Twenty-four pages in, five pages out: that's extraction. Somewhere in that long report sits the two-page summary and three exhibits somebody actually needs, and the answer is not "send the whole thing" — it's a new PDF that contains exactly the right pages. Here's the syntax real extractors speak, how extraction differs from its cousins, what quietly breaks when pages move, and how to do the whole job without uploading the file.
Extraction copies a chosen subset of pages into a brand-new PDF. The tool reads the original, takes the page objects you named, assembles them into a fresh document, and saves it. Nothing about the kept pages is re-rendered — text stays selectable and searchable, images keep their resolution, form fields and annotations ride along. The original file never changes; you get a second, smaller document.
That copy-not-rerender detail is what separates a good extractor from the screenshot-and-rebuild school of thought. A rebuild rasterizes your pages into images; the extract is still a real text PDF that you can search, copy from, and pass through downstream tools.
Every serious extractor speaks roughly the same dialect, and it's worth thirty seconds to learn fluently:
| You type | You keep |
|---|---|
4 | Page 4 |
2,5,9 | Pages 2, 5, and 9 |
5-7 | Pages 5, 6, 7 — ranges include both ends |
2,5-7,20 | Pages 2, 5, 6, 7, 20 — mix freely |
4-end | Page 4 through the final page, whatever its number |
Two habits save errors. Read the resolved list before you download — a good tool shows "2, 5, 6, 7, 20" as chips or text so you're approving the exact pages, not your syntax. And know your error cases: pages before 1, pages past the end of the file, and backwards ranges like 7-5 should be rejected with a message, not silently produce a wrong file.
The outputs of extract and delete can be identical; the difference is which list is shorter to write. Pick the shorter list.
Features that point across pages:
What doesn't break is the content: fonts, images, selectable text, and any form fields on the pages you kept arrive byte-faithful.
Yes, and for confidential documents that's the only sane route. Copying page objects is a local operation — no server needs to see the file. Our Extract PDF Pages tool parses your selection, shows the resolved list before you commit, and builds the output with pdf-lib in your browser. Type 2,5-7,20 on a 24-page file and the download is a 5-page PDF, original untouched; type 20,2,5-7 and the same pages arrive in your order, summary-behind-exhibits if that's the story you want. When the job is really "reorganize everything," that's Organize PDF territory instead.
Password-protected files stop any extractor before page one: the document is encrypted end to end, and nothing can be copied out of it without the password. Unlock it first (in whatever word processor or PDF editor owns it), then extract. Digitally signed documents are the subtler case — a signature cryptographically covers the exact bytes of the exact pages, so any rebuild, extraction included, breaks the signature's validity. That is the signature doing its job: it is telling the next reader that the document changed since it was signed. If you need an extract of a signed contract, keep the signed original for the record and treat the extract as a working convenience copy, clearly separate from the authoritative file.
Open a PDF, type 2,5-7, watch the resolved list, download the extract. Free, private, in-browser.
Extract PDF Pages →Extraction is subset selection with a download button: learn the three-token syntax, approve the resolved list, keep the original. Check bookmarks and prose cross-references in the output, choose extract or delete by which list is shorter, and keep the whole operation client-side whenever the document isn't public.