What a watermark actually protects, and what it does not

Stamping DRAFT across a document is one of the oldest habits in office work, and a good one. It stops a preliminary version being quoted as final, which is a genuinely expensive mistake to make.

Where it goes wrong is when a watermark is treated as security. It is not, it was never designed to be, and building a process around the assumption that it is leads to unpleasant surprises.

What a watermark is good at

Its real function is changing behaviour. A reader who sees CONFIDENTIAL on every page knows the document should not be forwarded casually. A reader who sees DRAFT knows not to act on the numbers. A reader who sees SPECIMEN knows the document is not a real certificate.

That is a communication problem, and a watermark solves it well because it is impossible to miss and it appears on every page, including the page someone prints in isolation and leaves on a desk.

• Marking status: draft, superseded, review copy, not for distribution
• Marking sensitivity, so a reader knows the handling expected
• Marking samples and specimens so they cannot be mistaken for the genuine article
• Discouraging casual reuse of images and previews
• Identifying which copy went to whom, when each recipient gets a distinct mark

That last use is more interesting than it looks. Giving each recipient a subtly different watermark means a leaked copy identifies its source, which changes behaviour more effectively than any general warning.

What it cannot do

A watermark is drawn on the page like anything else, and anything drawn can be removed by someone sufficiently motivated. Depending on how it was applied, it can be deleted as an object, painted over, cropped out, or eliminated by rebuilding the page from an extraction of the text underneath.

It also does nothing to prevent the document being copied, forwarded, printed or read. Every one of those still works exactly as before. The watermark asks the reader not to; it cannot stop them.

And it does not prevent the text being extracted. The words remain in the file underneath, selectable and copyable, because a mark drawn over text does not remove it — the same principle that makes black rectangles a failed way to redact.

Applying one that works

Since the purpose is communication, the design decisions all follow from readability.

1. Place it diagonally across the middle of the page, where it cannot be cropped off without losing content.
2. Keep it light enough to read the text through, and dark enough to survive photocopying.
3. Apply it to every page, not only the first, because pages get separated.
4. Use words with unambiguous meaning: DRAFT, CONFIDENTIAL, SPECIMEN, NOT FOR DISTRIBUTION.
5. Flatten afterwards so the mark becomes part of the page rather than a removable object.

Flattening is the step that turns a trivially removable annotation into something that requires real effort to strip. It will not defeat a determined adversary, but it does defeat someone who simply opens the file and presses delete.

When you need actual protection

If the requirement is that people genuinely cannot read, copy or print the document, a watermark is the wrong tool entirely.

Password protection prevents opening without the password, which is real protection provided the password travels separately from the file. Permissions passwords restrict printing and copying, though they rely on the reader honouring them and not every reader does. Beyond that you are into rights management systems, which control access centrally and can revoke it later.

Use a watermark to tell people how to treat a document. Use encryption to stop them reading it. Do not expect either to do the other job.

The two combine perfectly well. A confidential report can be both encrypted and watermarked: the encryption controls who opens it, and the watermark tells them what to do once they have.