PDF to Word

How to use PDF to Word

Key features of PDF to Word

About PDF to Word

PDF to Word reads each page's text with pdf.js, groups it into lines by baseline and puts each line's words in order from left to right, whatever order the PDF drew them in. On a page with a clear gutter between two columns, the left column is read first. A new paragraph starts at a clearly larger gap between two lines, at an indented line, or after a line that stopped short although the next word would have fitted. Short lines that each start with a capital or a digit, such as an address, stay separate.

The detection is heuristic, so a poem, a complex table or an unusual layout may come out with lines joined or split differently from the original.

When to use PDF to Word

A supplier sends a contract as a PDF and wants your changes back. Convert it, open the .docx, switch on Track Changes and edit the clauses; the headings show up in Word's navigation pane. Keep the original PDF next to it, because the page layout is not reproduced.

Only the PDF of an old report, thesis chapter or article survives and you want to reuse its paragraphs in a new document. Convert it and copy the text across, restyling it in the new file's template.

If you need only the raw words, for a search, a script or a word count, PDF to Text is quicker and has no options step. A table of figures is better served by PDF to Excel, which puts the values into cells; here each table row becomes one line, with tabs at the wide gaps between cells.

A scanned letter or a photographed page has no text layer, and the tool stops with a message that says so. Run it through OCR PDF first, which adds the text layer this tool reads. OCR text comes out plainer: it has no bold or italic, list bullets are often lost, a heading is recognised only from about one and a half times the body size, and a short title above two columns can come out split. In a file that mixes typed and scanned pages, each scanned page is marked with a line saying it has no text layer.

Tips and troubleshooting

• Failed to convert to Word: No password given means the PDF needs a password to open. The tool has no password field; remove the protection and convert the copy. Restrictions on printing or copying do not block the conversion.
• Failed to convert to Word: Invalid PDF structure means the dropped file is not a PDF. Drag and drop accepts any file type, unlike the picker.
• Bullets become a Word bullet list, nested by their indent. Numbered items keep their printed numbers, with a hanging indent, so Word will not renumber them. A line that starts with a number or a hyphen, with no list item next to it, stays an ordinary paragraph.
• With Start each PDF page on a new Word page on, a paragraph that runs over a page end stays in two parts. With it off and headers and footers left out, the parts are joined when the page's last line reaches the right margin and the sentence carries on.
• Leave out running headers, footers and page numbers only removes lines set apart from the text near the top or bottom edge: a page number such as 3 or Page 3 of 9, or the same line, numbers aside, on three or more pages.
• A word hyphenated at the end of a line is joined without the hyphen when the next line starts with a small letter. A compound such as self-aware that breaks at its own hyphen loses it too, unless the file has the same word with its hyphen elsewhere.
• Columns are recognised only when most lines on each side fill their column; any other page is read line by line across. The result card has no Preview Document button, since a .docx cannot be previewed here.
• If you only need headings and paragraphs for a notes app or a repository, PDF to Markdown gives a .md file from the same text layer.
• Not sure whether a PDF is a scan? The guide how to make a scanned PDF searchable starts with a quick check, then shows how to add the text layer this converter needs.

How PDF to Word works in your browser

pdf.js, running in a worker fetched from cdnjs.cloudflare.com, returns each page's text as runs with a position, a size and a font. Bold and italic come from the font's name, which pdf.js reports only after it has worked out how to draw a page; the script asks for that, with images skipped, for each page that uses a font it has not seen yet. A gap wider than about a sixth of the font size becomes a space, so a word drawn in pieces stays one word. Hebrew and Arabic lines are read from right to left.

JSZip then writes the package: [Content_Types].xml, the relationships, word/document.xml, word/styles.xml with the Normal, Heading 1 to 3 and List Paragraph styles, and word/numbering.xml for the bullets. Characters that XML does not allow, such as control codes, are removed first, because one of them makes Word refuse the file. The PDF and the text stay in the tab's memory; no server sees them.

Frequently asked questions