PDF → Word

Convert PDF text into an editable .docx document. Free, no upload. 1 GB max · Up to 1 GB · Processed locally, never sent to a server.

PDF is a destination format: it freezes a layout so that it appears identically everywhere. The day you have to correct an amount in a quote whose source file has vanished, that very quality turns against you. PDF → Word then recovers what can be recovered, which is the text, and delivers it in an editable .docx document.

The tool opens the document with PDF.js, asks each page for its text layer, rebuilds lines and then paragraphs from the vertical position of every fragment, and writes a minimal .docx that JSZip assembles on the fly. The file produced takes the name of the PDF, and the note tells you how many paragraphs were extracted.

It is worth saying plainly what this tool is not. It is not a reconstruction of the layout: fonts, sizes, bold, images and table grids do not make it into the Word document. What you get back is clean text, cut into paragraphs, which you then format yourself. Services that promise the full layout run structure recognition engines hosted on their own servers; here everything happens inside the tab, starting from the text layer.

How to use it

  1. Check that the PDF contains text Open it and try to select a word. If selection is impossible, the document is a scan and extraction will find nothing at all.
  2. Drop the file The drop zone reads “Drop your file here”, with “Choose a file” below it and a line stating a ceiling of 1 GB and local processing. Complex layouts, columns and images included, come out simplified.
  3. Click “Run” Progress shows the page under way out of the total. Each page is walked exactly once, to collect its text fragments and their coordinates.
  4. Open the .docx and format it The document arrives in A4, with margins of 2.5 cm and eight points of spacing after every paragraph. The styles are yours to apply.
  5. If there is no text, go through OCR If fewer than five characters are detected, the tool stops and points you to OCR — scanned PDF, which recognises the text of pages that have become images.

How the paragraphs are rebuilt

PDF.js does not return lines but fragments, each one carrying its position matrix. The tool rounds the vertical coordinate of every fragment, groups into the same line all those that fall within four units of each other, then glues the pieces back together, reducing runs of spaces to a single one.

The split into paragraphs then follows a punctuation rule: a line ending in a full stop, an exclamation mark, a question mark, a colon, a semicolon or a closing guillemet closes the current paragraph, as does a blank line. A heading with no final punctuation therefore ends up stuck to the paragraph that follows it, and a line ending on an abbreviation cuts too early. Worth noting for English documents: the closing quote mark the rule tests for is the French guillemet, so a line finishing on a closing double quote does not by itself end a paragraph. Between two pages, a page break is inserted.

What the .docx file really contains

A .docx is a ZIP archive of XML parts. This one holds the strict minimum:

There is no stylesheet, no numbering and no media folder. Each paragraph is written without any formatting, followed by 160 twips of spacing, and the closing section declares a page of 11,906 by 16,838 twips, which is an A4, margins of 1,417 twips included.

The package respects the structure a .docx is expected to have and opens in software that reads the format: your headings will become headings again once you apply a style to them, and not before.

Documents that come through badly

Two column layouts are the awkward case. Since lines are rebuilt by height, two columns sitting at the same level end up end to end on a single line: the text is complete, but the reading order is not.

Tables lose their grid: the cells of one row become a single sentence. Headers, footers and page numbers are extracted like any other text and reappear scattered through the document. Finally, some generators position characters one by one without ever encoding a space: the extracted text then shows words run together.

When this is not the right tool

If it is the layout you care about, do not convert at all: ask for the source file, or annotate the PDF instead. If the document is a scan, the detour through OCR is unavoidable. And to recover an illustration, PDF → JPG exports the whole page as an image, which you then only have to crop.

The round trip is not neutral either. The Word → PDF tool, which goes the other way with mammoth, renders the document into an image before slicing it into pages: its text is no longer selectable afterwards. A document converted and then converted back does not return to its starting state.

Frequently asked questions

Why are my headings no longer headings?

The generated document contains no stylesheet at all, and each paragraph is written without formatting. The word processor applies its default style to everything, so the hierarchy is yours to recreate.

Where have the images from the PDF gone?

They are not read, because only the text layer is walked. To get them back, export the pages concerned with PDF → JPG, then insert them into the Word document.

Two paragraphs have merged, how is that explained?

The break depends on the punctuation at the end of a line. A line that ends neither in a full stop, nor a semicolon, nor a closing guillemet carries on the current paragraph.

Does the paragraph count announced match my document?

It matches the number of blocks formed by the algorithm, page breaks excluded, and it can differ from the way the PDF looks to be divided.

Are accents and special characters preserved?

Yes: the text is taken exactly as PDF.js returns it, with the XML reserved characters escaped before writing. A character that was badly encoded in the original PDF will stay wrong.

My PDF is a scan, can extraction be forced?

No: below five detected characters, the tool stops. Run OCR — scanned PDF instead, then paste the recognised text into a blank document.

Similar tools

Discover other tools

Categories