We Checked 44,000 Resumes Against an ATS Parser. Only 5.8% Passed Cleanly

Between November 2025 and September 2026, 44,255 resumes were uploaded to OwlApply's free ATS checker. The checker does what an applicant tracking system does: it extracts the text from the file, splits it into sections and fields, and runs 34 checks on what it finds. This report is the aggregate of those 44,255 runs. No individual resume is quoted; every number below is a count or an average across the whole set, and the underlying files were processed only by the checker itself.

The short version: only 5.8% of PDF resumes came through the layout check with nothing flagged, one in three used a table for layout, and the single most common content problem was not a missing keyword but a listed skill that appears nowhere else in the document. Everything in this article is a number you can cite; the methodology and its limits are at the end.

How resumes actually score

The checker grades on a 100-point scale across layout, structure, content and keywords. The distribution is not a bell curve centred on "fine": it is a hump between 50 and 70, with a long tail below.

Scores cluster in the 50–70 band: 12,792 resumes (28.9%) landed in the 60s and 10,147 (22.9%) in the 70s. Only 1,991 resumes (4.5%) scored 90 or above, and 207 scored a perfect 100. Recruiters do not see a score, but they do see the two things the score measures: whether the parser found the fields, and whether the bullets say anything measurable.

What breaks in the file: layout findings on 22,901 PDFs

Since May 2026 the checker inspects the PDF itself, not only the extracted text: column geometry, tables, fonts, header and footer bands, images, hyperlinks and glyphs. 22,901 files were analysed this way. The table lists how many of them triggered each finding; a file can trigger several.

Two of these numbers deserve a second look. The table layout finding is critical and hits one resume in three, which is a direct consequence of template marketplaces: most "modern" templates position content with invisible tables. And the two-column layout, the problem every ATS guide warns about, is only 8.8% of files; the far more common damage comes from tables, headings and fonts that look perfectly ordinary on screen.

What breaks in the content: 23,239 resumes since late May 2026

Once the text is out of the file, the checker looks at what it says. These findings come from the 23,239 checks run since 28 May 2026, when the current issue model was introduced, so the shares are comparable with each other.

The 63.4% figure is the most useful one in this report. It is not about missing keywords: these candidates have the skill on the page. They list it in a Skills block and then describe their jobs without ever using it. Keyword-matching systems count the term once; a recruiter searching for evidence finds none. Moving three skills from the list into the bullets where they were actually used is the highest-return edit most people can make.

The three "section not detected" rows (experience 11.3%, education 7.2%, skills 6.0%) are the quiet failures. The content exists, the candidate can see it, and the parser filed it under nothing, usually because the heading was creative, sat inside a table, or was rendered as an image.

Where resumes lose the most points

Averaging each scoring category across all 44,255 checks shows where the gap between a typical resume and a strong one really is. Layout basics are mostly fine; evidence is not.

Personal information and section titles score well because the checker only asks whether they are present and standard. The two categories at the top of the table are where the score is decided, and both are about the same thing: proof. A bullet that says what happened and by how much, using the words the posting uses.

The fixes the checker recommends most often

For every resume the checker ranks the edits by expected gain. Across the 23,171 checks with ranked recommendations, these came up most:

• Add quantified achievements — recommended to 54.1% of resumes. The average resume in the set had almost no bullets with a number in them.
• Demonstrate listed skills in context — 30.4%.
• Trim the skills list — 27.1%. Long lists dilute the terms that matter and trip the stuffing detector.
• Complete education entries (degree, school, year) — 26.4%.
• Add the missing keywords from the target posting — 25.5%.
• Start bullets with a strong verb — 24.8%.
• Complete experience entries (title, company, dates) — 22.9%.
• Add a country code to the phone number — 19.4%, a reminder that a large share of these candidates apply across borders.
• Standardise date formats — 18.4%.
• Add a summary — 17.1% had none.

So what does "ATS-friendly" mean, in numbers?

Reading the two halves of the data together, a resume that would have passed every major check in this set looks like this:

You can test any file against the same 34 checks with the free ATS resume checker; it shows the extracted text line by line, which is the fastest way to see whether your own resume is in the 5.8% or the 94.2%. If you are starting from scratch, the ATS-friendly resume templates are built to the profile above, and the job description match tool handles the keyword half for a specific posting.

Methodology and limitations

• Source: 44,255 uploads to the public OwlApply ATS checker between 3 November 2025 and 14 September 2026. Files are uploaded voluntarily by job seekers; no resumes from the OwlApply builder or from user accounts are included.
• What was measured: the checker extracts text from the file the way an ATS does, maps it to sections and fields, and applies 34 rules covering layout, structure, content and keywords. Layout rules run on the PDF geometry itself and were introduced in May 2026, so layout figures cover the 22,901 files analysed since then; content-issue shares cover the 23,239 checks since 28 May 2026, when the current issue model shipped; score and grade figures cover all 44,255.
• Privacy: only aggregate counts and averages were computed. Uploaded files are processed by the checker and are not read by anyone; the analysis queried check results, not resume contents.
• Selection bias: people who use an ATS checker suspect a problem. The set almost certainly over-represents resumes with issues compared with all resumes submitted to employers, so treat the shares as an upper bound for the general population and a fair picture of active job seekers.
• Not a vendor's parser: the checker models common parser behaviour (Workday, Greenhouse, Lever, Taleo, iCIMS and SmartRecruiters all handle clean single-column PDFs; all lose data on the failure modes above), but no third-party tool has access to any vendor's internal ranking. A score here is a diagnostic, not a prediction of any specific system's verdict. For how the systems themselves work, see how ATS works in 2026.
• Citing this report: you are welcome to quote any figure with a link to this page. The dataset is updated as new checks accumulate; the numbers above are as of 14 September 2026.

Frequently asked questions