What to expect when converting PDF to Word

PDF to Word is one of the most requested conversions and one of the most oversold. Understanding why it is hard tells you exactly what you can expect from it.

PDF does not store documents the way you think

A Word file stores meaning: this is a heading, this is a paragraph, this is a table with four columns. A PDF stores appearance: this glyph goes at these coordinates in this font at this size.

By the time a document becomes a PDF, the structure has been discarded. What remains is a very precise description of where every mark sits on the page. There is no heading in a PDF, there is text that happens to be larger.

Which means conversion is inference

Turning that back into an editable document means guessing at structure from geometry. Text noticeably larger than its surroundings was probably a heading. A wide vertical gap probably separates paragraphs. Text starting with a bullet character is probably a list item. Fragments lining up in columns across several rows are probably a table.

These rules work well on ordinary documents and badly on unusual ones. A two-column academic paper reads straight down and interleaves the columns. A table whose columns do not quite line up comes across as prose. A caption in a slightly larger font becomes a heading.

What you should reasonably expect

Reliable: the words, in the right order, on ordinary single-column documents. Headings identified as headings. Bulleted lists as lists. Simple tables as tables.

Unreliable: multi-column layouts, complex tables, footnotes, anything where reading order differs from position on the page.

Not attempted by anything honest: exact fonts, spacing, images in position, colours, page design.

The useful mental model is that you are recovering the content to work with, not producing a copy of the original that happens to be editable.

When it will not work at all

A scanned document has no text to recover. The pages are images, and conversion will produce an empty document or a document full of nothing. Run recognition first to create a text layer, then convert.

This catches people out because a scan looks identical to a digital PDF on screen. The test is whether you can select a word with your cursor.

Doing it well

Convert, then read the result against the original before editing. Fix the structure first, headings, lists, tables, and the formatting afterwards, because reformatting content you are about to restructure is wasted work.

If the document was originally made in Word and you have any route back to that file, use it. Conversion is for when you do not.

What the conversion is inferring, in detail

It helps to know the specific rules, because then you can predict the failures.

Lines are formed by grouping text fragments that share a baseline. Fragments are joined with a space when the horizontal gap between them exceeds roughly a quarter of the font size, which is how PDFs that store each word separately are reassembled.

Headings are identified by comparing a line's height to the document's median. Noticeably larger means a heading, and how much larger sets the level.

Paragraphs are broken by vertical gaps. Consecutive lines close together are joined; a wide gap starts a new block.

Lists are recognised by a leading bullet character or a number followed by a full stop or bracket.

Tables are found where fragments line up at consistent horizontal positions across several consecutive rows.

Every one of those is a heuristic, and knowing them tells you exactly which documents will convert badly.

Documents that convert well and badly

Well: single-column reports, letters, contracts, articles, anything produced from a word processor with conventional structure.

Badly: academic papers in two columns, newsletters with sidebars, forms, anything with text in boxes positioned around a page, and tables whose columns are visually clear but not actually aligned.

Not at all: scanned documents, which have no text to recover.

Cleaning up efficiently

Fix structure before formatting. Restyling text you are about to reorganise is wasted work.

Read the converted document against the original once, looking specifically at the boundaries: where sections start, where lists begin and end, where tables sit. Those are where the inference goes wrong, and errors elsewhere are rare.

Then apply styles. If the document has real headings, applying a Word heading style to each one takes minutes and gives you a navigable document.

The question worth asking first

Does the original still exist? Conversion is for when it does not.

If the document came from a colleague, a client or a system that produced it from a Word file, asking for that file takes one message and gives you something conversion can never match. It is worth doing before spending an hour on cleanup.