Why PDF to Word loses your tables

The short answer

Almost every PDF stores the position of every character on the page and nothing about what any of it meant. There is no marker saying “this is a table” or “this is a heading”, only coordinates. A converter has to infer the structure from where things sit, and different converters infer differently. That is the whole reason results vary so much.

What a PDF actually stores

A Word file is a description of intent. It says: this paragraph uses the Heading 2 style, this run of text is a table cell, this list restarts its numbering. Where the words land on a page is worked out afterwards, which is why the same .docx reflows when you change the margins.

A PDF is the opposite. It is the finished page: this glyph at this coordinate, in this font, at this size. The structure that produced the page was used and thrown away. Converting back to Word means reconstructing information the file no longer contains.

There is an exception, and it is worth knowing because it explains why some documents convert beautifully and others fall apart. A PDF can be tagged (meaning it carries a hidden outline that names its headings, its tables and the order the page should be read in). It is the same information a screen reader needs, which is why accessible PDFs have it. A converter handed a properly tagged file is reading rather than guessing. But tagging is optional, most exporters do not do it well, and nothing on the outside of a file tells you whether yours has it. So the honest default is to assume your PDF is untagged, because it probably is.

A Word file is the recipe. A PDF is a photograph of the meal. You can work out a lot from the photograph, but you are working it out rather than reading it.

The three things a converter can do

Almost every tool advertised as “PDF to Word” is doing one of these. They are different jobs with different costs, and knowing which one you are being sold explains most of the disappointment.

• Text extraction: reads the characters and puts them in order. Fast, can run on your own device, and keeps nothing but the words. No tables, no images, no heading styles.
• Layout reconstruction: infers tables, headings, reading order and image placement, then builds a real document. Far heavier than extraction, and in practice every tool that offers it does the work on a server.
• OCR: for a scan, which is a picture of a page with no text in it at all. Nothing can be extracted until the characters are recognised first.

A tool doing the first will never give you the second, however it is marketed. If a converter is free, instant and never asks you to upload anything, it is extracting text.

What we measured

We built a deliberately hostile two-page document and ran it through three converters. It contained two-column justified body text, a table with a merged header cell, a chart image, six headings across three levels, nested lists, a hyperlink and a running header with page numbers: every element a text extractor is known to destroy. Then we counted what came back.

Two findings mattered more than the counts, and neither is visible when you open the file.

“Reconstruction B” lost text. In the two-column section it spliced the end of a left-hand column onto a word from the right, and a sentence present in the original was simply absent from the output. Nothing warned us. We only found it because we were searching for specific sentences on purpose.

The other finding is the one worth carrying away. A converter that cannot rebuild a layout can still look flawless by dropping every paragraph into a fixed box at the coordinates it had on the page. The result is pixel-perfect and close to unusable: text does not reflow, backspace does not pull the line up, and editing one sentence pushes the rest out of its box. None of the three did this, but many tools do, and you cannot see it in a screenshot.

How to check a converted file in under a minute

Four checks, no software beyond the Word file you already have. Each one catches a different failure, and the first two catch the ones that cost you most.

1. Search for a sentence from the middle of a two-column page. Copy a line out of the original PDF and search for it in the Word file. Two-column pages are where converters splice one column into the other, and a missing sentence is invisible until you go looking for it.
2. Click into a paragraph and press Enter. If the text below moves down, you have a real document. If nothing moves, every paragraph is sitting in its own fixed box. It will look perfect and fight you on every edit.
3. Open the navigation pane. In Word, View then Navigation Pane. An empty pane means your headings came back as ordinary bold text, so there is no outline and a generated table of contents will be blank.
4. Click a single table cell. If one cell highlights, the table was rebuilt. If the whole table selects at once like a photograph, it is a picture of a table and none of the numbers are editable.

Which one you actually need

If you want to quote a report, re-use a few paragraphs, or get the wording out of something somebody sent you as a PDF, text extraction is the right job and the cheapest one. It runs in your browser and your file never leaves your device.

If you have to keep the document (edit an invoice, update a contract template, fix a number in a table), you need reconstruction, and you are accepting two things: in practice the file goes to a server, and the result needs checking against the list above.

If the PDF is a scan, neither works until OCR has read the characters. A scan has no text in it to extract, so a text-only converter has nothing to find. Some say so, and some hand back an empty document without comment, which is the version that wastes your afternoon.

PDF to Word on PDFsansar does text extraction, in your browser. It gives you the words and not the design, and it says so on the page rather than after you have tried it. If a scan has no text layer it refuses instead of handing you an empty file.

Common questions

If your pages are also the wrong way round, sideways scans have a separate cause and a fix that loses nothing. And if the PDF will not let you select its text at all, that is a permission flag, not a lock.