How to watermark a PDF, step by step

A watermark on a PDF copy does one thing well. It writes on every page who the copy was for and why, so a copy in the wrong inbox shows at once it was never meant to go there. I treat it as a return address rather than a lock.

What it doesn’t do is hide anything. Names, numbers and dates stay readable underneath, by people and by software. So before any watermark goes on, go through the details a recipient rarely needs to see and take those out of the copy.

How to watermark a PDF in five steps

These five steps hold whatever you use. The program only changes the menus, and each one has its own section further down. Don’t shuffle the order: the wording chosen in step one is what makes a stray copy traceable, and no menu fixes a vague line afterwards.

1. Decide what it will say. Say who receives it, why, and the month, with nothing about yourself. If the PDF is a scan of an ID, blank out the fields they have no use for first: taking a number out of a PDF properly is a different job from labelling it.
2. Pick the program. If the document still exists as a Word or LibreOffice file, mark it there and export. If it’s already a PDF, use Acrobat, Preview on a Mac, or a browser tool for a short scan.
3. Place it across the content. Large, partly transparent and over the middle of each page, where cropping it out would take the content with it.
4. Save a new PDF. Keep the original unmarked and separate, and give the marked copy a name that says who it went to.
5. Check it before you send it. Open the file you’re attaching, page by page, and confirm the wording matches this send, not the previous one.

Step five is the one I refuse to skip. A copy still marked for last month’s landlord is worse than an unmarked one, because it points the blame at the wrong recipient.

What to write on a copy of an ID or bank statement

The wording does more work than the font or the angle. A single word such as COPY fits any misuse at all. A line naming a recipient and a reason does the opposite: on a copy that lands with the wrong company, it shows at once that it wasn’t sent to them.

Keep it under one line. Our watermark tool caps the text at 80 characters, which is a sensible limit anywhere, because a long line forces a small font. And leave yourself out of it: no licence number, no phone, no birthday. Adding them repeats an identifier on every page instead of protecting it.

Some people ask a chatbot to add the watermark for them. I can’t justify sending an ID photo to one for that: what happens to pictures you give ChatGPT is a separate question, and a local tool avoids it.

Diagonal, repeated or a single stamp: where it should sit

Placement decides whether the mark survives a pair of scissors. A small stamp in a corner can be cropped off in seconds and the copy is as good as clean. A mark that crosses the middle of the page runs over the content itself, so removing it means retouching what the recipient needed to read.

For a bank or employer you already deal with, one diagonal across the page is enough. For an app, a marketplace buyer or an agent you haven’t dealt with before, the repeated pattern is my recommendation, since it covers the whole page and leaves no clean corner to cut out.

A watermark only sits on top of the page. If the PDF began as a Word draft, clear out the tracked edits before you export, or old wording can travel inside the watermarked copy.

Adding the mark in Word or LibreOffice before you export

If the document still exists as a Word file, Microsoft’s route is Design, then Watermark. Windows offers presets such as DRAFT and CONFIDENTIAL, plus Custom Watermark for your own text; the Mac dialogue adds font, size and orientation. Then save as PDF. Details in the body need Word’s own redaction steps before that save.

One catch is easy to miss. Word’s Document Inspector checks watermarks together with headers and footers, in one category called Headers, Footers, and Watermarks. Remove All on that row can take your watermark out with them. The rest of the inspection, field by field, is safe to run.

LibreOffice is free and gives you two routes. If one of your copies started life as a spreadsheet, names in the cells need clearing before export, because a watermark won’t touch them. For a PDF that matters, I prefer Format, Watermark over the export option: the export mark can’t be moved or resized.

Acrobat’s Watermark tool, and why it has a Remove button

Acrobat calls this a watermark and keeps it under its editing tools. On desktop, open the PDF and select Edit, then Watermark, Add in the left pane. The source is either typed text or a file such as a logo, and you set opacity, rotation, size and position before confirming.

Apply to Multiple Files puts the same watermark on a batch of PDFs. For a stack of copies going to one recipient, nothing else covered here is quicker, and I’d rather set the wording once than retype it file by file.

