How to split a PDF

Three different jobs called splitting

People say "split" for three things, and picking the wrong one is most of the wasted time:

Extract: take pages 12 to 18 out and keep only those. One file out.

Divide: cut a 300-page document into chapters. Many files out, and the cut points are what you care about.

Burst: one file per page. Usually for scanned batches: forty invoices that arrived as one document and need to be forty documents again.

All three are the same operation underneath, copy a set of pages into a new document, but they want completely different controls, and a tool that only offers one of them makes the other two tedious.

Page numbers lie, and here is how

The single most common error is off by one, and it has two separate causes.

The cover page. A document whose printed page 1 is the second sheet in the file is entirely normal, title page, then contents, then page 1. The PDF knows nothing about the printed numbers. Ranges are always positions in the file, counting from the first sheet.

Zero versus one. Programmers count from zero and documents count from one. Any tool written carelessly will be one out somewhere. The test takes five seconds: ask for the range 1-1 and see whether you get the first sheet or the second.

Ours counts from one, the way the page thumbnails are labelled, and a range that runs past the end is clamped rather than refused, asking for 10-999 of a twelve-page document gives you pages 10 to 12.

Splitting should not re-render

The same warning as merging, for the same reason. A split done properly copies the page's drawing instructions into a new document: text stays text, vectors stay vectors, the file stays searchable and stays roughly proportional in size.

A split done by rendering each page to an image and wrapping it in a new PDF produces something that looks right in a viewer and is much worse, nothing selectable, nothing searchable, blurry at any zoom you did not anticipate, and frequently larger than the original.

Check the same way: open a result and try to select text.

Why the output is bigger than a fraction

Take 10 pages out of a 100-page, 10 MB document and the result is rarely 1 MB. Usually it is more, sometimes much more, and nothing has gone wrong.

PDFs share resources across pages. One embedded font is stored once and used by every page; so is a logo that appears on all of them. Copy any single page out and it needs its own copy of every font and image it references. Ten pages carrying a full font set can easily come to a third of the original file.

The practical consequence: bursting a document into one file per page produces a folder considerably larger than what you started with. That is expected.

Getting the ranges right the first time

• Commas separate, hyphens span. 1-3, 7, 10-12 is six pages in three

groups.

• Order is yours. 5, 1-2 is legitimate if you want page 5 first.
• One page is a range. 7 and 7-7 are the same thing.
• Check against the thumbnails, not the printed numbers. This is the cover

page problem, and looking is faster than counting.

For a long document being cut at chapters, typing thirty ranges by hand is not the right approach. Cutting every N pages, or splitting at marked points, is the same job without the transcription errors.

What does not survive

• Bookmarks and links that point at pages which are not in the output have

nowhere to go.

• Form fields come with their page, but a field referenced by a script

elsewhere in the document will lose the other half of that relationship.

• Encryption has to be removed before the pages can be copied, and no

split tool can do that without the password.

Doing it privately

A split is a local operation on a local file. Uploading a scanned passport or a bank statement to a stranger's server in order to keep pages 2 and 3 of it is a poor trade, and an unnecessary one: the work is small enough that your browser can do it, which is what this one does.