Merging, splitting and reordering PDFs: workflows for contracts, scans and submissions

A signed signature page has to go back into a 40-page agreement. Twelve photographed receipts need to become one expense report. A double-sided document went through a single-sided scanner and now reads 1, 3, 5, then 6, 4, 2. A portal wants a single numbered PDF under a size limit, cover first. Each of these is a page-level edit — combine, extract, rearrange, stamp — and each has a way of going quietly wrong: bookmarks vanish, form fields freeze, a “removed” page leaves traces, a merged file balloons to three times the size of its parts. This guide covers what the site’s tools actually do to the bytes and the checks that catch a broken result before someone else does.

Which operation do you need?

Four tools cover the page-level jobs and share one engine: pages are copied from source documents into a new one by pdf-lib, and previews are drawn by pdf.js. Nothing is re-rendered, so text stays selectable and images keep their pixels.

None of the four rotates a page, and none opens an encrypted file. When you need more than one, work in this order: merge, then reorder or delete, then compress if you must, then number last, so the numbers reflect the final page set.

Merging: order, page sizes, and what does not survive

The merge tool accepts any number of files ending in .pdf(anything else is silently skipped by the picker filter), lists them numbered, and joins them top to bottom; the button stays disabled until two are listed. Each input’s pages are appended whole, in their own order, into a fresh document.

Mixed page sizes and orientations

Every copied page keeps its own MediaBox, CropBox and Rotateentry, so an A4 page (595 × 842 pt) followed by a US Letter page (612 × 792 pt) followed by a landscape scan simply stays that way; viewers cope and printers apply “fit to page”. The case that looks wrong is a photo added as a page: the image to PDF tool sizes each page to the image at one point per pixel, so a 4032 × 3024 phone photo becomes a 56-inch-wide page that dwarfs the A4 pages around it. Crop and downscale first — the phone-scanning guide covers how.

Bookmarks, forms and links

Copying works page by page, and some things are not on a page. The outline (bookmarks panel), the AcroForm dictionary that makes fields fillable, named destinations, page labels and the document title all live at document level, and none is carried across. Concretely: the merged file opens with an empty bookmark panel; a fillable form arrives with its widgets drawn as they last appeared but no longer editable — and a field with no stored appearance shows no value at all; a contents link that pointed to a named destination is dead. When form content matters, flatten first: open the filled form and print it to PDF (macOS Preview, Chrome and Edge all do this), which bakes the values into the page.

Encrypted inputs

pdf-lib cannot decrypt. The merge tool loads with encryption ignored, so a password-protected input either trips Could not merge these files. One of them may be corrupted or password-protected. or, worse, comes through as pages whose content streams are still ciphertext while the new file has no /Encrypt dictionary — they render blank. Run it through Unlock PDF first (it needs the current password and saves name_unlocked.pdf), merge, then re-protect. The passwords guide explains why an owner-password file still needs this step.

Why the merged file is bigger than its parts

Merging never recompresses an image, so the output is roughly the sum of the inputs — plus one hidden cost. Fonts are copied per source document and nothing deduplicates across sources. Twenty one-page invoices from the same accounting system, each embedding the same 300 KB font subset, merge into a file carrying twenty copies: 6 MB of fonts for twenty pages of text. The fix is not compression but re-exporting from the generating system as one multi-page document, where the font is embedded once.

Splitting: ranges, single pages, and ZIP naming

Drop one file on the split tool and it reports name.pdf — 24 page(s) before you do anything — the quickest page-count check on the site. The range parser accepts comma-separated numbers and dashed spans with or without spaces (1-3, 5, 8-10); a reversed span like 10-8 is fine; numbers beyond the last page are dropped without comment; duplicates collapse. Two inputs stop it: anything the pattern does not match, such as Unrecognized page range: "3–5" (an en dash pasted from a document, not a hyphen), and a range with nothing left in it, No valid pages in the requested range. The output is name_extracted.pdf, and the pages come out in ascending order whatever you typed — 5, 1 gives page 1 then page 5. Splitting cannot reorder.

