How to remove metadata from PDF and Word files, step by step

Every Word file and PDF keeps a small record about itself: who created it, who saved it last, the template it grew from and, for a phone photo dropped inside, where the picture was taken. None of it prints, and all of it travels.

This guide clears that layer (file properties, PDF metadata blocks, picture data) in Word on Windows and Mac, Acrobat, Preview and LibreOffice. Hiding text on the page is a different job, set out in our Word redaction walkthrough. This one is easier to forget, I’d guess, because nothing on screen prompts it.

What hidden data does a Word document or PDF carry?

Three layers sit outside the visible text. Properties describe the file: author, last saved by, company, title, template, dates. Custom fields hold whatever a document system or a colleague added, often a matter number. Pictures keep a separate record written by the camera.

Ireland’s Data Protection Commission (DPC) says as much in its note on redacting records. File properties can hold the name of whoever created or edited a file, creation dates and revision history. Some phones and cameras add the photographer’s username, the time, the location and even lens settings.

For our test of 28 September 2026, we made up an advice letter as a .docx, redacted its body, filled in the properties and read the file back. The page never named the client; the file did, five times. A page-only sign-off doesn’t reassure me, as one leaked property raises the breach question.

Check a .docx for names before it leaves

Seeing what’s there comes before clearing it. The checker below opens a Word file inside your browser and lists the names stored as author and last saved by, plus every name attached to tracked changes and comments. Nothing is sent anywhere: the reading happens on your machine.

It also counts tracked changes, comments and hidden text, because each one carries a name. Settling those is a task of its own, covered in the guide to clearing tracked changes for good. The checker doesn’t read the title, subject, keywords, company field, template, custom properties or pictures.

So a clean result here is a first pass, not a verdict. If a name turns up that nobody expected, I’d consider the whole file suspect and run the full clean in the next section on a copy.

Document Inspector in Word, and the Mac workaround

On Windows, Microsoft’s Document Inspector does most of the work. Its Document Properties and Personal Information check covers the Summary, Statistics and Custom tabs of the properties dialog, plus usernames, template names, email headers, routing slips and server properties. Microsoft advises running it on a copy, since removed data can’t always be restored.

Word for Mac is another story. Microsoft’s own help says it doesn’t have a document inspector, so properties are changed by hand through File, then Properties. The same page says Word for the web can’t view or change properties at all, so this job needs the desktop app.

The Mac gap is my main concern, because a hand-cleared file depends on someone remembering every tab. In an office that runs on Macs, those steps belong, in my view, in the firm’s written rules for AI tools rather than in memory on a deadline.

How to remove metadata from a PDF in Acrobat

Adobe documents two routes. The light one edits properties: Menu on Windows or File on macOS, then Document Properties, change or delete the fields, check the Additional Metadata section, press OK and save. Adobe warns that deleted metadata can’t be brought back and that some permissions block the edit.

The thorough one lives in the Redact a PDF tool. After any redactions are applied, you select Continue to sanitise and remove hidden information. Adobe says that covers items such as comments, metadata and hidden layers, and that the result is saved to a new file.

For a PDF leaving the office, the second route gets my vote, because it doesn’t depend on someone spotting every field. Black boxes over words on the page are a separate matter, and covering text in a PDF so it can’t be copied back has its own guide.

Free tools: Preview on a Mac and LibreOffice

Apple’s Preview is a viewer first. Its Inspector, opened with the Info button in the toolbar, shows general details such as the author name, and a More info panel shows where a photo was taken. Apple documents removing keywords through Tools, then Show Inspector, but describes no way to clear the author.

LibreOffice goes further. Its security options include Remove personal information on saving, which strips user data from file properties, comments and tracked changes, swaps author names for labels such as Author1 and resets time values. Under File, Properties, General, Reset Properties zeroes the editing time and resets the creation date.

I’m happy to inspect a PDF in Preview but wouldn’t rely on it to clean one. For a .docx, LibreOffice is a fair free choice, and it opens spreadsheets too; hidden sheets and names in Excel files have their own guide.

Why a PDF made from Word can still name the author

Exporting to PDF creates a new file, but its properties can start as copies of the Word file’s. Adobe lists the author’s name, the title, the creation date and keywords as typical PDF metadata. A US federal court, the Southern District of Alabama, warns filers that Word metadata can survive the conversion.

