Why PDF to Word breaks tables and columns

A PDF does not contain a document

It is tempting to think of a PDF as a document file in the same family as a Word file. It is not. A Word file is a description of a document's structure: this is a heading, this is a paragraph, this is a table with three columns and four rows, these cells are merged. The layout is computed from that structure when you open it, which is why a Word document reflows when you change the page size.

A PDF is the output of that computation, frozen. It is closer to a set of printing instructions: place this glyph at this coordinate in this font at this size, draw a line from here to here, fill this rectangle with this colour. There is no "table" object in a typical PDF. There is a set of glyphs positioned in a grid, and some lines drawn around them.

Converting a PDF to Word therefore cannot be a translation. It has to be an inference: the converter looks at where the marks are and reasons backwards about what they were supposed to mean. That inference is very good on simple documents and gets progressively less certain as the layout gets more interesting.

What the converter is actually doing

Our PDF to Word conversion runs on pdf2docx. It reads the text runs out of the PDF along with their positions, font sizes, and styling, then groups them into blocks based on how they sit relative to each other. Runs that share a baseline become a line. Lines with consistent spacing become a paragraph. Where it finds ruling lines or strong column alignment, it proposes a table and fits the text into cells.

On long documents the work is split across several worker processes, so a 200-page PDF does not take proportionally longer than a 20-page one.

Everything in that description is a heuristic. None of it is reading metadata, because there usually is not any to read.

The four things that break, and why

• Tables without ruling lines. If a table's columns are held apart by whitespace rather than drawn borders, the converter has to guess where the column boundaries are from alignment alone. Ragged content — a cell with two lines of text next to cells with one — makes that guess harder and is the usual cause of "my table became paragraphs".
• Multi-column layouts. Academic papers, newsletters, and magazine layouts store text in reading order that is not top-to-bottom. A converter that walks the page in the wrong order produces text that alternates between columns mid-sentence.
• Merged and nested cells. A merged cell spans a grid position that, from the outside, just looks like a missing boundary. Nested tables compound it. These need manual repair more often than not.
• Tables that break across pages. To the PDF these are two unrelated tables on two unrelated pages. There is nothing in the file connecting them, so they come back as two tables.

Scanned PDFs are a different problem entirely

If your PDF came from a scanner, a phone camera, or a fax, it does not contain text at all. It contains a photograph of text. There is nothing for pdf2docx to extract, so the conversion returns a document full of images and no editable words.

The fix is optical character recognition — running the image through an engine that recognises the shapes as characters and writes a real text layer. Use the searchable-PDF option on the PDF first, then convert the result to Word. It is two steps, but it is the difference between an editable document and a scrapbook.

You can tell which kind you have in about two seconds: open the PDF in any viewer and try to select a sentence with your cursor. If a selection highlight appears over the words, there is a text layer. If you get a rectangle over the whole page, it is an image.

How to get a better result

• If you only need the data, not the document, convert the PDF straight to a spreadsheet instead. Table extraction aimed at rows and columns does better on tabular pages than a general document conversion does.
• If you only need the words, convert to plain text. It is faster, it never invents structure, and it is the right input for anything you are going to reformat from scratch anyway.
• If the PDF was generated from a Word file that still exists somewhere, find that file. No conversion will ever beat the original.
• If you must convert a complex layout, expect to do cleanup, and budget for it. Fix the table structure first, then the styles — doing it the other way round means redoing the styles.

What a realistic expectation looks like

For a letter, a report, a contract, a CV, or any document that is mostly one column of prose with the occasional simple table, a PDF to Word conversion should land close enough that you are proofreading rather than rebuilding.

For a financial statement with merged header cells, a scientific paper in two columns with floating figures, or a designed brochure, the honest expectation is that you are getting a head start, not a finished document. That is worth knowing before you start, because it changes whether converting is the right move at all.