Does metadata really matter?

It depends on the document. On a public report, the author field and the internal file path in the title can reveal more about an organisation than intended. On a personal document sent to one recipient, it usually matters very little.

Does compressing or converting a PDF remove metadata?

Not reliably. Some operations rewrite the document and drop fields; others copy them straight through. Check the output rather than assuming.

What is hidden inside a PDF you are about to send

A PDF carries author names, software fingerprints, timestamps, and sometimes earlier versions of its own content — none of it visible when you read the pages.

A PDF is a container, and the pages you look at are only part of what is inside. None of the following is exotic — it is present in ordinary files produced by ordinary software.

Document metadata

Every PDF has an information dictionary, and most also carry an XMP metadata block. Between them they typically record:

Title — often not the document's name but the source filename, sometimes including a full path such as C:\Users\jsmith\Clients\Acme\draft-3-final-REVISED.docx.
Author — the account name of whoever created the source document.
Subject and keywords — occasionally left over from a template.
Producer and Creator — the exact application and version used, which fingerprints your organisation's software estate.
Creation and modification timestamps, sometimes with timezone, which reveal when work was actually done.

You can see and edit all of this with PDF Info. It takes ten seconds and is the single easiest privacy improvement available on a document you are publishing.

Content that is present but not visible

Text underneath graphics. A black rectangle drawn over a paragraph hides it from your eyes. The text objects underneath are still in the page content stream, and selecting or extracting text recovers them exactly. This is the mechanism behind essentially every publicised redaction failure. See redaction versus covering.

Cropped image regions. Cropping a placed image in many design tools changes the visible window, not the stored data. The full original photograph — including whatever was outside the crop — remains embedded in the file.

Layers and hidden content. Optional content groups can be switched off for display while remaining fully present in the file.

White text. Text set in white on a white background is invisible and perfectly extractable.

A quick way to test all of these at once: run the document through PDF to Text and read the output. Anything that appears there is recoverable by anyone, regardless of what the page looks like.

Revision history

PDFs support incremental updates: rather than rewriting the file, an editor can append changes to the end and add a new cross-reference table. It is efficient and it is how signature workflows preserve integrity.

The side effect is that the earlier state remains in the file. A document edited this way can contain the previous version of a page, an earlier annotation, or a form value that was later changed. Standard viewers show only the current state, but the bytes are there.

Saving a "flattened" copy — one full rewrite rather than an append — removes this history. Any operation that rebuilds the document, such as running it through Compress PDF, generally has this effect, though it is worth verifying rather than assuming.

Annotations, comments and form values

Comments, sticky notes, highlights and review marks are separate objects from page content. Depending on the viewer and its settings, they may not be displayed — a colleague's blunt note on a draft can travel to the client unseen by you.

Form field values are likewise stored separately. Printing to PDF or flattening merges them into the page; simply saving does not.

Attachments and embedded files

The format allows arbitrary files to be embedded inside a PDF — a spreadsheet behind a report, a source file behind a diagram. These do not appear in the page view and are easy to forget.

What to do about it

Proportionality matters. An invoice to a regular supplier does not need forensic treatment. A document going to a client, a regulator, a court, or the public does.

For a document that matters:

Open PDF Info and clear or correct the title, author, subject and keywords.
Extract the text with PDF to Text and read it. Anything sensitive that appears there is not hidden, whatever the page looks like.
Check for annotations by opening the comment panel in a full viewer.
If any region was covered rather than removed, redo it properly — redaction versus covering explains the difference.
Rewrite the file once, so incremental revision history does not travel with it.

The full pre-send sequence is in what to check before sharing a sensitive PDF.

All of these checks run in your browser here, which means inspecting a confidential document for leaks does not require uploading it to find out what it leaks.