Extracting pulls the pages you name out of a larger PDF and leaves the rest behind. Give ranges the way you would say them — 1-3, 7, 10-12 — and each comma-separated group becomes its own file, or ask for every page separately if you want them individually. The pages are copied rather than re-rendered, so the extracted file is visually identical to those pages of the original, with selectable text and unchanged image quality. Everything happens in your browser, so a confidential document is never uploaded and there is no size limit on what you can work with. This is the tool for sending someone exactly the three pages they asked for: the signature page of a contract, one invoice out of a monthly batch, the section of a report that concerns them and nothing else.
Extracting against splitting
Both take ranges and both produce new PDFs; the difference is intent and therefore coverage. Extraction takes a subset and discards the remainder — you want three pages out of ninety. Splitting divides a document into parts you intend to keep, so the ranges usually account for everything.
The practical consequence is what you do with the output. Extracted pages are usually about to be sent to someone; split parts are usually about to be filed. If your ranges add up to the whole document, splitting is the tool you actually want.
Sending exactly what was asked for
Sharing a whole document because someone needs one page of it is a small, routine information leak — salary tables in an appendix, other clients' names in a schedule, internal comments in the notes. Extraction is the least effortful fix available.
Do check what remains on the pages you extract. A page carries whatever is on it, including headers, footers, watermarks and any marginal annotation, and those travel with the extracted copy. If something on a kept page is sensitive, redaction is the next step.
What comes with the pages
Content, images, selectable text, per-page annotations and links whose targets are inside the extracted range all come across intact. Fonts used by those pages are embedded into the new document, so it renders correctly on a machine that does not have them installed.
Links pointing at pages you did not extract cannot resolve and are dropped, along with bookmarks whose targets are gone. Interactive form fields belonging to pages outside the range go too. Everything that remains is self-contained, which is the point: the output has to work on its own.