Merging and splitting PDF files

Combining two PDFs sounds like joining two piles of paper, and for the pages themselves that is roughly true. Everything around the pages is where the interesting problems live. A PDF page does not carry its own fonts, its own colour profiles or its own form definitions — it references them from shared tables held by the document. Merge two files and those tables have to be reconciled; split one and they have to be divided. Whether a tool does that carefully or crudely is the difference between a merged file that is the sum of its parts and one that is twice the size, has lost its bookmarks and shows a blank signature block where a filled form used to be. Note before you read on: our own tool compresses, converts and handles passwords, and does not merge or split. This page exists because those operations sit next to the ones we do, and because knowing what they break tells you what to check afterwards.

What merging really has to reconcile

Four things travel with a document rather than with a page. Fonts are the first: each source file embeds its own subsets, and a naive merge keeps both copies even when they are the same typeface, which is the usual reason a merged file is larger than the two originals added together. A careful merge de-duplicates identical font programs. The second is the outline tree, the bookmarks panel — two documents have two roots, and something has to decide whether they become sibling branches or whether one set is simply dropped, which is what most quick tools do. Third are named destinations, the internal anchors that make cross-references clickable: two files can easily both contain an anchor called "section1", and one of them will lose. Fourth is metadata, which cannot be merged in any meaningful sense, so the result normally inherits the first file's title and author and carries them silently into a document that is now something else.

Forms, signatures and annotations

Interactive elements are where merges do real damage. An AcroForm field is a document-level object; two files each with a field named "date" cannot both keep it, and the usual outcome is that one field's value follows the other's, so filling one changes both. Some tools flatten forms on merge, turning fields into printed-looking text, which fixes the collision by removing the interactivity — acceptable if you are archiving, disastrous if the recipient still has to fill it in. Digital signatures are simpler and harsher: a signature covers the byte range of the document it signed, so any merge invalidates it. That is the signature doing its job. If you need a signed document inside a compilation, keep the signed original alongside as a separate file, because no tool can merge it and keep the signature valid.

What splitting does and does not carry

Splitting is subtraction and is usually gentler, but it has its own trap. If a tool extracts pages 5 to 10 into a new file, those pages still reference the whole document's shared resources, and a lazy implementation copies the entire resource table across — leaving you with a six-page file that carries the fonts and colour profiles of the original two hundred. The result opens correctly and is far larger than it should be. Bookmarks pointing outside the extracted range have to be pruned or they become dead links, and internal cross-references to removed pages will silently go nowhere. When you split, check the output size against the page ratio: six pages out of two hundred should not be half the original file, and if it is, run the result through a compressor or use a different tool.

Page order, rotation and the boring failures

Most merge complaints are not corruption but arrangement. Pages that were scanned upside down carry a rotation attribute rather than rotated pixels, and some tools apply that attribute while others discard it, so a merged file can suddenly contain sideways pages that looked fine before. Mixed page sizes are the other one: combining A4 and US Letter, or portrait and landscape, produces a document that is correct but prints badly, because the printer scales to the first page or asks per page. Neither is a bug to be fixed after the fact — decide before merging whether you want a uniform page size, and if you do, normalise the odd pages first rather than letting the printer improvise.

Choosing a method

For a one-off, the operating system may already have it. Preview on macOS merges by dragging page thumbnails between open documents. On Windows, printing to the Microsoft Print to PDF driver merges reliably but re-renders everything, which flattens forms and can soften text — fine for an archive, wrong for a document that must stay crisp. Command-line tools give the most predictable results for repeat work, and are worth learning if you assemble documents regularly. Web services are convenient and are also the option where your document leaves your machine, which is a real consideration for contracts, medical records and anything with personal data in it. There is no universally right answer; there is a right answer per document, and it depends mostly on how sensitive the content is and whether interactivity has to survive.

Compress afterwards, not before

If the merged document is destined for email, compress the finished file rather than each input. Compressing the parts first and then merging leaves duplicated resources in place and re-encodes some images twice, which costs quality without saving much. One pass over the assembled document handles the whole thing once, de-duplicates what the merge left behind, and gives you a single number to check against your mail server's limit. The same logic applies to splitting: extract the pages you need and compress the extract, rather than compressing a two-hundred-page file to get at six pages of it.

A short check after every merge or split

Open the result and confirm five things. The page count is what you expected. The first and last pages of each source section are present and the right way up. Any form fields still accept input, or have been deliberately flattened. Bookmarks either work or are gone — not present but broken. And the file size is proportionate: notably larger than the inputs combined means duplicated fonts, and a split file nearly as large as its parent means the resource table came along for the ride. Two minutes of checking catches everything described on this page, and none of it is visible from a thumbnail.