The same menu has Watermark, Remove, which works on one file or several. That doesn’t bother me, as long as nobody mistakes the mark for protection. If a number must not leave your hands, Acrobat Pro has a separate Redact tool that deletes the text rather than writing over it.

Labelling a PDF on a Mac with Preview, for free

Preview has no command called Watermark. What Apple documents is the Text tool in the markup toolbar: type your text, drag the box where you want it, and change the font, style and colour with Text Style. Repeat that on every page, because nothing copies it across for you.

Saving is the step that decides everything. Apple says File, Save, File, Export and Export to PDF all leave annotations editable, so anyone can move or delete your line. To flatten it, choose File, Print, then Save as PDF. I’d send a Preview copy no other way.

Printing to PDF writes a new file, and new files pick up properties of their own. Before it goes out, the Mac’s route to a PDF’s hidden fields is a quick check.

A free watermark for an ID photo or scan, in your browser

A passport page or a driving licence needs a different tool. Nonimo’s photo tool at foto.nonimo.ai works in the browser and the photo stays on your device. It takes a photo (HEIC, JPG, PNG or WEBP) or a PDF and can join the front and back of a card into one PDF.

The panel offers three layouts (Diagonal, Repeated and Horizontal), four suggested lines, three fonts, colour, opacity, size and an angle with quick 0°, 15°, 30° and 45° buttons. You download PDF, PNG or JPG; a PDF adds a small line naming nonimo.ai. It can also blank out the document’s number and the birth date.

The tool sits on country pages, each tuned to one document, so filter the table by where yours was issued. For most ID copies, the repeated layout with the number covered is the combination I’d back.

Why OCR and AI read straight through the mark

A semi-transparent line over a page is easy for software to read past. Google’s Cloud Vision documentation describes text detection that extracts text from any image. In a PDF that still has a text layer, nothing even needs recognising: the words sit in the file under the mark.

AI models read both. Anthropic’s documentation says that when you send a PDF to Claude, each page is converted into an image and the text from each page is extracted alongside it. A watermark changes neither. Even our tool’s output, which flattens each page into a picture, can be read back by OCR.

So the detail itself has to go. The GDPR’s minimisation rule, in EU and UK law, asks for data “limited to what is necessary”; Australia’s OAIC suggests blanking out what an agent doesn’t need. That means removing the text from the PDF, then marking the redacted copy. I don’t trust any other order.

One copy for a person, clean text for an AI

The watermarked PDF is for a human recipient: a landlord, an employer, a bank. If the next step is pasting that document into a chatbot to summarise it, the watermark adds nothing. Were the document mine, the model would get the text with names and numbers already swapped out.

Which assistant matters too. Read what OpenAI does with documents you upload before a payslip goes into ChatGPT.

For the text half, Nonimo takes the names and numbers out of the document before a model sees it, while the photo tool stays with the copy a person receives.

Sources

• Adobe Help, “Add watermarks to PDFs”: definition, the difference from stamps, Edit, Watermark, Add, and Apply to Multiple Files.
• Adobe Help, “Remove watermarks”: Watermark, Remove for one PDF or several.
• Microsoft Support, “Insert a watermark”: Windows presets, the Mac dialogue, and the Word for the web limit.
• Microsoft Support, “Remove hidden data and personal information by inspecting documents”: the Headers, Footers, and Watermarks inspector in Word.
• LibreOffice Help, “Watermark”: Format, Watermark, its five settings and the page style limit.
• LibreOffice Help, PDF export “General”: Sign with watermark, fixed position and size, not stored in the source.
• Apple Support, “Annotate a PDF in Preview on Mac”: the Text tool, and flattening with File, Print, Save as PDF.
• Google Cloud, “Detect text in images”: OCR that detects and extracts text from images.
• Anthropic, “PDF support”: each page converted to an image, with its text extracted alongside.
• EUR-Lex, Regulation (EU) 2016/679 (GDPR): Article 5(1)(c), data minimisation.
• OAIC, “Tenancy”: blanking out details a real estate agent doesn’t need, and government related identifiers.

Accessed 10 November 2026.