Split every page (ZIP) gives one file per page: name_split.zip holding name_page_1.pdf, name_page_2.pdf and so on. Drop several PDFs at once and the range box disappears, every page of every file is split, and the archive is split_pages.zip. Two consequences. The numbers are not zero-padded, so a file manager sorts page_10 before page_2 — rename to page_01, page_02 before feeding them back into a merge in bulk. And each single-page file carries its own copy of every font the page uses, so a 60-page text document with a 400 KB embedded font becomes 60 files of at least 400 KB each; the ZIP dwarfs the original, and that is expected. There is no every-N-pages option: to halve a 24-page file, extract 1-12 and then 13-24.

Reordering and deleting pages

The reorder tool takes one file, renders every page as a 120-pixel-wide JPEG thumbnail, and lays them out left to right. Each card shows #3 (was 7) — new position and source page — with ← and → buttons that move it one place and ✕ to drop it (the last page cannot be removed). Save reordered PDF copies the surviving pages in the new order into name_reordered.pdf. The “was” labels are your verification: read them left to right before saving. Thumbnails at 120 pixels hide detail — a page saying only “This page intentionally left blank” in 10-point type looks identical to a truly blank scan — so extract a doubtful page with the split tool and open it full size before deleting.

Fixing a duplex scan

A double-sided document through a single-sided feeder produces the fronts first — 1, 3, 5, 7, 9, 11 for six sheets — and then, after you turn the stack over as a block, the backs in reverse: 12, 10, 8, 6, 4, 2. Merged fronts-first that is 1, 3, 5, 7, 9, 11, 12, 10, 8, 6, 4, 2. Restoring it in the reorder tool takes 30 single-step moves for six sheets, and the count grows with the square of the sheet count, so it only suits a handful of sheets.

For anything longer, combine split and merge. Drop both scans on the split tool and press Split every page; split_pages.zip now holds fronts_page_1.pdf through backs_page_6.pdf. In the merge tool add them one at a time in interleaved order — fronts_page_1, backs_page_6, fronts_page_2, backs_page_5… — check the numbered list, and merge. Twelve picks instead of thirty moves, and it scales to a hundred pages. Check the scanner software first, though: most driver dialogs have a “reverse page order” or “duplex from single-sided” option that makes the fix unnecessary.

Page numbers, and the cover-page problem

Add Page Numbers stamps every page of every file you drop, in 11-point Helvetica, dark grey, baseline 24 pt (a third of an inch) above the bottom edge — bottom centre, or bottom left or right at a 36 pt inset. Start at takes any whole number from 0; the “Page N of M” box turns 7 into Page 7 of 24. Helvetica is a standard-14 font every reader carries, so it is referenced rather than embedded and the file barely grows. One file downloads as name_numbered.pdf; several arrive as numbered.zip, and every file starts from the same Start at value — numbering does not continue from one file into the next.

Two traps. M is the start value plus the page count minus one, so a 10-page file started at 5 reads Page 5 of 14— right when the file continues a 4-page volume, surprising otherwise. And numbers are drawn in the page’s unrotated coordinate space: a page whose landscape orientation comes from a /Rotate 90 entry, typical of scanner output, gets its number along a side edge, reading sideways. Look at one landscape page before sending a batch.

There is no skip-cover option. The workaround is three short runs: extract page 1 as the cover and 2-24 as the body with the split tool, number the body starting at 1 (or 2 if the cover should count), and merge cover then body. Number only after you have finished deleting pages, or a removed page leaves a gap.

Bates numbering

Legal productions ask for Bates numbers: a fixed prefix plus a zero-padded counter, unique across the whole production and continuing from one document into the next — ABC000001 on the first page of the first exhibit, ABC000347 on the last page of the ninth. This tool does not do that: no prefix, no padding, no carry-over between files. It can give a continuous plain sequence across a set — merge the exhibits, number once starting where the previous production ended, split back into per-exhibit ranges — but true Bates stamps need dedicated legal software.

Before you submit: a checklist

Portals and clerks reject files for a short, predictable list of reasons. Run down it with the file you are actually about to upload, not the working copy.

