Dev.to Security 🔐 Cybersecurity 👁 0 📖 4 min read

Your Word Document Remembers More Than Your Words

You finished the document. You checked the spelling, fixed the wording, saved it — and sent it. What you didn't check is the part you can't see: who started the file, who touched it afterwards, how many times it was sav

You finished the document. You checked the spelling, fixed the wording, saved it — and sent it.

What you didn't check is the part you can't see: who started the file, who touched it afterwards, how many times it was saved, how long people spent inside it, and every sentence that was argued over and withdrawn.

A .docx file isn't a single blob. It's a compressed archive of XML parts, and several of those parts exist purely to describe the document rather than hold its words. Rename a copy to end in .zip and open it — the properties live in a folder named docProps.

What an Office file keeps about you

Three layers, and they leak in different ways:

  • Core properties — creator name, last modified by, title, subject, keywords, category, and creation and modification stamps accurate to the second.
  • Application properties — the organisation the software was registered to, the template the document grew from, a revision counter, cumulative editing time, page and word totals, and the producing application's version string. Generated automatically, rarely considered.
  • Custom properties — anything a business process attached: matter numbers, client codes, classification labels, workflow identifiers. The most organisation-specific, and often the most sensitive, because they were never meant to leave the building.

Then there's the revision layer, separate from properties and far more damaging. Tracked revisions hold withdrawn wording alongside the name of whoever proposed it. Comment threads hold candid internal discussion. Hidden text holds whatever somebody parked behind a formatting attribute. A document can show an empty author field and still contain a rejected paragraph naming a price you no longer offer.

What we found when we looked inside

We built a plain test document — a few paragraphs, a tracked suggestion and a comment from a second account, the suggestion accepted, then saved — and unzipped the result to read the property parts directly:

What we looked for Still present?
Creator name Yes — taken from the signed-in account profile
Last modified by Yes — named the second contributor
Revision counter Yes — incremented on every save
Total editing time Yes — recorded in minutes
Template name Yes — present even for a blank start
Accepted revision text No — correctly merged into the body
Comment thread Yes — required manual deletion

Nothing here is a flaw in the software. Every entry exists to support collaboration, version tracking and search. They become a problem only at the moment the file crosses an organisational boundary — which is the moment nobody is thinking about them.

The order that actually works

Microsoft ships a cleaning function, Document Inspector, and it is capable — provided you use it in the right order:

  1. Duplicate the file first. What follows deletes content permanently, and undo won't rescue you afterwards.
  2. Resolve the revisions yourself. Accept or reject every tracked change, then delete the remaining comments — before inspecting, so the text that survives is the text you chose.
  3. Reveal hidden text. Show all markup and formatting marks, and delete anything concealed that should not travel.
  4. Run the inspection. File > Info > Check for Issues > Inspect Document. Tick the categories — properties and personal information, comments and revisions, headers and footers, hidden text, invisible content — and press Inspect.
  5. Remove what it found. Each category gets its own Remove All button. Read what each one covers; removing headers and footers also removes page numbering.
  6. Check the Info panel. Confirm the author, company and last-saved-by entries are now blank.

Two caveats: the inspection isn't available in the browser-based editions of Office, only in the desktop applications. And saving the document again after inspecting re-stamps the modification time and may reinstate the author name from your account profile — inspect last, not first.

The PDF mistake

Converting to PDF feels like flattening, and visually it is. Underneath, the export typically carries the title and author across into the new document information dictionary and adds the name of the producing application. You've changed container and kept the attribution. If a PDF is the deliverable, clean it as a PDF.

If Office isn't installed

Because the modern formats are archives of XML parts, a browser can open the package, rewrite the property files and repackage it without any desktop software. That's what our document metadata remover does — it opens the package inside your browser, clears the core, application and custom property parts, and hands back a rebuilt file. The document never travels to a server.

One honest limitation: a property cleaner rewrites properties. It does not adjudicate your tracked revisions or delete your comment threads, because those are editorial decisions about the text itself. If the revision layer matters, resolve it in Word first and clean the properties afterwards — and confirm the result in a reader that wasn't involved in producing the file.

The short version

  • Properties name you; the revision layer remembers the argument. Clearing one does nothing to the other.
  • Resolve revisions and delete comments yourself first — the inspector is a safety net, not the editing step.
  • Inspect last, not first. A later save undoes some of the work.
  • Clean at the point of sending, not when you finish writing.
  • Start sensitive documents from blank. Reusing a previous file inherits its properties — and sometimes its custom fields.
  • The full guide has a one-page "Before you send it: the anonymize checklist" — author, company, tracked changes, comments, hidden text, headers and footers, custom properties, and a final re-check in a fresh reader — run through it in order before the file goes out: https://privacymetadata.com/blog/remove-metadata-from-word-documents/

Two things leave an Office file: the properties that name you and the revision layer that remembers the argument. Three steps — resolve, clear, confirm — take less time than writing the covering email.

Originally published at PrivacyMetadata.

📰 Read the original article on Dev.to Security

Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.