Remove PDF metadata

Every PDF carries an information dictionary: title, author, subject, keywords, creator, producer, and two timestamps. That is what a reader shows under Document Properties, and it is what most tools clear when they say they remove metadata.

Modern PDFs carry the same information a second time, as an XMP packet — a block of XML embedded in the file. Clearing the dictionary and leaving the packet produces a document whose properties dialog is blank and whose bytes still contain the author's name in readable text. Both are removed here.

The producer and creator fields name the software, usually with a version number: which word processor, which export plug-in, which PDF library, sometimes which operating system. For a document sent outside an organisation that is a small piece of infrastructure disclosure, and for a document meant to look like it came from somewhere else it is the detail that gives it away.

The timestamps are the other half. A PDF created at 02:14 on a Sunday says something about how it was produced, and a modification date days after the creation date says the document was revised after it was apparently finished.

The content. Text stays text, images stay images, links stay links, and the document opens exactly as it did — this rewrites the file around the pages rather than rebuilding them.

That also means anything embedded *in* the content survives: a scanned signature is still a picture of a signature, a photograph placed on a page still carries whatever its own metadata said, and text hidden under a black rectangle is still text. Removing document properties is not redaction, and no tool that claims otherwise should be believed.

A PDF records the application that created it and the one that produced the final file — frequently with version numbers. A document exported from Word names Word; one from an invoicing system names the system, which can identify the vendor a business uses.

The title field is the sharper edge. It is often inherited from the source document rather than set deliberately, so a PDF sent to a client can arrive with an internal filename or a working title in its properties, visible in any viewer’s document information panel.

These are different operations and conflating them is the mistake worth naming. Stripping metadata clears the fields describing the document: author, producer, title, dates, keywords. It does not touch anything drawn on the pages.

Redaction removes content from the page itself, and it is a separate job with a separate failure mode — drawing a black rectangle over a name adds a rectangle and leaves the name underneath, retrievable by copying the text. If something on a page must not be read, metadata removal is not the tool.

The page content, obviously, but also anything embedded in it: fonts with their own names, images with their own EXIF, attached files, and any structure the document carries. Some of that can identify a source as clearly as the metadata did.

Embedded images are the one most often missed. A photograph placed into a document brings its own metadata with it, including the camera and possibly the location, and clearing the document properties does not reach inside it. For a document assembled from photographs, the images deserve their own pass before they go in.