Remove PDF metadata

What is actually in there

Take a blank Form W-9 downloaded from the IRS. Nobody has filled it in and there is nothing personal on the page. Its metadata still names the internal department code that produced it, the software it was authored in and that software's version, three separate timestamps accurate to the second, and two identifiers: one for the document and one for that particular copy of it.

That is a file with nothing to hide. The interesting cases are the ones that do. A contract that went through five drafts carries the name of whoever saved it last. A resume built from a template names the template. Producer and creator fields fingerprint the software down to the version, which says more about an organization than most people expect. Timestamps say when the work really happened, which is not always when it was supposed to have happened.

The identifiers are the quiet ones. A document identifier is meant to survive being saved, so two files that came from the same original still carry the same value. That is a link between the copy you sent and the copy you kept, and between the version you published and the draft you did not.

A PDF keeps its metadata in two places

This is the part most tools skip. There is the document information dictionary, which is the old mechanism and holds title, author, subject, keywords, producer, creator and two dates. Then there is the XMP packet, a block of XML attached to the document that carries its own copy of those same facts plus things the dictionary has no room for: the identifiers, the authoring tool, sometimes an edit history.

Nothing keeps the two in step. An editor can update one and leave the other, so a file can genuinely say two different things about itself, and the older answer is usually the more revealing one. That is why the tool above shows you both before it changes anything: the information dictionary, the XMP packet with its size in bytes, every field either record names, and where they disagree, the disagreement called out at the top with both values side by side.

A disagreement is itself worth seeing. It tells you the document has been through something that only understood half of its metadata, which is a fact about the file you would otherwise never learn.

How we know this matters

Because we got it wrong. The library this is built on writes metadata through setters that reach the information dictionary and stop there, and an earlier version of this tool called those setters, saved the file, and reported the document clean.

Measured on that same W-9: after the strip, the XMP packet came back byte-for-byte identical, all 4,033 bytes of it, with eleven fields still readable. Among them the department code in dc:creator, Designer 6.5 in both pdf:Producer and xmp:CreatorTool, three timestamps, and both identifiers. The tool had cleared the record nobody inspects and left the record any forensic tool reads first.

Unlinking the packet turned out not to be enough either. Removing the reference from the catalog leaves the stream object sitting in the file, pointed at by nothing and perfectly readable to anyone who opens the bytes in a text editor. So the object is overwritten with an empty dictionary first and unlinked second. Same story as the redaction engine, learned the same way.

What gets removed

• The document information dictionary: title, author, subject, keywords, producer, creator, and the creation and modification dates.
• The XMP packet, both the document's own and any a page carries separately, including the document and instance identifiers and any edit history in it.
• Application-private data left in the file by page-layout and design software, which is where working state ends up and where it is never expected.
• Authorship on comments and markup: the reviewer's name, the timestamps, the subject line, and the name where it is repeated inside rich-text comment bodies. The comment itself stays, because a sticky note's text is content and deleting it would change what the document says.

The file identifier is the exception, and it is replaced rather than deleted. It is a pair of values in the trailer; the first is meant to survive every save, which is exactly what makes it a durable link between versions of one document, and the second changes each time the file is written. Some readers and signing tools expect the pair to be there, so removing it outright would be worse than useless. Instead both halves are overwritten with the same fresh random value. Setting them equal matters: a pair whose halves differ is itself a disclosure, because it says the file has been modified since it was created.

Form fields are left working. Stripping metadata here does not flatten anything, so a fillable form goes in and comes back fillable, minus its metadata.

What does not get removed

A photograph inside a PDF keeps its own metadata. An embedded image carries its EXIF block, which can name the camera model, the moment the shutter fired and sometimes the GPS coordinates where it happened, and that block belongs to the image rather than to the document. Clearing the document's fields does not reach inside it. If your PDF was assembled from phone photographs, strip the pictures before they go in.

The page content is untouched, and that is the one worth being clear about. Metadata is what the file says about itself, not what it shows. If a name, an address or an account number is visible on the page, stripping metadata does nothing to it and was never going to — that is redaction, it is a different operation with a real cost attached, and it is covered on how to properly redact a PDF.

Common questions

How do I remove PDF metadata?

Drop the file into the tool at the top of this page. It reads both places a PDF keeps metadata and shows you every field it found before touching anything, so you can see what is actually in the document. Press the button and you get the same file back with the information dictionary emptied, the XMP packet gone, application leftovers dropped, the names taken off any comments, and a fresh file identifier. Your form fields still work and the page content is untouched.

Can I remove PDF metadata online?

This page does it without uploading anything. Most online metadata tools send the document to a server, strip it there and send it back, which means a copy of the file exists on someone else’s machine for a while — an odd trade when the reason you are stripping metadata is that you would rather not hand your document around. Here the page loads and the work happens in the browser afterwards. Wait for it to settle, then take the network away first: reading the metadata and stripping it both still work. Loading any other page on the site is a fresh request, so that part does need a connection.

How do I remove PDF metadata for free?

There is no paid tier, no sign-up, no daily cap and no watermark. Nothing runs on a server, so there is no per-file cost for anyone to recover from you. The site is paid for by ads, which come from Google and set cookies in your browser the way ads do anywhere. Your document is not part of that: it is never uploaded, and closing the tab is the end of it.

How do I remove PDF metadata on a Mac?

The same way as anywhere else, in the browser, because this is a web page rather than an application. Nothing is installed and there is no Mac build to download. macOS gives you no command for clearing PDF metadata on its own: Preview will show you some of what a file carries under Tools, then Show Inspector, but showing it is as far as that goes, and re-saving out of Preview is not a way to clear anything because it is not trying to. Command-line tools can do it if you already have that tooling set up, which is a developer answer to a problem that usually is not a developer’s.

Can I remove all metadata from a PDF?

Not all of it, and a tool that tells you otherwise is overselling. What comes out here is everything the document records about itself: the information dictionary, the XMP packet on the document and on any page carrying its own, application-private leftovers, the names and timestamps attached to comments, and the file identifier, which is replaced with a fresh pair rather than deleted so the file stays valid. What no document-level tool can reach is metadata inside an embedded image. A photograph placed into a PDF keeps its own EXIF block, which can name the camera and sometimes the coordinates where the shutter fired, and that block belongs to the picture rather than to the document. If your file was assembled from photographs, strip the images before they go in.