Until 24 September 2026, a page you deleted with our PDF tools could still be inside the file you downloaded. We found it while testing for this guide. We deleted page 10 of a 32-page NIST report, and the saved file still held page 10, text and all, where no viewer would show it. A five-page extract of the same report carried 18 hidden pages. The cause, and the fix, say a lot about what survives when you merge, split or rearrange a PDF, so we will start there.
How a deleted page stayed in the file
Pages in a PDF are linked to from other places: a table of contents entry, a bookmark, a form field, the accessibility tags that tell a screen reader what is on each page. Take a page out of the page list and those pointers still lead to it. When the file is saved, anything something still points to is kept. Three links in the report led to page 10, so page 10 stayed.
Split and merge had the same fault in a different shape. The library our tools use copies a page together with everything it points to, and a contents page points to every section, so copying pages 1 to 5 copied 18 other pages as well, invisibly. The five-page file weighed 1.42 MB. The internal links on those pages led nowhere, because they now pointed at the hidden copies. Merging the Transformer paper with the report left all 142 of their internal links dead in the same way: citations, figure references and contents entries.
Our tools now cut every reference to a page that is no longer in the file before saving, repoint internal links at the pages in their new places, and drop links to pages that were left out. After the fix, the five-page extract is 509 KB with nothing hidden in it, the file with page 10 deleted is 104 KB lighter, and all 142 links in the merged file land on the right page. If you split or trimmed a sensitive PDF with our tools before 24 September 2026, run it again.
What survives each operation
With that fixed, we ran each page tool over three files: the NIST Cybersecurity Framework report (made in Word, with 12 bookmarks, 47 internal and 56 web links, and accessibility tags), the “Attention Is All You Need” paper (22 bookmarks, 95 internal links), and the IRS W-9 form (27 form fields, Adobe XFA form data and a usage-rights signature).
| Tool | How it works | Bookmarks | Internal links | Form fields | Accessibility tags |
|---|---|---|---|---|---|
| Rotate PDF | Edits the file in place | Kept | Kept | Kept, with XFA and signature | Kept |
| Delete PDF pages | Edits the file in place | Kept; ones to deleted pages stop working | Kept; links to deleted pages removed | Kept | Kept |
| Add page numbers | Edits the file in place | Kept | Kept | Kept, with XFA and signature | Kept |
| Organize PDF, same order | Edits the file in place | Kept | Kept | Kept | Kept |
| Organize PDF, new order | Builds a new file | Lost | Kept and repointed | Form list lost | Lost |
| Split PDF | Builds new files | Lost | Kept within each file; others removed | Form list lost | Lost |
| Merge PDF | Builds a new file | Lost (34 of 34) | Kept and repointed (142 of 142) | Form list lost | Lost |
The split between the two groups is the rule to remember. Tools that edit the file in place keep everything they don’t touch. Tools that build a new file carry pages across, and whatever belongs to the document rather than to a page stays behind: the bookmark tree, the accessibility tags, the list of form fields and the XFA form data.
The W-9 shows what “form list lost” means in practice. Merged with the report, its 23 boxes on page 1 were still drawn and pdf.js still offered them for typing, but the document no longer listed any form fields, where the original listed 27. A reader or a form tool that works from that list finds no form. Reordering the W-9’s pages also dropped the XFA data, and the file went from 141 KB to 56 KB.
If you need a form to stay a form, only rotate, delete pages or add numbers to it.
Flatten a filled form before you merge it
Filling the form in first is not enough on its own. We filled the W-9 with our Fill PDF form tool (a name, a business name and one ticked box), merged it with the 48-page NIST AI Risk Management Framework, and did the same again with a copy flattened first.
| W-9 going into the merge | Form fields listed after merging | Answer boxes left on page 1 | Name in the page text | Merged file |
|---|---|---|---|---|
| Filled, fields kept (84 KB) | 0 of 23 | 23 | No | 1.96 MB |
| Filled and flattened (77 KB) | None needed | 0 | Yes | 1.96 MB |
The unflattened answers didn’t vanish from view: Chromium’s built-in viewer still showed the name and the tick, a little greyer than the printed text. But they lived only in the 23 leftover boxes, which no longer belonged to any form. The name wasn’t part of the page’s text, so PDF to text and search missed it, and a form tool found nothing to read or change. Flattened first, the answers became ordinary page text and came through the merge like everything else on the page. Flatten PDF does that for forms that are already filled; Fill PDF form can flatten as it saves.
Splitting can make more bytes, not fewer
Split the 32-page, 1.52 MB report into single pages and you get 32 files that add up to 4.90 MB, more than three times the original. Each one-page file carries its own copy of the fonts it uses, so even the plainest page came out at 79 KB. Page ranges cost less than singles, and which pages you take matters more than how many: pages 1 to 5, with the cover photo, came to 509 KB; pages 6 to 10, with a large diagram, to 598 KB; pages 21 to 25, all text, to 137 KB.
Every file these tools save also leaves without its author, app and date details; what those look like in real PDFs is in what your PDF says about you. To shrink a file rather than cut it, see how much a PDF really shrinks, or go straight to Merge PDF and Split PDF.