PDF to Word

Not closely. The DOCX carries the words as plain paragraphs, with the pictures on lines of their own between them; fonts, sizes, colours, tables as tables, link targets and drawings made of lines and shapes stay behind. A paragraph set wholly in bold or italics keeps it; a bold word inside a sentence does not. The original's indentation and spacing are not kept, and a picture painted over another, or over text, comes through on its own line instead of on top. What the page paints over a picture, such as a box over a name, is drawn into that picture, so what it covers stays covered. Each PDF page with text or a picture begins on a new Word page, and running headers and page numbers come through as ordinary paragraphs.

Mostly. A plain JPEG photo up to 2,400 pixels on its longer side goes into the DOCX byte for byte. A larger one is scaled down to 2,400 pixels on that side, a black-and-white scan to 3,600, and a picture that is not a plain JPEG is re-saved, as a PNG or, for a photograph, a JPEG. A one-colour stencil comes through in its colour, and a picture the page repeats comes through each time it is shown. A picture the page shows only part of is cropped in Word the same way, so the hidden part can be uncropped there, unless something is painted over the picture: then it is re-saved as the page shows it, at its own resolution, and what is covered is not in the file. A picture too large for our server, or one it cannot decode, is left out, and the note beside the download names its page, except on a page our server cannot read before it is drawn, as in a damaged PDF, where a picture can be left out with no note. Past 2,000 pictures, the size of one file, or the time one conversion may take, a document's remaining pictures are left out too and counted in the same note; the text still comes through whole.

A scan has no selectable text, only a picture of its words. In a PDF that also has typed pages, a scanned page comes through as that picture, on its own Word page, with no editable words, and beside the download the panel says how many pages had text and names the ones that came through only as pictures (up to 50, then a count of the rest). If no page has any selectable text, the job is turned away instead, since a document of page pictures would have nothing in it to edit; lettering converted to vector outlines has the same effect. Tick “Recognise text in scans (OCR)” to have the words as text: our server runs OCR first, keeping pages that already have text as they are, and converts its result. OCR in the wrong language does not fail, it produces plausible-looking gibberish, so choose every language printed on the page. Arabic, Hebrew, Persian, and Urdu are not offered for it: written right to left, their recognised words come out of the conversion out of order. OCR PDF reads them in order.

In reading order. A page set in two columns is read one column at a time. A table whose cells each fit on one line comes through as one paragraph per row, with its cells separated by tabs rather than placed in Word table cells; a table whose cells wrap onto a second line is read column by column instead. For real cells, use PDF to Excel or Extract Tables; for plain words with no upload at all, PDF to Text runs in your browser.

Not directly. Our server cannot read text it cannot decrypt, so the job stops at once, without retrying, with a message that says the PDF is password-protected and a link to Unlock PDF. If you know the password, remove it with Unlock PDF and convert the decrypted copy. Unlock PDF cannot recover or bypass a password you do not have. A PDF that opens without a password and only restricts editing converts as it is, pictures included. A PDF whose structure our server cannot read is rebuilt first, the way Repair PDF does, and the rebuilt copy is converted, with a note beside the download that says so; one that cannot be rebuilt into a readable file is not converted, and the message suggests Repair PDF.

No. Our server erases your uploaded PDF when the conversion ends. The DOCX stays retrievable for 60 minutes, then the link lapses and a daily storage rule wipes the leftover file. No sign-up is involved.

Up to 100 MB per file.

Its selectable text and its pictures. The text arrives as paragraphs in reading order: lines that wrap in the PDF are joined into one paragraph, a page set in two columns is read one column at a time, and a table whose cells each fit on one line comes through one row per line, its cells separated by tabs. Each picture on a page, such as a photo, a screenshot or a scan, comes through on a line of its own between the paragraphs above and below it, at the size the page shows it, scaled down if it is wider or taller than the Word page, with anything the page draws over it, such as a box or an arrow, drawn into it. Past 2,000 pictures, or past the size of one file or the time one conversion may take, a document's remaining pictures are left out and counted in the note beside the download; the text still comes through whole. Tables, fonts, colors, drawings made of lines and shapes elsewhere on the page, and the page layout are not carried over, and bold or italics only where a whole paragraph is set in them, so the text arrives in Word's default style, ready to format — not as a copy of the page.

Yes, with “Recognise text in scans (OCR)” turned on. On its own a scanned page is a picture with no selectable text: a PDF with none at all is refused, since a document of page pictures would have nothing to edit, and in a PDF that mixes the two, each scanned page comes through as its picture, with no editable words. With the switch on, our server runs OCR PDF's engine first, in the languages you pick, and converts its result in the same job: each page's scan then comes through as a picture, followed by its words as text — recognised lines are not always joined back into paragraphs, so a scanned line can arrive as a paragraph of its own.