Watermarking a PDF - And What It Doesn't Protect You From

A watermark is one of the oldest ways to mark a document - stamp "DRAFT" across a page, or a company logo, and anyone who looks at it knows something about where it came from or what state it's in. It's also one of the most misunderstood: a watermark is a visible marker, not a security measure, and treating it as one is where the trouble starts.

What a watermark is actually good for

Three things, realistically: signaling a document's status at a glance ("DRAFT", "CONFIDENTIAL", "SAMPLE"), attributing a document to its source (a logo, a name), and discouraging casual reuse - someone screenshotting a watermarked page gets a screenshot with your mark on it. None of these require the watermark to be hard to remove; they only require it to be visible while the document is being read or shared normally.

What it doesn't do

A watermark added by drawing text or an image onto a page is exactly that - content drawn on top of the existing page. Anyone with a PDF editor can select and delete it, or paint over it entirely (cropping does not work for a stamp in the middle of the page; see below). It does nothing to the underlying text or images, which remain fully selectable, copyable, and extractable regardless of what's stamped over them. This is the same lesson as why hiding text in a PDF usually doesn't remove it - a visual change to how a page looks is not the same as a change to what the page actually contains.

If what you actually need is to stop people opening a document without a password, that's encryption - a different problem, which HerePDF's password protect tool handles. Stopping copying or printing is another matter: those restrictions are honoured only by readers that choose to honour them; see password-protected PDFs: what browser tools can and can't do.

What we measured: how a watermark sits in the file

"Not a security measure" is easy to say, so we checked it. We watermarked every page of 388 real documents (5,409 pages from a fixed collection of 2009-era US government PDFs; of the files with 3 to 40 pages and at most 3 MB, the 400 with the lowest SHA-256 of their thread/name) with HerePDF's watermark code three ways at the tool's default settings: a centred text stamp, a tiled text stamp and an image. Then we did the least a person could do to get it off, and read the results back.

One edit removes it. HerePDF draws the stamp as its own content stream, added after the page's content, so a page's content list reads: start, the original streams, end, stamp. Taking the last entry off that list removed the stamp and left the page's own streams byte for byte unchanged on 5,409 of the 5,409 pages with a centred text watermark, 5,409 with a tiled one and 5,409 with an image, and the text a reader sees was the same as in the original on 5,409 pages. Tiling made no difference: however many copies are on a page, they are drawn by that one stream. This is the structure of a file made by HerePDF; another tool may lay its watermark out differently.

The text is still text. A text watermark can be found by search and copied like any other text: it was extractable on 5,409 of 5,409 pages with the centred stamp and 5,409 with the tiled one. Underneath it, the page's own text was still extractable on 5,409 and 5,409 pages. A watermark changes how a page looks, not what it contains.

Cropping is not the way round it. We measured the least of the page you would lose by trimming one side until every stamp is outside it. For a centred stamp the typical loss was 66% of the page, and it was under half on 0 of 5,409 pages. That is not an accident of the settings: a stamp centred on the page cannot be trimmed away from one side without cutting past the middle of the page. For the tiled stamp it was 100%: there is nowhere to trim to. So a watermark is not weak because it can be cropped out; it is weak because it can be deleted.

The limits: these are old government documents, not yours, and only the default settings were tried. 12 of the 400 files are not counted: 10 were encrypted and 2 have damaged pages that the tool cannot draw on. We tested removal of HerePDF's own output with a library and did not try the many other ways to remove a watermark, nor any tool that rebuilds a page from a picture. The file names are not published, only counts; the data and the method are there to check against.

Opacity, rotation, and tiling

A watermark that's too opaque or too large makes the underlying page hard to read, which usually defeats the document's own purpose. Somewhere around 20-40% opacity keeps a mark clearly visible without obscuring the content under it. The classic 45° diagonal placement is a convention, not a requirement - it just reads as "watermark" instantly to most viewers, the same way a diagonal red stripe reads as "urgent."

Tiling is often suggested as the stronger layout, and it does one thing: nothing can be cropped away, because the mark is on every part of the page. Against the removal that works (above), a single stamp and a tiled one are equally easy to take off, so tiling raises the bar for cropping, not for editing. HerePDF's watermark tool supports both layouts. If the point is to make removal hard, the stamp has to stop being a separate piece of the file: turning each page into a picture (the PDF to image and Image to PDF tools) does that, and also gives up the page's text.