Pull out the pages you want as a new file.
Add a PDF, then say which pages you want — single numbers, ranges, or a mix, like 1-3, 7, 12-15. The pages are copied into a new document in the order you listed them, and the original is left untouched.
Add the PDF and the panel reports its page count, which is the number your ranges have to stay inside. Nothing is uploaded.
Commas separate entries, hyphens make ranges. So 1-3, 7, 12-15 keeps eleven pages. Anything you leave out is simply not copied.
The output follows the order you write, not the order in the original. Entering 5, 1, 3 produces a three-page document with page five first. This makes the tool a reordering tool as much as an extraction tool, which is useful when a scanner has produced pages back to front — write 10-1 and the whole thing comes out reversed.
The new document contains copies of the chosen pages with text, links and fonts intact. The original is untouched on disk. Open the result before you send it, because an off-by-one in a range is easy to make and invisible until someone else finds it.
A PDF page is an object that points at the resources it uses. Extracting copies the page object and everything it references into a new document. The page is not re-rendered or re-encoded, so text stays selectable, vector art stays vector, and image quality is untouched. The operation is fast because it is fundamentally a graph copy.
A page references shared resources. If every page of a report uses the same embedded font and the same header logo, those objects are stored once for the whole document. Extract a single page and it must bring its dependencies with it, so a one-page extract from a two-megabyte report is not one-fiftieth of two megabytes. It is that page plus the fonts and images it needs, which can be most of the overhead.
This is the point people get wrong, and it matters. Content that is visually off the page — outside the crop box — travels with the page object. Metadata attached to the page travels too. Annotations and comment threads generally travel. If a page was created by drawing an opaque rectangle over sensitive text, the text is still there in the file and can be recovered by selecting it. Extraction is about which pages you send, not about what is inside them.
A page object is a hub with references radiating out from it, and extraction follows those references. Annotations attached to the page — comments, highlights, sticky notes, stamps — are page-level objects and come across. So do link annotations, though an internal link whose target page was not extracted becomes a link to nowhere. Form field widgets are page-level but their field definitions are document-level, so an extracted form page carries the visible box without the machinery behind it. Page-level metadata travels. Content positioned outside the crop box travels, because the crop box only controls what is displayed. If a page was assembled by placing a full-page image and then cropping the view to part of it, the whole image is in the extracted file.
Ranges here refer to physical page positions, counting from one, not to whatever numbers are printed on the pages. A report with roman-numeral front matter will have a printed page 1 that is physically page 7. Trust the page count the tool reports and the position in the file.
Something in the range falls outside the document or is not a number the parser recognises. Check for a typo, a page number beyond the end, or a stray character. En-dashes pasted from a word processor are accepted, but other punctuation is not.
Ranges are inclusive at both ends, so 1-3 is three pages. A duplicate entry across two ranges is extracted twice rather than skipped, which can also throw the count. Read the ranges back as a list and count them.
Shared fonts and images. See the section above — a page brings its dependencies. Run the result through PDF Compress to drop anything genuinely unused.
Write the ranges around them: 1-3, 5-8, 10-20 for a twenty-page document. There is no exclusion syntax, and being explicit is less error-prone anyway.
Internal links point at page objects. If the target page was not extracted, the link has nowhere to go. External links to websites are unaffected.
A desktop editor shows you page thumbnails and lets you select visually, which is genuinely better when you do not know exactly which pages you want. Typing ranges is faster once you do. For a long document you are unfamiliar with, open it in a reader first, note the page numbers, then come back.
Every operating system lets you print to PDF and specify a page range, which achieves something similar and needs no tools at all. The difference is that printing re-renders the page through the print pipeline. Text usually survives, but form fields flatten, some links break, and the output can be larger. Direct extraction copies the original objects and is cleaner.
Same trade as always. They work, they are free, and your document spends time on someone else’s disk. For a page from a legal filing or a medical letter, that is worth avoiding for the sake of a tool that does the same job locally.
Commas separate entries and hyphens make ranges, so 1-3, 7, 12-15 keeps eleven pages. Order matters: 5, 1 puts page five first.
Not in one pass. Run the tool once per range you want, changing the page numbers each time.
Yes. Pages are copied whole, so text, links, and embedded fonts all survive. Nothing is rasterised.