Load a .docx and its pages are laid out in front of you. Add more documents and their pages join the same grid. Drag everything into the order it should be in, drop what shouldn't be there, and save one Word file. Text, formatting, tables and pictures come across as they are.
Merging Word documents by hand is a bad afternoon. Select all, copy, paste, and the headings come through the wrong size, the numbered list restarts at one, the picture lands in the margin and the page breaks end up somewhere nobody asked for. Then you do it four more times because there were five documents. This does the same job as a rearrangement instead of a retype: the pages of every document you load are laid out together, you put them in order, and one file comes back.
This is the part worth reading, because it is the one thing about editing Word documents that surprises people.
A .docx file does not contain pages. It contains one long stream of paragraphs and tables, and Word decides where the pages fall while it is drawing the document — from the fonts installed, the paper size, the printer driver, the widow-and-orphan settings. Open the same file on two machines and the page breaks can land in different places. There is no "page 4" stored in the file to go and fetch. That is why no tool can rearrange the pages of a Word document the way it can rearrange the pages of a PDF, and why anything claiming otherwise is either running a copy of Word on a server or hoping you won't check.
What a .docx does contain is every page boundary somebody wrote down, and this tool works from those. There are four:
When the document you load has any of these, the page cards you see match them, and this page says which kind it found. When a document has none of them — a plain flow of prose that Word paginates on the fly — it comes back as a single card, and says so rather than inventing boundaries. If you need that document split, open it in Word and put page breaks where you want them; they are what makes a page a thing that can be moved.
This follows from the above, and it is a real change to your document, so it is worth being plain about. If the pages you dragged around were only ever rendered apart — Word's record of its own layout, not breaks anyone typed — then the file that comes back has real page breaks written in where those boundaries were. It has to: the whole point of moving page 7 above page 2 is that it stays there, and without a break in the file Word would simply reflow the text back into one stream and undo the work.
So the document you get back is more firmly paginated than the one you put in. Add a paragraph to the middle of it later and the text will not spill onto the next page the way it used to — it will push against a break instead. That is the right trade for assembling a finished document, and the wrong one for a draft you are still writing.
This is the job usually called merging, and merging is the blunt version of it: two files, end to end, in whatever order you happened to upload them. What people want is nearly always narrower — the signed page from this one, in place of the unsigned page in that one. So a second document isn't appended and sealed: its pages arrive as cards alongside the ones already there, carrying the file they came from, and you keep what you want and drop the rest. Ten documents can feed one file, and a page can come in more than once.
Pages keep the paper they were set on. A landscape page dropped into a portrait document stays landscape, and a page from a Letter document stays Letter inside an A4 one, because resizing it would mean re-laying-out everything on it. Word handles a mixed document perfectly well — it is the same mechanism as a landscape appendix. Headers and footers travel with their pages too, so a page from another document keeps the header it had rather than silently adopting yours.
The document you upload first is the one being built on, and its styles are the ones the finished file keeps. So if both documents define Heading 1 — and nearly every Word document does — incoming pages take on the look of the document they are joining. That is almost always what you want when you are assembling one report out of several: consistent headings rather than a patchwork. This page tells you how many styles it had to reconcile before you save, so it is never a surprise.
Everything a page carries directly — a bold word, a font size set by hand, a colour, a table's borders, an indent — comes across untouched, because that formatting is part of the page rather than a reference to something outside it. Styles the other document uses that yours has never heard of are brought across as they are. Bulleted and numbered lists keep their own numbering rather than folding into a list they aren't part of, and footnotes come across with the pages that reference them.
Comments on pages from another document. A comment is anchored by an id into a list of comments belonging to the file it was written in, with its own threads and its own authors. There is no id an arriving comment could take here that would still mean the same conversation, and a comment that silently reattaches to the wrong sentence is worse than one that is gone — so they are stripped, and this page says when it has happened. Comments in the document you started from stay put, because they are anchored to text, not to page numbers, so moving pages around doesn't disturb them; the exception is a comment whose text was on a page you deleted, and that one is counted for you too.
Tracked changes come across as tracked changes — they are not accepted or rejected on your behalf. If you want a clean file, accept them in Word first.
A document that needs a password to be opened can't be read here, and won't be: there is nothing to show you. Open it in Word with the password, save an unprotected copy, and rearrange that.
Every document you upload — the one you're building on and any you take pages from — is read for its page list, used when you ask, and deleted within the hour by a sweep that runs whether or not you come back. Nobody opens it, and nothing out of it is stored — the usage log keeps the tool name, the file size, and nothing else. The text previews on the cards are built on the server and sent to your browser; they are not written down anywhere.