Every PDF carries information about itself that nobody puts there on purpose. Who made it, what software they used, when they made it, when they last touched it, and sometimes considerably more than that. It is invisible when you read the document and trivially visible to anyone who looks.
For most documents this does not matter. For a CV sent to a competitor of your current employer, a tender response, an anonymous submission, a legal filing, or a document leaving a company, it can matter a great deal.
What is actually in there
The document information dictionary is the familiar part: title, author, subject, keywords, the producing software, and creation and modification dates. The author field is frequently a real full name or a corporate account, because it defaults to whoever was logged into the machine.
XMP is the second, larger part — an XML block carrying the same kinds of data and more. It can be attached to the document, to an individual page, or to an object inside it, which means removing the document-level block does not necessarily remove all of it.
Then there are private blocks left by authoring applications. InDesign and similar tools write their own structures, which can include a user name, a machine name, or an internal document identifier that links this file to others produced on the same system.
Filum shows you everything it finds before you remove anything, because the first useful thing a metadata tool can do is tell you what is there.
Unlinking is not deleting
This is where most metadata removers quietly fail. A PDF is a collection of numbered objects with a table of cross-references. Removing a metadata entry can mean either of two things: unlinking the reference so nothing points at the object any more, or deleting the object itself.
Do only the first and the data is still in the file. It no longer appears in a reader's properties panel, so the document looks clean — and anyone who opens the raw bytes in a text editor can still read the author's name. The tool reports success and the information is still there.
Filum deletes the object as well as the reference, and verifies it against the actual output bytes rather than by asking the PDF library whether it believes the field is gone. Those are different questions, and only one of them is about your file.
The tool that adds its own
There is a particular irony worth knowing about. Most PDF libraries stamp a producer name and a fresh modification date into every file they save. A metadata remover built the obvious way therefore removes your metadata and adds its own — and now the document announces which website processed it and when.
Filum loads the document in a mode that suppresses that, so the output carries no producer at all, neither yours nor ours. It is a small detail that is the entire point of the tool.
The honest limit, and how to check
A photograph inside a PDF carries its own EXIF — camera model, settings, sometimes GPS coordinates — inside the image data, which is separate from the PDF's own metadata. Removing it means re-encoding the picture, which would change the image. Filum leaves photos untouched and tells you on screen when the document contains any, and only when they are real JPEGs: a screenshot pasted into a PDF has no EXIF, and warning about data that is not there would be its own small dishonesty.
That is a smaller promise than "everything removed", and it is the true one. If a document contains photographs whose EXIF matters, clean the images before assembling the PDF.
To check the result, open the cleaned file in the tool again — it will report that there is nothing to remove. Any PDF reader's document-properties panel shows the same fields. And if a file has no metadata to begin with, the tool says so and produces nothing rather than handing you a re-saved copy that merely looks processed.
All of it runs inside your browser. A tool whose purpose is that your file carries information you do not want other people to have would be a strange thing to upload to a server.