Word to PDF

Why this conversion normally requires an upload

Almost every free Word-to-PDF service works the same way: your file is uploaded, a copy of LibreOffice on a server opens it and prints it to PDF, and you are given a download link. It produces good output, because LibreOffice is a full word processor with the same layout engine assumptions Word has. It also means your document — which, for the files people actually convert, is often a contract, an offer letter, a medical report, a signed agreement or a CV with a home address and phone number on it — sits on somebody else's disk for a while.

This page takes the harder route so that no upload happens. A .docx file is not a mysterious binary; it is a ZIP archive containing XML. The archive is unpacked in your browser, the document XML is read into a structured model of headings, paragraphs, runs, lists, tables and images, and that model is then typeset onto PDF pages by a layout engine running in the same tab. Nothing crosses the network. You can verify it the direct way: load this page, disconnect from the internet, and convert a file anyway.

The trade-off is honest and worth stating up front. A server running LibreOffice reproduces Word's pagination closely. This reflows your document instead — the words, the structure and the images are all there, but a line may break in a different place and a page may end a paragraph earlier or later than Word would. For the vast majority of documents that is invisible. For a form that must land exactly on a grid, it is not, and Word's own File → Save as PDF is the right tool for that job.

What survives the conversion

Headings keep their level, whether the document used Word's built-in Heading 1–6 styles, a renamed style with an outline level set, or a translated locale's own heading names. That matters more than it sounds: heading levels are what give the PDF a usable structure for screen readers and for anyone skimming with a search.

Bold, italic and underline are carried run by run, which is why emphasis inside a sentence survives rather than being flattened. Hyperlinks are kept as real, clickable annotations with the underlying URL intact — not just blue text. Bulleted and numbered lists keep their nesting up to five levels, and the numbering format follows the document's own definition, so a list numbered a, b, c does not come back as 1, 2, 3.

Tables are rebuilt rather than screenshotted. Columns are sized from their content, so a narrow quantity column stays narrow; a header row is shaded and repeated at the top of each page it continues onto; and a row too tall for the space left on a page is split across the break instead of being clipped. Images embedded in the document are re-encoded where necessary and placed inline at a width that fits the text column.

What is deliberately not carried across: headers and footers, footnotes and endnotes, comments and tracked changes, text boxes and floating shapes, columns, charts, SmartArt, and the exact fonts. Anything the reader could only partially represent is reported in a warning above the download rather than quietly dropped — if your document has footnotes, the tool says so.

Fonts, Hindi, and other non-Latin scripts

A PDF either embeds the fonts it uses or relies on the fourteen standard fonts every PDF reader has built in. Embedding a font copies it into the file, which is both large and, for many commercial fonts, a licensing question. This tool uses the standard fonts — Helvetica, Times or Courier, your choice — which is why a converted document is usually tens of kilobytes rather than megabytes.

Those fonts can only draw Western European characters. That is fine for English, and for the curly quotes, dashes, accented letters and currency symbols a word processor inserts automatically, all of which convert cleanly. It is not fine for Devanagari, Bengali, Tamil, Arabic, Chinese, Japanese, Korean or emoji, none of which exist in those fonts.

So the tool has a switch, on by default: "Draw non-Latin text as an image". With it on, any line the standard fonts cannot write is rendered by your browser's own text engine — which shapes Hindi conjuncts and CJK correctly — and placed on the page as a picture of that line. The document looks right; the text on those particular lines is not selectable. With the switch off, those characters are dropped and the tool tells you exactly how many and which script they belonged to. Neither option is perfect, which is why you get to choose rather than being surprised.

Old .doc files, and other formats that need one more step

Only .docx works here. The older .doc format from Word 97–2003 is a completely different, binary, undocumented-for-decades structure, not a ZIP of XML, and reading it in a browser is not realistic. If you have one, open it in Word, LibreOffice or Google Docs and use Save As to produce a .docx — a ten-second step that every one of those programs does for free.

The same applies to OpenDocument .odt files and to RTF: any word processor can save them as .docx. Apple Pages needs File → Export To → Word. In each case the tool recognises the file and says which step to take, rather than failing with a generic error.

There is no file size limit, because there is no server paying for bandwidth. The real ceiling is your device's memory, since the archive, the parsed model and the finished PDF all live in RAM at once. A 200-page report with a few dozen images is comfortable on a laptop; a document containing hundreds of high-resolution photographs may be happier converted in two halves and merged afterwards.