• One file, right name. Merge everything the form asks for, then name it so it sorts and identifies itself: 2026-09-06_Surname_Contract_signed.pdf. ISO dates sort; spaces and accented characters are what upload scripts choke on.
• Right page count. Drop the final file on the split tool and read the count beside the name, or run qpdf --show-npages file.pdf (on a Mac, mdls -name kMDItemNumberOfPages file.pdf). Compare with what the form expects.
• No blank or duplicate pages. Open it in the reorder tool and scan the thumbnails; duplex scans of one-sided originals leave a white back after every page. Remove them and save.
• Consistent orientation. A sideways thumbnail means the scanner rotated that page. None of these tools rotates, so rescan it or rotate it in a viewer before merging.
• Searchable text when required. Drop the file on the OCR tool: if it answers This PDF already contained selectable text, so it was read directlyyou have a text layer; if it starts recognising pages, you are holding a pure scan. The site’s OCR returns .txt, not a searchable PDF, so a portal that demands one needs the scanner’s own OCR pass — see the OCR guide.
• Under the size limit.Compare the file’s size against the limit in the upload dialog before pressing anything. Gmail refuses attachments over 25 MB; most portals cap at 10 MB or lower. If you are over, the right fix depends on what is inside.

Keeping the file small

A merged file is large for one of two reasons, and they need opposite treatments. If the inputs were scans — one big image per page — the compress tool is the right instrument: it re-renders every page as a JPEG at a chosen density (Light 150 dpi, Balanced 110 dpi, Strong 72 dpi) and rebuilds the file, or, given a Target size (KB), walks down a quality-and-resolution ladder until it fits, reporting The target size could not be fully reached; the smallest possible version was saved. if even the last rung misses. An A4 page scanned at 300 dpi is 2480 × 3508 pixels; at 110 dpi it is 909 × 1286, roughly an eighth of the pixels, which is where the savings come from.

If the inputs were generated documents — contracts from Word, invoices from a billing system — compression makes them worse: each page becomes a picture, the text stops being selectable, and 30 KB of text and fonts becomes 200 KB of JPEG. The size problem in a text-born merge is the duplicated-fonts effect above, and the remedy is upstream. The PDF size guide shows how to tell which kind of file you have. Either way, compress before numbering, so the numbers are stamped as real text onto the final pages.

Common mistakes

• Trusting a deleted page to be gone.Removing a page in the reorder tool drops it from the page tree, and a page nothing references is not copied. But the copier follows every reference from the pages you keep, and an internal link — a contents entry, a “see page 12” cross-reference — pointing at the removed page drags that page’s objects into the output as unreachable data: invisible in a viewer, present in the bytes. The tell is a file that lost a page but barely lost any size. For a page that must be gone for legal reasons, run the result through the compress tool, which rebuilds the document from rendered page images and carries nothing else, or use a proper redaction tool.
• Merging scans without checking for text. Photographed receipts merged into one PDF are still photographs; a reviewer cannot search them and a portal that checks for text will bounce the file. Decide before merging whether the destination needs a text layer — OCR at the scanner is far easier than OCR afterwards.
• Merging a filled form and losing the answers. Fields whose values were never rendered into an appearance stream — common with forms filled in a browser preview — come through empty. Print to PDF first, then merge, then open the result and check the values are visible.
• Numbering, then editing. Delete a page after stamping and the sequence jumps from 6 to 8; add one and two pages share a number. Stamped numbers are page content — there is no un-stamp — so they go on last, onto a copy you will not touch again.
• Sending the working copy. Every tool here writes a new file with a suffix — _extracted, _reordered, _numbered, merged.pdf — and leaves the original alone, so a three-step edit leaves four files. Rename the final one deliberately and delete the intermediates, or you will attach agreement_reordered.pdf when agreement_reordered_numbered.pdf was the one you checked.
• Assuming the metadata came along. A merged or extracted file is a new document: title, author and subject are empty and the producer reads pdf-lib. Portals that display a document title show the filename instead — another reason the filename should carry the meaning.

All of these edits, including the ones on pages you would least like to upload anywhere, happen inside your browser tab; the tools keep working with the network off once the page has loaded.