There’s a second trap. A PDF can store its properties twice, in the document information dictionary and in an XMP packet, a block of XML. Our 28 September test used an invented one-page PDF made with reportlab, with both filled, then one copy with only the dictionary emptied and one with both removed.

Read with pypdf, the half-cleaned copy still named the author and the client in its XMP. So the sequence I stand by is clear Word first, export, then sanitise the PDF. A clean copy can still be forwarded, so on anything sensitive, a watermark naming the recipient is a sensible extra.

Photos inside the file: GPS and camera details

A picture placed in a .docx is stored as its own file inside the package. In our test, python-docx kept the photo exactly as supplied, EXIF block and all, and Pillow read the GPS point straight back. We didn’t test what Word itself does on insertion, so don’t assume the data has gone.

Microsoft’s Inspector page doesn’t list camera data among what its checks remove. Treat each picture as its own task. AI assistants are no clearer: what ChatGPT, Gemini, Claude and Copilot say about uploaded photos leaves the EXIF question open.

When a photo only has to show something, such as damp on a wall, a screenshot of it does the job and leaves the camera record behind. That swap makes sense to me whenever the location could point to someone’s home.

Upload the file or paste the text into an AI tool?

Paste three paragraphs into a chatbot and it gets three paragraphs. Attach the file and properties, custom fields and pictures go too, read or unread. Outside Temporary Chat, ChatGPT saves uploads to Library or Space, and deleting the chat leaves them there; what else ChatGPT keeps depends on your plan.

The GDPR’s data minimisation principle (EU and UK law), in Article 5(1)(c), asks for personal data limited to what’s necessary for the purpose. A colleague’s name in the author field does nothing for a summary. The DPC notes that exporting to plain text strips out various kinds of metadata.

For most AI tasks, text is enough, and I’d send nothing more. If you’re in Ireland, Practice Direction HC 142 has deponents declare that AI didn’t produce their evidence, and a note that the file was cleaned fits alongside that record.

Clean text and a clean copy with Nonimo

Nonimo’s desktop app, on a Mac or a Windows PC, takes a Word file or a PDF and gives back two things: the text, with each client detail swapped for a label, ready to paste, and a redacted copy of the file. Open Nonimo and drop in the copy you were about to upload.

In the Word copy, the properties that point to people are emptied or redacted and pictures lose their EXIF data, as the table shows. A PDF copy is a fresh file holding the text alone, without the original layout. Labels are masking, not anonymity: de-identified and anonymous data are different things.

The template name and the file name are left for you, and I’d rename any file still called after the client. The app does its swapping on the computer it’s installed on, so the text it works on never passes through our servers; the security page written for IT teams has the rest.

Sources

• Microsoft Support, Remove hidden data and personal information by inspecting documents, presentations, or workbooks: what the Document Properties and Personal Information check covers; inspect a copy.
• Microsoft Support, Inspect document: the Mac edition’s manual route through Properties; Word for the web can’t show or edit properties.
• Adobe, How to remove metadata from a PDF: Document Properties steps, typical PDF metadata, no undo, permissions.
• Adobe Experience League, Remove Sensitive Information: Continue to sanitise and remove hidden information, saved to a new file.
• Apple, View information about PDFs and images in Preview on Mac: Info button, General info and More info, including where a photo was taken.
• Apple, Assign keywords to a PDF or image in Preview on Mac: removing keywords through Tools, Show Inspector.
• LibreOffice Help, Security Options and Warnings: Remove personal information on saving.
• LibreOffice Help, General (File properties): Apply User Data and Reset Properties.
• Data Protection Commission, Redacting Documents and Records: file properties, camera metadata, plain-text export stripping metadata.
• US District Court, Southern District of Alabama: Best Practices, Redaction of Information: hidden metadata, revision history included, can remain after conversion to PDF.
• OpenAI Help Center, Using Library to manage files in ChatGPT: uploaded files saved automatically; deleting a chat leaves them in Library.
• Regulation (EU) 2016/679, Article 5(1)(c): data minimisation.

Accessed 10 November 2026.