PDF metadata: what your invoices and contracts reveal about you

Photo metadata gets all the attention, but the file format that quietly carries the most about you into other people's hands is the PDF. Invoices, quotes, NDAs, CVs, contracts: the business world runs on PDFs, and almost nobody inspects what is inside them beyond the visible page. Two separate metadata systems live in a typical PDF, and between them they can expose your real name, your employer's file server structure, your software versions, your working hours, and sometimes the document's entire pre-export history.

If you have ever sent a "final_v3_REALLYFINAL.pdf" to a client, this is the article for you.

The two layers: Document Info and XMP

The older system is the Document Information dictionary, sometimes called DocInfo. It is a simple set of key-value pairs embedded near the start of the file: Title, Author, Subject, Keywords, Creator, Producer, and a set of creation and modification dates. Creator records the application that authored the document, like Microsoft Word or AutoCAD. Producer records what generated the final PDF, like a Word export pipeline or a print driver. The dates are exact to the second, in the author's timezone, which is more information than most people realise they are handing over.

The newer system is XMP, Adobe's Extensible Metadata Platform. It is a block of XML, usually RDF, that can carry the same fields as DocInfo plus much more: full edit history in some authoring tools, document identifiers, colour profiles, and tool-specific panels. Acrobat and Creative Cloud applications are enthusiastic XMP writers. Critically, a PDF can contain both layers, and cleaning one does not clean the other. Plenty of "metadata removal" guides only handle DocInfo and leave the XMP packet sitting there with your name in it.

What actually leaks in real documents

Having looked at a lot of business PDFs, these are the recurring finds:

• Real author names behind corporate anonymity. The invoice comes from accounts@ but the Author field says the junior accountant's full Windows username. A contract drafted by outside counsel sometimes still carries the lawyer's name in metadata even though the letterhead says the firm.
• Internal file paths. Some authoring tools embed the original document path, which can reveal server names, shared drive letters and folder structures. In one leak-reviewing exercise this writer saw a path that disclosed the client's internal project codename, which was moderately awkward for all parties.
• Software and version fingerprinting. "Producer: Microsoft: Print To PDF" is harmless. "Creator: Adobe Illustrator 28.0" on a "hand-drawn" diagram, or a five-year-old Word version on a "freshly drafted" letter, tells the recipient something you may not have intended to say. Metadata dating is the same trick journalists use on government PDFs.
• Working hours and timelines. Creation date 23:47 on a Sunday, modification date two minutes before sending. Fine in most cases. Occasionally decisive in a dispute about when a document was actually prepared.
• Hidden content that was "deleted". This one is not technically metadata, but it belongs in the same risk bucket. Redacting by drawing a black box over text leaves the text in the file, selectable and searchable. Proper redaction tools remove the underlying text objects. Every year, some organisation learns this publicly.
• Attachments and embedded files. PDFs can carry embedded documents that do not appear as pages. A source file attached during drafting and forgotten is a classic.

How to inspect a PDF before you send it

The gold standard is the free command-line tool ExifTool: run exiftool document.pdf and you get every field in both metadata systems in seconds. If command lines are not your thing, the document properties dialog in any PDF reader shows the DocInfo layer, though it may not show XMP. Browser-based checkers, including our own tooling here at ExifData.io, will read both layers and show you the combined picture without installing anything. Whichever route you take, the discipline that matters is checking the actual file you are about to send, not the draft you remember.

A useful paranoid habit: inspect the file from the recipient's side of the transaction. Ask what an adversary, a competitor, or a due-diligence team would conclude from the metadata alone. Most of the time the answer is nothing interesting. You want to know on the occasions when it is not.

How to actually remove it

Options, from bluntest to cleanest:

Re-printing the PDF to a new PDF. Printing to PDF regenerates the file and typically replaces metadata with the print driver's defaults. It works for DocInfo, but it does not reliably clear XMP, it can degrade quality and clickable links, and rasterising this way is a sledgehammer. Acceptable in a hurry, not as policy.

ExifTool strippers. exiftool -all= -overwrite_original document.pdf clears both metadata systems and preserves the document. It does not remove hidden text, attachments, or previous incremental revisions of the file, and PDFs edited over time can accumulate old object versions that a simple wipe leaves behind.

Sanitise functions in Acrobat and equivalents. Acrobat's Redact tool has a "remove hidden information" pass that strips metadata, attachments, hidden layers and prior saved versions in one action. For contracts with redactions, or anything legally sensitive, this is the level you want: removal, not just clearing of the visible fields.

Set your authoring tools to leak less by default. Word lets you remove personal information on save and inspect the document before exporting. Do this before the first PDF ever leaves, and there is less to clean later.

The habit, not the tool

Photos leak where you were; we have covered that at length, including what screenshots quietly contain. Documents leak who you are, how you work and when you did it. The defence is identical in both cases: a check step between "finished" and "sent". If your routine already includes the quick pass described in our checklist for files before sending, add "inspect the PDF" to it. Ten seconds per document, and the awkward finds stop being possible.