The resume table problem: what 22,901 PDFs looked like to a parser
Yesterday I posted about the three steps a tracking system runs on a resume PDF: extract text, detect sections, match. Several people asked the obvious follow-up: how often does step one actually fail on real files? We now have a number. Between May and September 2026, 22,901 PDF resumes went through the free checker I run, and since May the checker inspects the file itself (column geometry,…
A recent study analyzed the performance of a resume parser when examining 22,901 PDF files. The parser checks for text extraction issues, section detection, and matches. Between May and September 2026, only 5.8% of the PDFs had no layout problems. A majority, 79.8%, had at least one major or critical finding.
The most common issue was two-column or sidebar layouts, found in 8.8% of files. This causes problems because the parser reads left to right across both columns, potentially combining unrelated sections. Tables used for layout were flagged in 32.7% of files, which is almost four times as common as the two-column layout. This occurs because most people use word processor templates, which are table-based, and the parser treats these tables as separate cells when reading the PDF.
Another significant issue was tables used for layout, found in 32.7% of files. Tables are often used for layout purposes, but the parser reads them row by row, potentially causing errors in section detection and job title association.
The findings also revealed that decorative bullets, icons, or emojis in the resume were found in 48.9% of files, which can break section detection. Font issues, such as non-standard or under 9 pt fonts, were found in 46.0% of files, potentially shifting glyph widths and causing extraction problems. Non-standard section headings were flagged in 39.2% of files, as they may not match the parser's expected headings.
The study also found that 32.5% of files had hyperlinks with hidden URLs, which could cause issues if the file is ever converted to text or printed. Inconsistent or unsafe date formats were found in 31.4% of files, as parsers may infer patterns from the first date and cause later entries to be blank. Contact details in the header or footer band were found in 22.1% of files, as many parsers only read the body layer.
Images in the skills area, such as skill bars or rating dots, were found in 16.3% of files, as these are often deleted when the image is stripped. Ligature or private-use glyphs were found in 11.5% of files, which can cause minor extraction issues. The full report, including the content findings and methodology, is available at https://owlapply.com/en/blog/ats-resume-statistics.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
