The string that did the work was 1-3,14-62,150-168. Everything else was naming. I keep saying that because people arrive wanting a philosophy of “splitting PDFs” and leave with a file called part-1.pdf that starts mid-sentence.
Three ranges from one Tuesday report
The source was a 180-page environmental report, 96 MB, mostly scans of maps plus a text front section. The client’s portal accepted 25 MB. Email to the reviewer was capped at 20 MB after the gateway’s own encoding. Nobody needed all 180 pages in one attachment.
1-3 cover, transmittal, contents 14-62 findings the reviewer actually had to mark up 150-168 the two map plates named in the email
Viewer page 1 was the cover. The printed footer on that cover said nothing. By the time the body started, the footer said “page 1” while the viewer said 14. If I had typed the footer numbers into Split PDF, the extract would have begun in the table of figures. I have done that. The reviewer sent it back with a single line: “this starts at a caption.”
Empty range field behavior on this tool is “you probably meant pages.” Do not rely on that. Type the ranges. Commas separate items. A hyphen is inclusive. 8 is one page. 8-8 is the same page. 8-2 is nonsense and should be treated as a typo, not a creative request.
What the extractor copies
The page script uses pdf-lib. It parses your list into zero-based indexes, copies those pages into a new document, and downloads that document. It does not re-render. A text page stays a text page. A 12 MB map plate stays a 12 MB map plate, which is why “I split it” is not the same sentence as “it will fit in email.”
That map plate was the whole lesson. 150-168 came out at 41 MB. Splitting removed the other chapters and did nothing to the pixels inside the plates. I then made a compressed sibling of that one extract, and I left the split-but-uncompressed plate in the project folder labeled maps-master.pdf. The reviewer received maps-email.pdf. Same geometry, softer gradients, under the cap.
| Ask | Range I typed | Result size | Next move |
|---|---|---|---|
| Transmittal only | 1-3 | 0.8 MB | Send as-is |
| Markup chapter | 14-62 | 11 MB | Send as-is |
| Named plates | 150-168 | 41 MB | Compress a copy, keep this file |
| “Every 20 pages” with no outline | nine files, two of them starting mid-paragraph | varied | Do not do this if the PDF has headings |
Cuts that create legal nonsense
A contract split between the last operative clause and the signature block is a favor to nobody. Include the signature pages with the section they close. Same for an exhibit cover: if the cover says “Exhibit C” and the pages after it are the exhibit, they travel together. I write the ranges in English on a sticky note first — “cover through contents, findings, the two plates” — and only then translate to numbers. The English version catches the stupid cut.
Password-protected files do not open for extraction. Unlock a copy you are allowed to open, split that copy, and do not email the unlocked master to the same list you were trying to keep small. Splitting is a distribution tool. It is not a confidentiality tool.
The original 96 MB file stayed in /archive. I did not delete it after the three extracts looked right. Extracts are working copies. Six weeks later someone asked for appendix D, which I had not extracted. The archive file was the only place appendix D still had its original bookmarks and the only place the page numbers matched the printed volume. That request took four minutes because the master was still there. It would have taken a day if I had “cleaned up.”
Memory, and the tab that dies at page 90
One file in memory is easier than five files in a merge, but a single 96 MB scan can still hurt a phone. If the render or the copy stalls, the file is too heavy for that device, not “corrupt.” Move the job to a desktop browser. pdf-lib’s copy is fast once the bytes are loaded. The slow part is getting a huge ArrayBuffer to stay alive.
If you only need one page for a slide, split that page out before you convert to JPG. Rendering all 180 pages at scale 2 is how laptops sound like hair dryers. One page is a two-second job.
Using a split as the input to a question
I stopped pasting 180-page PDFs into Chat with PDF. The model does not get smarter because you gave it more paper. I split findings into their own file, ask about sampling method and the one table of exceedances, and I open the PDF beside the answer. When the answer cites a page, I check that page. When it shrugs, the sentence usually is not in the extract, which is a better failure than a confident paragraph about a chapter I did not include.
After the comments came back I did not merge the marked-up chapter into the archive master. The master stays clean. The marked-up extract is a review artifact with its own name and date. Recombining too early is how you lose track of which stamps are current.
Names that survive a shared drive
report-findings-p14-62.pdf beats split-output.pdf. Put the viewer range in the filename. Future you will not remember that “final-v3” meant the maps. A one-line text file next to the extracts, with the ranges and the date, is enough of a lab notebook. I have started jobs from that note three quarters in a row.
Blank pages at a section break might be intentional printer spreads. Look at them before you delete them in Organize PDF. I removed a “blank” once that was the back of a double-sided map legend. The reviewer printed duplex and the legend vanished. That was a me problem, not a tool problem.
If a portal wants under 10 MB and your smallest coherent chapter is 11, you are in compression territory, not in “split harder” territory. Finer splits that break a table across files make the upload smaller and the document worse. Stop at the semantic edge. Then decide whether a JPEG rebuild of that one piece is acceptable for that one recipient.
When the portal and the reader disagree
A university upload form wanted one PDF under 10 MB. The examiner wanted the bibliography attached. Those are different readers. I uploaded findings plus bibliography as two files because the form allowed a second slot, instead of crushing both into one JPEG brick. Where the form allows only one slot, I put a cover line on page 1: “maps supplied separately; this file is the text.” Pretending one upload can be both the archive and the form-filler is how the maps get ruined and the text gets rejected for size at the same time.
Page labels in some PDFs (the ones that show “iv” or “A-3” in a reader) are not what the range parser sees. It sees the page index from getPageIndices, which is the count from the start of the file. If a cover is unnumbered, it is still page 1. I check the index in the thumbnail strip, not the footer, every time the document looks “professionally numbered.” Those documents are the ones that lie.