PDF vs DOCX resumes: a reproducible parsing comparison
Four generated files, two named parsers and eight text anchors. All four preserved the tested anchors and their order in this small experiment.
The direct answer
In this test, neither format lost the eight selected anchors. Single-column and two-column DOCX files passed Mammoth extraction; PDFs exported from those files passed PDF.js extraction. This is evidence about these exact fixtures and parser versions. It does not establish that every ATS parses either format correctly or that two-column resumes are universally safe.
What we tested
We created one fictional resume in two layouts using python-docx, then exported each DOCX through LibreOffice to a text-based PDF. The content uses normal paragraphs and a native column break, not floating text boxes or scanned images. On October 23, 2026, Mammoth 1.12.2 extracted DOCX text and PDF.js 5.4.624 extracted PDF text under Node 22.23.2. Whitespace was normalized before exact matching.
| File | Parser | Anchors found | Relative order |
|---|---|---|---|
| single-column.docx | Mammoth 1.12.2 | 8 / 8 | Preserved |
| single-column.pdf | PDF.js 5.4.624 | 8 / 8 | Preserved |
| two-column.docx | Mammoth 1.12.2 | 8 / 8 | Preserved |
| two-column.pdf | PDF.js 5.4.624 | 8 / 8 | Preserved |
Reproduce the result
Download all four files, parsing-anchors.json and compare-parsing.mjs into one directory. With Node 22, install the pinned parsers using npm install mammoth@1.12.2 pdfjs-dist@5.4.624, then run node compare-parsing.mjs. It writes a plain-text extraction for each file and parsing-results.json with SHA-256 hashes, positions, counts and parser versions. The supplied results are the output of the executed test, not a predicted outcome.
What the check cannot tell you
Presence and order do not prove that an application assigned employment dates, employer names and skills to the correct structured fields. The sample has one fictional resume, two layouts and two extraction libraries; it includes no commercial ATS, OCR, image-only PDF, tables, complex headers or unusual fonts. Successful extraction does not score relevance or predict hiring decisions.
Use the employer’s instructions first
Submit the requested format. If both are allowed, export a text-based document, inspect it visually, copy its text into a plain-text editor, and review the application’s parsed fields before submitting. Keep a clean DOCX source and recheck the final PDF after each layout change. If you see missing text or a confusing order, simplify the layout and test again.