Two ways in: drop a .docx and get semantic HTML plus every embedded image as a real file, or paste straight out of Word or Google Docs and get the mso junk stripped out. Markdown comes with both. Your file is read in this tab and never uploaded.
The document is unzipped and read inside your browser, and nothing about it is sent anywhere. The page downloads its converter code from a public CDN, part of it when the page opens and the rest the first time you convert a Word file, and it records an anonymous count of page views and conversions. Nothing else leaves this tab.
The ZIP always puts the files in an images/ folder. This box only changes the src written into the HTML, so you can point it at wherever you will actually upload them.
Non breaking spaces, zero width characters and soft hyphens are always replaced or removed. They are invisible, they are what make pasted text refuse to wrap properly, and there is never a reason to keep them in web copy.
unsupported/ folder in the ZIP and the image tag is left out rather than shipped broken.Copy a paragraph out of Word, paste it into a page builder, and you paste in far more than the words. Word puts a second, private version of the document on the clipboard: mso style properties, class names like MsoListParagraphCxSpFirst, empty o:p tags, hardcoded font sizes in points, and bullets that are not lists at all but paragraphs with a bullet character glued to the front. It looks fine for about a day, then the site is restyled and every pasted block ignores the new design because it carries its own inline styling. That is the mess this page removes.
The paste tab takes whatever your clipboard is holding and runs it through a strict allow list. Only real content tags survive, and only a handful of attributes on them: the link target, the image source and alt text, table spans. Everything else goes. Word list paragraphs are read back into real nested lists, and Google Docs formatting, which arrives as font-weight:700 on a span rather than a bold tag, is converted back into semantic markup so bold stays bold after the styles are stripped.
The file tab is a different job. A .docx is a zip archive, so it is opened directly and the document body is read from the XML rather than from any HTML Word generated. Heading styles become real heading tags, numbering becomes ordered lists, tables become tables, footnotes become a linked list at the end, and every embedded picture is pulled out as an actual PNG, JPEG, GIF or SVG file, named and numbered, with the HTML pointed at wherever you say you will upload them.
Most converters either drop images or bake them into the page as enormous base64 strings. Base64 is handy for a single quick paste and there is a switch here for it, but it inflates the page, cannot be cached by the browser, and cannot be given an alt attribute in a media library. Real files can. Duplicate images, the same logo used eleven times, are written once and referenced eleven times.
The same converted content is offered as Markdown, tables and all, which is what you want if the destination is a static site, a docs folder, a README or anything that stores content as text. Both formats come from the same cleaned structure, so they do not drift apart.
Word documents can contain things the web has no equivalent for. Equations, SmartArt and pasted Visio drawings are stored as objects or as Windows metafiles, and a browser cannot draw either. Rather than shipping image tags that render as a broken icon, those parts are written into a separate folder in the ZIP and left out of the HTML, and the page tells you it happened. Page furniture, headers, footers, fonts and colours are deliberately discarded. If you need a pixel copy of the Word page, convert it to a PDF instead.