Split PDF the right way: ranges, extraction, and file size

A PDF is not a linear stream of pages the way a Word document is a stream of text. Internally it is a collection of numbered objects: page dictionaries, content streams, font programs, images, and a cross-reference table that records the byte offset of every object so a viewer can jump straight to page 200 without reading pages 1 through 199. The pages themselves hang off a page tree, and each page dictionary points at the resources it needs, such as fonts and images.

When a splitter extracts pages 4 to 11, it walks the page tree, copies those eight page dictionaries into a new document, and then follows every reference each page makes: its content stream, the fonts those streams use, any images placed on the page, and annotation objects like hyperlinks. Everything reachable from the selected pages gets copied; everything else, including the other 52 pages and their resources, is left behind. Finally the tool writes a fresh cross-reference table and trailer for the new, smaller document.

Because the operation copies objects rather than re-rendering pages, nothing is recompressed or rasterized. Text remains real text, so it stays selectable and searchable, vector graphics stay sharp at any zoom level, and images keep their original compression. A split is lossless by construction, which is one reason it is safe to split legal and archival documents without worrying about degrading them.

People are often puzzled when they burst a 10 MB PDF into ten single-page files and the pieces total 30 MB. The cause is shared resources. In the original document, one embedded font program or one background image can be referenced by every page while existing in the file exactly once. Split that document into ten files and each file needs its own copy of that font or image, because a PDF cannot reference objects inside a different file.

The effect is strongest in documents built from a template: letterheads, slide exports, and forms where a logo, a font set, and a background graphic repeat on every page. A 200 KB font subset shared across 50 pages costs 200 KB once in the original but 200 KB in every split output. Scanned documents behave the opposite way, since each scanned page is its own full-page image with nothing shared, so split sizes track page counts almost exactly.

The practical takeaway: if your goal is to get under a size limit, split by range rather than bursting every page, and check the actual output size rather than estimating from the original. If a range is still too heavy because of scanned images, compression is the right next step, not a finer split.

Per-page bursting, custom ranges, and fixed-interval splitting solve different problems, and picking the wrong mode creates cleanup work. Bursting into single pages fits situations where each page is genuinely independent: a stack of scanned receipts digitized in one pass, individual certificates generated as one batch PDF, or invoices that need to be filed into separate client folders.

Custom ranges are the mode for structured documents. A range expression like 1-5, 8, 12-20 pulls the executive summary, one exhibit, and an appendix into a single new file while skipping everything between. This is the mode for contract work, where you might need pages 1 to 3 plus the signature page as one file for the counterparty and nothing else.

Every-N splitting shines when a scanner has concatenated many identical-length documents into one file. If a clerk scanned forty 3-page forms as one 120-page PDF, splitting every 3 pages reconstructs the forty original documents in one operation instead of forty manual range extractions. It only works when the interval is truly constant, so verify the page count divides evenly before trusting the output.

The most common trigger is an attachment limit. Gmail caps attachments at 25 MB and many corporate Exchange servers sit at 10 MB, so a full scanned mortgage package simply will not send. Splitting the package at its natural section boundaries, such as application, disclosures, and appraisal, gets each piece under the cap and, as a side benefit, lets the recipient open the one section they care about.

Legal and administrative workflows lean on extraction. Filing portals for some courts reject uploads over a fixed size, immigration applications ask for specific pages of a passport or bank statement rather than the whole booklet, and opposing counsel should receive the executed signature page, not your entire internal draft history. Extracting exactly the requested pages is also a quiet confidentiality control, because pages you never send cannot leak.

Publishing and teaching produce the rest. Instructors split a textbook PDF into per-week readings so students are not scrolling a 400-page file on a phone. Researchers pull a single article out of scanned conference proceedings. Accountants split a year-end binder so the auditor gets the general ledger section while payroll stays internal. In each case the split file becomes the distributable unit and the master stays intact.

Content on the selected pages survives completely: text, images, vector drawings, and link annotations that point at external URLs. Things that reference the document as a whole degrade predictably. Bookmarks pointing at pages you removed have nowhere to go. Internal links from a table of contents on page 2 to a section on page 40 break if page 40 is not in the output. Form fields on extracted pages usually survive, but calculation scripts that read fields on other pages will misbehave.

The classic mistake when splitting is confusing printed page numbers with PDF page positions. Books front-load roman-numeral material, so the page printed as 1 is often the seventh or eighth page of the file. Scanned documents add cover sheets. Always trust the position shown in the tool's page thumbnails or your viewer's page counter, not the folio number printed on the page image, and spot-check the first and last page of the output before sending it anywhere.

One more boundary worth knowing: splitting is not redaction. Extracting pages 1 to 5 removes pages 6 onward from the output file, but metadata such as the document title, author, and producer carries over from the original. If the concern is hidden or sensitive content rather than page count, handle that separately before distributing the split file.