Anonymize or redact a Word document without uploading it

What a Word file contains besides the visible text

A .docx file is not plain text, but a .zip package of several parts. Comments, headers, footers, notes: “Each of these additional parts is contained in the .zip package of the word-processing document”, Microsoft's technical documentation explains (our translation). Here is what the file can carry:

• The properties: author, manager, company, title, keywords. Word also records by itself “the name of the person who most recently saved a document” and the creation date.
• Comments: their text, and the names of their authors. Along with revision marks, these traces “can let other people see the names of the people who worked on your document”, Microsoft warns.
• Tracked changes: a tracked deletion stays in the file, struck through, until it is accepted. Turning off tracking changes nothing: “Word stops marking changes, but the colored underlines and strikethroughs are still in the document”.
• Headers, footers and watermarks: firm name, file reference. A field can also automatically display the author, the file name or its path there.
• Hidden text: a font effect that hides a passage on screen without removing it from the file.
• Embedded objects: an Excel sheet or a document inserted as an object. For Excel and PowerPoint, Microsoft states that these objects “can have data that is not visible”.
• Cropped images: “The cropped parts of the picture are not removed from the file and could potentially be seen by others”.
• The file name: “Durand lease – reviewed version.docx” goes out with the attachment.

Everything that follows is done in Word, on your computer: an online service would receive the whole file, with everything above. Same logic as for anonymizing a PDF locally rather than online.

On Windows: the Document Inspector

Word for Windows has a built-in tool for this hidden layer: the Document Inspector. According to Microsoft, it finds and removes comments, revisions, properties, headers and footers, watermarks, hidden text, custom XML data and invisible content. Run it last, once the names have been replaced.

1. Save a copy (File › Save As) and work on it. Microsoft strongly recommends this, because “you can't always restore the data removed by the Document Inspector”.
2. Accept or reject the tracked changes yourself (Review tab: Accept All Changes or Reject All Changes): that way you choose which version goes out.
3. Open File › Info › Check for Issues › Inspect Document.
4. Tick the categories to check, then click Inspect.
5. Next to each category to clean, click Remove All.
6. Click Reinspect: nothing should come up anymore. Save.

What the Inspector does not do

• It does not touch the visible text. Names written in the body of the document stay in place.
• It does not see everything that is hidden. Microsoft says so: “This inspector cannot detect text hidden by other methods (for example, white text on a white background).” Nor does it detect “objects covered by other objects”.
• The list Microsoft publishes for Word mentions neither embedded objects nor the cropped areas of pictures. Check them by hand. For pictures: Picture Format › Compress Pictures, box Delete cropped areas of pictures.
• It removes all the content of headers and footers, not just the names. If the layout must stay, clean them by hand.

On a Mac: no Inspector, a manual cleanup

On a Mac, Microsoft's help is clear: “Word for Mac doesn't have a document inspector.” It only suggests editing the properties by hand. The rest is cleaned item by item, on a copy (File › Save As):

1. Tracked changes: Review tab, Accept All Changes (or Reject All Changes). Hiding them is not enough: “Choosing the No Markup view only hides the changes temporarily”, Microsoft points out.
2. Comments: Review tab, arrow next to Delete, then Delete All Comments in Document.
3. Properties: File › Properties menu. In the Summary tab, clear the fields that name a person or an organization. In the Custom tab, select each unnecessary property, then Delete.
4. Cropped pictures: select the picture, Picture Format tab, Compress Pictures, then tick Delete cropped areas of pictures. Microsoft warns that this deletion “can't be undone”.
5. Headers and footers: reread them on every page.

What remains is what you cannot see, such as hidden text: that is the limit of the manual method. For a sensitive document, run the copy through the Inspector on a Windows computer before sending it.

