Why bookmarks and links disappear when you merge or split a PDF

Bookmarks are not on the pages

A PDF has a catalogue: one object at the top of the file that says where everything is. The pages hang off it in a tree. So do several other things, and two of them are what this article is about.

The outline — what your reader shows as bookmarks in its side panel — is a tree of entries hanging off the catalogue. Each entry has a title and a destination, and the destination points at a page object. The page knows nothing about it. There is no mark on page 14 saying "Chapter 3 starts here"; there is an outline entry elsewhere in the file that happens to point at page 14.

Named destinations are a lookup table, also on the catalogue, mapping names like chapter3 to places in the document. They exist so a link can say "go to chapter3" rather than "go to page 14", which means the link survives the chapter moving. They are also how links from outside the file reach a particular place in it.

Both are properties of the document. Neither is a property of a page. That one sentence is the entire explanation for what happens next.

These tools build a new document

Merge, split, extract pages, remove pages and reorder pages all work the same way, in your browser: create an empty PDF, copy the pages you want into it, save it.

They do not edit your file. Nothing is deleted from anything — your original is untouched on your disk, and what you download is a new document that happens to contain the same pages. That design is why these tools are fast, why they work offline, and why they cannot corrupt the thing you started with.

It is also exactly why bookmarks do not come through. The copy operation takes pages. The outline and the name table were never attached to a page, so there is nothing to bring: the new document gets a brand new catalogue with an empty outline, and your reader, finding no outline, shows no bookmark panel.

The same reasoning explains a few other things people notice:

• Page labels do not come through either — the roman numerals on front matter, appendix prefixes, and anything else your reader showed in its page box. Same structure, same reason. Why PDF page numbers do not match the pages covers labels in detail.
• Form fields and signatures do not survive intact either, and a signature could not be allowed to: it certifies a particular document, and this is a different document. How PDF forms work covers that side.
• Printed page numbers do survive, because they are ink on the page, and they will now be wrong. That is not a malfunction, it is the same fact seen from the other end.

What does survive

Everything that is genuinely part of a page comes across, because the page is what is being copied:

• All the visible content — text, fonts, vector graphics, images, at full quality. Pages are copied, not re-rendered, so nothing is recompressed and nothing loses sharpness.
• Web links. A link to a website is an annotation sitting on the page with the address written into it. It needs nothing outside the page, so it keeps working. This includes email links and anything else that is a plain address.
• Page size, rotation and boxes. Each page carries its own, so a bundle mixing A4 and Letter stays mixed, and a sideways page stays sideways. Why a rotated PDF goes back to how it was covers rotation.
• Comments and highlights, which are also page annotations — though a reply thread that references the document can end up incomplete.

And what does not: the outline, named destinations, page labels, document-level metadata such as title and author, attachments, article threads, and the document’s open action — the instruction that tells a reader to open at page 5 or at a particular zoom.

Metadata is worth a separate note, because it cuts the useful way: a new document is not born carrying the old one’s author name, software and revision history. What PDF metadata contains covers what that normally includes, and the metadata editor lets you check what came out.

The awkward case: links to another page in the same file

A contents page whose entries jump to chapters, or a "see page 12" cross-reference, is an annotation on the page like a web link — but instead of an address it holds a destination: a reference to another page object in the same document.

When the pages are copied into a new document, the annotation comes with its page. Its destination has to be copied too, and that destination is a page. So the copying machinery follows the reference and brings that page across as well — as a separate object that is not in the new document’s page tree.

We measured this on these tools’ own output rather than assuming it. Merging a six-page document that contains one internal link produces a file with seven page objects and six pages. The file is structurally valid — a checker finds nothing wrong with it — and the link still highlights when you hover it. It simply points at a page that is not in the document, so a reader has nowhere to take you.

So: treat internal links as not surviving. The clickable area may still be there, which is worse than a clean break because it looks like it works. This holds even when the target page is one you kept, and even for a straight merge where every page was copied.

Links that use a named destination fail more honestly: the name table is on the catalogue, so it is gone, and the link has nothing to resolve.

What to do about it

There is no tool on this site that writes an outline, and this page will say so rather than imply otherwise. What there is:

• Do the page work first, the navigation last. Merge, split and reorder while the document is still raw, and add bookmarks in whatever application will produce the final file. Doing it the other way round throws the work away.
• If the bundle needs navigation, build it from the source documents. A word processor or a desktop PDF editor that can write an outline is the right tool; these browser tools cannot, and a tool that claims to preserve bookmarks through a merge is doing considerably more than copying pages.
• Add printed page numbers after merging, not before. Add page numbers gives the bundle one honest sequence, and it has a separate setting for which sheet to start on so a cover page can stay unnumbered.
• Give the reader a contents page as text. Not a substitute for bookmarks, but in a merged bundle a first page listing what is where survives everything and needs no feature support at all.
• Check before you send. Open the result and look at the bookmark panel and one or two cross-references. It takes ten seconds and the alternative is finding out from the recipient.

How to combine PDF files and reordering and deleting pages cover the rest of what moves and what does not.