Deleting pages produces a copy of your PDF without the pages you deselected — cover sheets, blank versos from a duplex scan, an internal appendix, the boilerplate terms nobody needs. Work from the thumbnails, untick what should go, and download what remains. The pages that stay are copied at the object level rather than re-rendered, so text stays selectable, images keep their original quality and the visual result is indistinguishable from the source. Nothing is uploaded: the file is read into memory in your browser and the new document is assembled there, which means no size limit and no copy of your document sitting on anyone's server. The important thing to understand is that this removes the pages entirely rather than hiding them — the content is genuinely absent from the file you download.
The pages are gone, not hidden
A deleted page is not carried along invisibly. The output document is built from the pages you kept, so the removed pages have no representation in it at all — no content stream, no thumbnail, nothing recoverable by a determined recipient.
That distinguishes deletion from the sort of tidying that only changes what a viewer shows. It also makes it a reasonable step before sending a document externally, though for sensitive material inside a page you are keeping, redaction rather than deletion is the tool that applies.
Why the file might not shrink much
Removing half the pages rarely halves the file. PDFs share resources: an embedded font is stored once and used by every page that needs it, so dropping pages does not drop the font. Neither does it remove images still referenced by pages you kept.
Where deletion does save meaningfully is on scans, because each page carries its own image and that image is most of its weight. Deleting ten scanned pages from a forty-page scan takes roughly a quarter off. For a text document, expect the saving to be modest and compress separately if size is the goal.
Check the numbering afterwards
Deleting shifts every subsequent page up, which is fine for the pages themselves and awkward for anything that referred to them. A printed table of contents, cross-references in the text and page numbers drawn onto the pages all keep pointing at their old positions.
Nothing here can fix that, because those are page content rather than document metadata. If the document relies on internal references, either delete before the numbering is applied, or renumber afterwards with the page-numbers tool.