Replace names without falling into traps

For the visible text, the tool is still Find and Replace: Home › Editing › Replace on Windows; on a Mac, the magnifying glass in the search box, then Find and Replace. Replace each person with a stable label, “[PERSONNE_1]”, “[PERSONNE_2]”, rather than with Xs: the text stays understandable. Microsoft advises clicking Replace rather than Replace All, to “review each item before replacing it”. The traps to know about:

• Variants of a name: “Mr. Durand”, “Durand”, the first name alone, initials (“J.-M. D.”), a typo, a capital without an accent (“HELENE” for “Hélène”). Each spelling has to be searched separately.
• Case: if the option that distinguishes uppercase from lowercase is on, “DURAND”, at the top of a contract, escapes a search for “Durand”.
• Names that are also words: Petit, Blanc, Marchand… Limit the search to whole words and approve replacements one by one.
• Areas outside the body text: headers, footers, notes, text boxes and comments are separate regions of the document. After replacing, search again for each name and reread these areas one by one.
• What is not text: a name in an image, a scanned signature or a logo cannot be searched. Delete or replace the image.
• Track Changes: if it is on, each replacement becomes a tracked change, and the old name stays in the file, struck through. Turn it off before you start.

Redact a Word document: delete, never cover

Redacting means removing a passage, not hiding it. We found no redaction tool in Word's online help: the Inspector handles hidden data, not passages of the visible text. Two clean ways to do it remain.

1. Delete or replace the passage in Word. First turn off Track Changes. Delete the passage, or replace it with a label or a note such as “[redacted]”, then clean the file as described above. The passage no longer exists in the document. This is the right method when the text is going to an AI: see how to anonymize a document before sending it to ChatGPT or another AI. Check the result by searching for each removed word: no result should come up.

2. Export to PDF, then redact the PDF. Useful when the layout must stay. On Windows: File › Save a Copy, PDF format, then More options › Options; choose Document rather than Document showing markup, and untick Document properties. On a Mac: File › Save As, PDF file format, then Export. The export removes no word from the visible text, and a PDF is not frozen: “Many programs, including Microsoft Word, let you open and edit PDF files.” You then need to redact the PDF with a real redaction tool (on a Mac, see redacting a PDF on a Mac), then remove the PDF's metadata. This comes up, for example, with the documents in a lawyer's case file or the draft deeds of a notary's office.

3. Redact the .docx with AnonymFiles. The app redacts the Word document itself: passages detected or selected during review are replaced with labels, the layout is preserved, and the .docx archive is rebuilt from an allowlist, so comments, tracked changes, hidden text, fields, images, objects, macros and properties are never copied over. The details are below, in an entire folder of Word files with AnonymFiles.

Check the file before sending it

Whatever the method, check the final copy, not the draft:

1. Search for each name, each variant, each sensitive number: no result should come up.
2. Open the Review tab: no more tracked changes, no more comments. Changes and comments “that have not been accepted, rejected or deleted remain in the document”, Microsoft reminds us.
3. Check the properties: on Windows, File › Info, then Show All Properties; on a Mac, File › Properties.
4. For a PDF, select a redacted area, copy it and paste it into a text editor: nothing should appear.
5. Give the file a neutral name.

An entire folder of Word files with AnonymFiles

The manual method works for one file. For a complete folder, such as the client files of a chartered accounting firm, it becomes long and the risk of missing something grows. AnonymFiles is a desktop app, coming soon for Apple silicon Macs and Windows, that processes an entire folder on your computer. It reads PDFs, text or scanned, Word .docx files, emails, Excel and CSV spreadsheets, presentations, images and .txt and .md files. The copy of a Word file is a redacted .docx that keeps its layout.

One point to know for Word files: AnonymFiles produces a .docx redacted within the document itself, layout preserved, plus a text copy (Markdown) for the AI. The body text, headers, footers and notes are processed. The archive is rebuilt from an allowlist: comments, tracked changes, hidden text, fields, images, objects, macros and properties are removed, never copied over.

The app automatically detects names of people, organizations, addresses, emails, phone numbers (French format), IBANs (nine European countries), private dates and identification numbers. It replaces them with stable labels, [PERSONNE_1], [ORGANISATION_1]…: the same person keeps the same label in every document in the folder. Amounts and other passages are masked by hand, in one click, during review. No detection is perfect: each document is reviewed side by side with its original, and nothing is exported without approval.

At export, masked content is actually deleted; metadata, annotations, forms and links are removed; files get neutral names, with an index of the documents exported and excluded. The mapping table stays on the computer, never in the export. The app then gives you an instruction to paste into the AI so that it keeps the labels intact. Finally, you put the real names back into the AI's answer, on the computer, whether it is in text, Word or PDF.

No account, no telemetry, no online license check. On a Mac, the models are included: no connection. On Windows, they are downloaded once, at the first processing run, without any document text.