 5% | |
| Phone | 6.9% |
| LinkedIn URL | 62.8% |
| Location | 46.3% |
n = 218. LinkedIn and location are frequently just absent from the resume rather than mis-parsed, so read those two as “not present in a form the parser could use” — name, email and phone are the ones that should always be there.
The blended ATS score, 1,085 analyses
A separate, larger dataset. Every ResumeAtlas resume analysis produces one 0–100 ATS score covering parsing, structure and readability together. Median: 80.
12.5%
Below 70 — parser likely dropping or scrambling sections
49.7%
70–84 — readable, leaking points on layout
37.8%
85 or higher — clean
n = 1,085 analyses, March 2026 – June 2026.
What this means for your resume
- Use one column and no tables for anything that matters. Multi-column and table artifacts were the most-flagged problem in the set. Keep dates on the same line as the role, separated by a comma or dash — not a tab stop.
- Name your sections with the standard words. “Experience”, “Education”, “Skills” — not “My Journey” or “Toolbox”. A heading the parser cannot classify sends everything under it nowhere.
- Put name, email and phone in the body as plain text. Not in a header band, not as an icon, not inside an image.
- Pick one date format and keep it. Month YYYY – Month YYYY throughout. Mixed styles scramble your work chronology.
- Drop decorative characters. Skill-bar glyphs and contact icons land in the extracted text and can corrupt the field next to them.
Run these checks on your resume
The same analyzer used for this study — see what a parser extracts from your file.
Convert to a single-column format
Rebuild a two-column or table-heavy resume into a layout that parses cleanly.
Start from an ATS resume template
A structure with standard headings and single-column reading order built in.
What this study does not prove
- It is not an industry benchmark. One analyzer was tested — ResumeAtlas's, which models the text-extraction stage that systems broadly share. It is not a measurement of Workday, Greenhouse, Lever, iCIMS, Taleo, SAP SuccessFactors, or any other commercial ATS. Those are proprietary, differ from each other, and each employer configures its own.
- The sample is small and skewed. 218 resumes, signed-in users only, heavily software / data / engineering. Percentages will move as the dataset grows.
- PDF vs DOCX parse rates on the submitted resumes — the source file type is not stored. The page answers the question from 35 template pairs rendered both ways instead (see the PDF or Word card).
- Page-count effects — not stored.
- Bullet structure — text extraction normalises or strips bullet glyphs, so a low bullet count here is an extraction artefact, not the resume.
- Any specific commercial ATS (Workday, Greenhouse, Lever, iCIMS, Taleo, SAP). This benchmark tested the ResumeAtlas analyzer, which models the shared text-extraction stage — not any one vendor's parser.
- Hiring outcomes — nothing here records whether a resume led to an interview.
Reproducibility
The 12 checks and their exact pass / warn / fail thresholds are the same rules the ATS resume parser applies, and you can run any resume through it yourself. The aggregate is regenerated from source by a build script rather than hand-written, so the page moves only when the data does. Raw resumes are not published: they are personal documents submitted in confidence, and the study needs only the pass/warn/fail tallies, which carry no identifying content.
About ResumeAtlas
ResumeAtlas is a set of tools for checking and fixing how a resume reads to an applicant tracking system and to a recruiter. We ran this study because we already analyze resumes at volume and the parse-stage failure rates are not something the usual resume advice quantifies.
Citing this data
Free to use with attribution and a link back to this page.
ResumeAtlas, “What Breaks ATS Resume Parsing? A 2026 ResumeAtlas Benchmark” (2026-09-20). Analysis of 218 submitted resumes through the ResumeAtlas ATS-parse analyzer, June 2026 – September 2026, and 1,085 resume analyses, March 2026 – June 2026. https://resumeatlas.io/ats-resume-parsing-benchmark
Related from ResumeAtlas
Resume statistics 2026
The companion dataset — the gap between what postings ask for and what resumes contain, from 1,085 analyses.
What is a good ATS score?
Where 75, 80 and 85 sit across real checks, the three things called an ATS score, and why tools disagree.
ATS-friendly fonts, tested
One resume in 20 fonts and five extractors: the font changed nothing, 2pt of letter-spacing broke the headings.
Are Canva resumes ATS-friendly?
28 Canva templates parsed one by one: which pass, and the two mechanisms behind the rest.
How ATS scans and reads your resume
The four stages, from file ingestion to recruiter view — the mechanism behind these numbers.
Common resume mistakes that fail ATS
The formatting choices behind the flagged checks above, and how to undo each one.
30-point ATS resume checklist
A pre-submit checklist covering every failure category measured above.
ATS rejection reasons
When the file parses clean but the content still doesn't match the posting.
ATS resume parser
Run your own resume through the analyzer used in this study.