File Formats: PDF, DOCX, and Plain Text

“Should I send a PDF or a Word document?” is one of the oldest questions in job hunting, and most answers are a confident assertion in one direction with no reasoning attached. The honest answer depends on what each format stores, so let’s start there.

What a PDF actually is

A PDF is a description of a page. Its job is to guarantee that the document looks identical everywhere — same fonts, same line breaks, same spacing, on any device.

It achieves that by storing content as positioned drawing operations. Roughly: place this glyph run, in this font, at these coordinates. There is often no stored notion of “paragraph,” “column,” or “reading order.” Those are visual conclusions a human draws from the arrangement.

Consequences for parsing:

Extraction has to reconstruct order from geometry. Usually fine for a single column, unreliable for two. This is the root of the column problem covered in why your two-column resume breaks.
Words can arrive fragmented. Kerning and justification sometimes split a word into multiple runs. Good extractors stitch them back together; some leave you with Engi neer.
Ligatures and fancy typography can distort text. An fi ligature or a curly apostrophe usually survives, but not always cleanly.
A scanned or image-based PDF contains no text at all. If you printed your resume and scanned it, or exported it as an image, extraction returns nothing. This is rare but catastrophic when it happens.

Modern PDFs can carry structure — tagged PDF stores headings, lists, and reading order for accessibility purposes. Whether your export includes those tags depends on the tool and its settings, and whether a given parser uses them is anyone’s guess. Don’t count on it.

What a DOCX actually is

A DOCX is a zipped bundle of XML files describing a document, not a page. Its content model is a flow: paragraphs in order, each with a style, containing runs of text.

Consequences for parsing:

Reading order is stored, not inferred. The paragraphs are already in sequence. Nothing has to be reconstructed from coordinates, so the single biggest source of PDF extraction error simply isn’t present.
Headings can be identifiable. If you used real heading styles rather than manually bolding a line, the structure is explicit in the file.
Floating objects are still a problem. Text boxes, shapes, and headers/footers live outside the main flow and may be skipped, exactly as in a PDF.
Tables are still tables. Cell traversal order can still scramble a job entry.
Rendering varies. The same DOCX can look different depending on the reader and the fonts installed. Your careful one-page layout may reflow onto two.

So DOCX is structurally friendlier to a parser and visually less predictable to a human. PDF is the reverse. That is the whole trade-off, and it explains why the question has never been settled.

What plain text is good for

A .txt file has no formatting at all: no bold, no fonts, no bullets except the ones you type. Extraction is a non-event, because the file is the extracted text.

You will rarely submit one. But a plain-text version of your resume is genuinely useful in three places:

The paste-into-a-textbox field. Many application forms have a “paste your resume” box. Pasting from a formatted document often drags in stray characters and broken bullets; pasting from a clean text version doesn’t.
Email body applications. Same reason.
Testing. Producing a plain-text copy by hand and comparing it against what your PDF actually extracts is the fastest way to spot damage.

Keep one alongside your real resume. It takes ten minutes to make and it will save you an argument with a form field eventually.

Which one to send

If the form tells you, do what it says. This is not the place for independent judgement. A form that says “PDF only” may reject or mangle anything else.

If the form accepts either, send a PDF — provided you have run the extraction test on it and the text comes out clean. You get visual fidelity, and if the file parses well, you have given up nothing.

If your PDF fails the extraction test and you can’t fix the layout, send DOCX. Better a document that reflows a bit than one whose structure disintegrates.

If a recruiter asks for Word specifically, send Word. Agency recruiters frequently need to edit the file — adding their branding, removing your contact details before sending it to a client. That’s a workflow requirement, not a parsing question.

Never send: .pages, .odt, .rtf unless asked, image files of any kind, a link to a document instead of a document, or a Google Docs URL. Each of those creates a reason for someone to skip you that has nothing to do with your experience.

Exporting a PDF that behaves

If you’re going the PDF route, a few settings matter.

Export, don’t print-to-image. Use “Export as PDF” or “Save as PDF” from your editor, not a screenshot or a scan. Confirm afterwards that you can select the text with your cursor.
Embed standard fonts. Exotic fonts sometimes get substituted or converted to outlines. Outlined text is a picture and extracts as nothing.
Don’t password-protect it or restrict copying. A permissions flag that blocks text extraction will do exactly that to the parser.
Keep the file name sensible. Firstname-Lastname-Resume.pdf. It ends up in a folder with hundreds of others.
Check the file size. A resume with an embedded photo can balloon past what an upload form accepts. Under a megabyte is comfortable.

One myth worth retiring

You may have read that applicant tracking systems “can’t read PDFs.” That was closer to true fifteen years ago, when extraction libraries were weaker and PDF submission was less common. Today, PDF is the most commonly submitted resume format and mainstream parsers handle it routinely.

What’s true is narrower and more useful: PDF extraction is more sensitive to layout than DOCX extraction is. A single-column PDF is fine almost everywhere. A three-column PDF with a table and a floating text box is asking for trouble — and, importantly, so is the same design saved as a DOCX.

Format is the smaller variable. Layout is the bigger one. Get the layout right and the format question mostly stops mattering.