What a merge actually copies

Published 2026-11-02 · Last reviewed 2026-11-06

A merge looks like a copy, and that word is why so many people are surprised twice: once when a signed contract comes back without its signature panel, and once when the combined file is bigger than the three files they started with. Both follow from the same fact about the PDF format - a merge copies pages and nothing else - and both are easier to plan around once the mechanism is on the table. This page is about the mechanism, and about the bill the tab pays for it.

Nothing leaves the tab

The files are read through the browser's own file picker, so the bytes land in the tab you already have open. The page then opens each document, copies its pages into one new document in JavaScript, and hands the result back as a download made from a Blob. There is no upload step in that sentence, which is the part that matters for a payslip, a passport scan or a medical letter: the file never travels, so there is no server copy to be deleted, forgotten or breached.

It also means the work is bounded by your device. A merge of a few dozen pages is a fraction of a second; the slow part is the disk read, not the page copying. The Merge PDF tool is the thing this page explains, and it prints the page count and the size of every input before it writes anything.

Pages carry content, documents carry paperwork

A PDF is a container of numbered objects - pages, fonts, images, streams of drawing instructions - tied together by a page tree and a cross-reference table. Merging reads the page tree of each input, copies the page objects and the resources they point at into a new document, and writes a new table for the result. Nothing is re-rendered, so text stays text: the merged file is still selectable and still searchable, images keep their original resolution, vector art stays sharp at any zoom, and each page keeps its own size and orientation.

Everything stored once per document rather than on a page is left behind, and that list is exactly the one people notice afterwards: bookmarks and outlines, the fields of a fillable form, signature fields, layer visibility, and the structure tags a screen reader follows. A merged file is therefore untagged, and a signed document that has been merged is no longer a verifiable signed document - the visible signature survives as page content, the cryptographic proof does not. Password-protected files are refused rather than half-merged, because there is nowhere on this page to type a password and guessing at an encrypted document produces a broken file instead of an error.

What the job costs the tab

Memory is the real limit, and it is worth knowing the shape of it. The merge holds the input bytes and the output bytes at the same time - it has to, because the new document is written while the old ones are still being read - so the peak is roughly twice the job. The caps on this tool are 25 MB per file, 80 MB for the whole job, and 20 files at once, which puts the arithmetic peak near 160 MB in one tab before the parsed object graph is counted.

Those numbers are a budget, not a preference. A tab that runs out of memory does not raise a helpful error; it is killed, usually with the work gone and nothing on screen to explain it. Merging in batches keeps the same result and stays well inside the budget. It is also why the output can be larger than the inputs added together: each input brings its own fonts and images along, and nothing is deduplicated between them - two files embedding the same font leave two copies in the merge.

Where these numbers come from

• This site does the merging itself. Page copying, the order of the output and the count of pages per input all follow the same merging steps, and every release automatically checks them. The tool and this guide use the same steps, so they cannot print two different answers for one set of files.
• The limits are a single published set of caps, checked before a byte is read. The same limits that stop an oversized job are the ones quoted in the paragraphs above, and the byte figures printed here - 25 MB and 80 MB - are formatted from those limits rather than typed in by hand.
• The PDF part is a processing engine, not a guess. It runs in your browser under the MIT license. What it preserves and what it does not is a property of the PDF format's split between pages and document catalogue, not a setting this site chose.

What this guide does not cover

It does not tell you how to get bookmarks back into a merged file: they have to be added afterwards, in a PDF editor, because an outline is a document-level structure and a merge writes a new document. It does not compress anything either - merging is lossless and deliberately does not re-encode images, so the reduction people expect from a "PDF compressor" is a separate job, done on a separate page here. Nor does it model OCR: a scanned page is still an image of text after merging, however the file is arranged, and this page is about arrangement rather than recognition.

Put your own files through the Merge PDF tool to see the page counts and sizes described above, take a subset apart in the Split PDF tool, or make the result smaller with the Compress PDF tool once the order is settled.

Explanations and estimates only - not legal advice about a signed document. Read the full disclaimer.