What your PDF says about you before anyone reads a word

Where the information lives

There are three places, and they are independent of each other — emptying one does not empty the others.

• The document information dictionary. A small set of fields stored in the file: Title, Author, Subject, Keywords, Creator, Producer, CreationDate and ModDate. Every PDF reader can display these.
• An XMP metadata stream. An XML block that often carries a richer and more structured version of the same facts, sometimes including a history of edits and the tools used at each stage. It is a separate object from the dictionary, so removing one leaves the other.
• The structure of the file itself. Embedded thumbnails, attached files, optional content layers and, most importantly, the fact that many PDFs retain earlier revisions of pages in unreachable objects.

Creator and Producer mean different things, and both leak

These two fields are constantly confused because their names suggest the opposite of what they do.

• Creator is the application the content originated in — the word processor, the design tool, the scanner software. It tends to reveal how the document was actually produced, which is sometimes not how it presents itself.
• Producer is the software that wrote the PDF as it now exists. Every tool that rewrites a PDF stamps itself here, so the producer string is a record of the last thing that touched the file. A document produced by one application and then passed through three online converters carries only the last one.

Neither field is inherently sensitive, but together they date a document and identify a workflow. A file offered as an original scan, whose producer names a page-layout application, is telling a different story from the one it is being submitted with.

Timestamps are usually the most revealing field

CreationDate and ModDate are stored to the second, with a timezone offset. That is a level of precision nobody intends to publish. A contract whose modification timestamp is 02:14 the same morning it was sent at 09:00 raises a question the document did not intend to raise. A bundle of records in which every file was created within the same four-minute window is a bundle that was generated, not assembled over time.

Timezones leak location too. A document carrying an offset of +08:00 tells a reader something that a document with no visible location does not.

The revision-history problem

This is the one people find hardest to believe, and it is closely related to why black-box redaction fails. Editing a PDF normally appends a new version of the changed objects to the end of the file rather than rewriting it. That is a deliberate design of the format, and it is what makes incremental saving fast.

The consequence is that a paragraph deleted on page 4 may still be present in the file, in a version of page 4 that nothing now points at. Readers show the current version, so the document looks correct. A tool that walks the file structure can still find the old text. The same mechanism preserves an earlier version of a page that was later covered, cropped, or replaced.

The fix is a full rewrite: load the document and save it as a new file, which discards superseded objects along with whatever they contained.

What is usually not in a PDF

It is worth being precise, because the folklore is more alarming than the reality.

• Your name, unless the application wrote it — many do, from the registered user name of the machine, and that name is frequently a company or a shared workstation rather than a person.
• Your GPS location. A PDF has no standard field for coordinates. Location can survive indirectly if a JPEG containing GPS data was placed into the page without being re-encoded, which desktop publishing software normally does not do.
• Your IP address or machine name in any standard field. Producer strings occasionally include a version number and a licence name, and a handful of tools have historically written a hostname, but this is not a general property of the format.
• The contents of pages you replaced, unless the incremental-update situation above applies.

Checking your own files in about a minute

• Open the file in a PDF reader and find Document Properties — usually under a File, View or Info menu. Everything in the information dictionary is displayed there, usually as a plain list.
• Ask what the Creator and Producer fields say about the software and the pipeline. If either names an application you would not want to disclose, that is the whole finding.
• Compare the creation and modification dates with what the document claims about itself.
• If the stakes are higher, use a dedicated metadata reader such as ExifTool, which also dumps the XMP stream and the file structure rather than just the friendly fields.
• Copy the file, then re-save it through a tool that rewrites the document, and check its properties again to see what changed.

Removing metadata locally — what this site does and does not do

Being straight about this matters, because a privacy-focused PDF site that quietly preserved an author name would be a bad joke. The honest answer is that the behaviour differs by tool, and you should check rather than assume.

Tools that build a new document from the pages — compression, page extraction, cropping, image-to-PDF, splitting — produce a file with a fresh, generic metadata dictionary. The original title, author, creator application and creation date are not carried over. Measured on a Word-exported test file whose dictionary read Author: MC SYSTEM, Title: 集合的概念, Creator: Microsoft Word 2013, the compressed copy carried none of them: only a generic producer string and the date it was produced.

The text-to-outlines tool behaves differently, and deliberately so. It rewrites the existing document rather than building a new one, so the original information dictionary survives intact — title, author, creator and original dates all remain — and only the producer string is replaced. If you outline a file and send it on, the authoring history goes with it.

What no tool here does is offer you a metadata editor. There is no button that clears a specific field, because doing that properly means understanding which of the three storage locations you are trying to empty, and a partial job is worse than none — it creates the impression of removal without the fact of it. If you need fields cleared deliberately, use a metadata editor that shows you the dictionary and the XMP stream side by side.

One crude method that does work locally: render the pages to images and rebuild a PDF from them. The rebuilt file carries none of the original metadata and no text layer at all. It is heavy-handed — you lose text, links and form fields — but for a document that only has to be looked at, it is a complete and verifiable reset.

Before sending any document that matters, open its properties and read what is there. Two seconds of looking beats discovering it in somebody else's hands.