Convert a PDF to Word in your browser

No, and no browser tool can make it. You get the text of the document in reading order, in paragraphs, with headings marked as headings, bold and italic where the fonts recorded them, and the photographs back among the paragraphs they sat between. You do not get the layout: no tables as tables, no columns, no text boxes, nothing wrapped beside a picture, no exact positions, and not the original typeface. If what you want is a document that looks identical, you want the PDF you already have.

The photographs do. Each one is placed in the text where it was drawn, at the size it was drawn, shrunk to fit if it ran wider than the Word page's margins, and centred on its own line. Four kinds do not come across. Vector artwork, which is what most charts, logos and diagrams in a PDF are, is drawn with lines and curves rather than stored as a picture at all, so there is nothing in the file to take. An image covering the whole page is left behind, because that is a scan or a background rather than an illustration, and a Word file full of page photographs cannot be edited. Shapes used as stencils and images tiled across a page are skipped. And nothing is placed beside text: a picture goes on its own line, so text that wrapped around one in the PDF comes out above and below it.

Close, but they are not the original files. A PDF hands its images to a browser as decoded pixels rather than as the JPEG or PNG it stores inside itself, and there is no way for a page to reach those original bytes, so each picture is encoded again on the way into the Word document. On anything but a very large picture, whichever of PNG and JPEG comes out smaller is used, and past about four megapixels a photograph is always smaller as a JPEG and simply gets that. The exception is a picture with transparency, which only PNG can hold. For a photograph that means a second pass of JPEG at high quality: noticeable if you go hunting for it at 400%, and not otherwise. It also means the camera metadata that was in the original does not survive, and neither does a colour profile, so a CMYK image converted for print can shift slightly.

It comes out as ordinary paragraphs, in the order the cells were drawn. A PDF has no idea it contains a table: it has text at coordinates and, usually, some lines drawn near it, and reconstructing rows and columns from that is guesswork that goes wrong quietly. Producing a plausible looking Word table with the wrong cells in it would be worse than what this does, which is to hand you the contents and leave the structure to you.

Because lines are read across the page and then down it, so the first line of the left column is followed by the first line of the right one. Detecting columns reliably means telling a two column layout apart from a table, a sidebar and a figure caption, from position alone, and the failure mode is a document whose sentences are interleaved without saying so. Turning off "Join lines into paragraphs" at least keeps each line intact.

Because a scanned or photographed PDF has no text in it at all. Every page is an image, and the words in it are dark pixels rather than characters, so there is nothing for this to extract and a .docx made from it would be genuinely empty. Putting the page photographs into the document instead would be worse, not better: it would look converted, and it would be a stack of pictures nobody can edit, search or read aloud. So a page sized image is left behind wherever it turns up, and a document that is all of them is refused with this message. Reading it needs optical character recognition, which is a different job with its own error rate, and this tool does not do it rather than pretending the page came out blank.

No. Adobe has worked on this problem for two decades, reconstructs tables, redraws vector artwork and still gets tables wrong. What this does is the part that can be done in a browser tab with nothing uploaded: read the text layer, group it into paragraphs and headings, put the photographs back where they were, and say which of that was inferred. For a report, a letter or an article you want to edit, that is usually the whole job. For a form, an invoice or anything built out of a table, it is not.

If you only want the words, yes, and it is the better tool: reading the text layer is exact, nothing is interpreted, and the file opens anywhere. This page exists for when you need to edit the document in Word, which means paragraphs that reflow, headings that show up in the navigation pane, and bold that stays bold. Everything this adds on top of the text extraction is inference, so if the inference is not worth anything to you, take the text.

No. The PDF is decoded and the DOCX is written by your own browser, in this tab. There is no upload endpoint in the site for a file to go to, which you can check in your network tools: convert something and watch that no request carries it. Disconnect from the network after the page loads and it still works.

Yes, every time, and there is no way to turn it off. A Word document is written from scratch out of the text and pictures this tool extracted, every picture is encoded again from its pixels, and no properties part is written into the file at all, so EXIF and GPS, the XMP packet, embedded colour profiles, and any Content Credentials or AI-generation tags the file arrived with do not survive. There is no switch for this, because there is no way to keep them. What no compressor can remove is a watermark written into the pixels themselves, such as the invisible marks some image generators add: those survive re-encoding here and everywhere else.