How to anonymize a Word document (DOCX) step by step

Word documents are the most common thing people want to run through an AI assistant: contracts, HR letters, meeting minutes, reports. They are also the files most likely to be full of names, companies, addresses and reference numbers. This guide shows how to anonymize a DOCX in a couple of minutes with a browser tool, what the manual alternatives look like, and how to keep the result useful.

What "anonymize" means here

For AI use, the goal is rarely to black out text. A blacked-out contract is hard to summarize because the model no longer knows who is who. What you usually want is pseudonymization: replace each real value with a consistent label, so "Jane Doe" becomes "Person 1" every time it appears, "Acme GmbH" becomes "Company A", and the structure of the document stays readable. If you want the finer distinctions, see pseudonymization vs anonymization vs redaction.

Option 1: anonymize the DOCX in your browser

Unformation processes the file on your device. Nothing is uploaded, which you can check with the browser's Network tab or by turning Wi‑Fi off before you start. The flow is four steps: upload, choose a rule, review, download.

Step 1: upload the file

Open the DOCX anonymizer and drop the file onto the page (or pick it with the button). You can add up to 10 files at once, each up to 50 MB. The text is extracted locally from the document's XML.

Step 2: choose a rule

Pick how detected values should be replaced. Labels ("Person 1", "Company A") are the default and the best choice for AI tasks because they keep the document readable. Redaction ("[REDACTED]"), fake values, masks ("J*** D**") and hashes are available too, and you can set a different rule per category, for example masks for card numbers and labels for names.

Step 3: review what was found

The tool lists every candidate it detected, grouped by category: people, organizations, emails, phone numbers, addresses, national IDs, IBANs, dates and so on. Detection uses three layers:

• Patterns and checksums for structured data (emails, phones, IBANs, ID numbers with check digits).
• Your dictionary: names and terms you add yourself, which are always replaced.
• An optional language model that runs in your browser and finds names and companies that patterns cannot. It is a one-time download (about 110 MB for English, about 180 MB for the multilingual model), cached for later use.

Untick anything you want to keep (for example, your own company name in a template) and add anything that was missed. Each value gets a label such as "Person 1" or "Company A". Labels are consistent: the same input always maps to the same output within the document and across the batch.

Step 4: download the anonymized DOCX and the mapping table

Choose "Download DOCX". You get a Word file with the same formatting, and you can also export a mapping table (CSV or JSON) that lists which original value became which label. Keep that table on your device; it is what lets you turn "Person 1" back into the real name later with the de-anonymizer.

What stays intact

Because replacements happen inside the document's text runs, the rest of the file is untouched:

Option 2: do it manually in Word

You can anonymize without any tool. It takes longer and it is easy to miss things, but it works for short documents.

Find and Replace

1. Press Ctrl+H (Cmd+H on a Mac).
2. Enter the real name in "Find what" and the label in "Replace with", for example "Jane Doe" → "Person 1".
3. Turn on "Match case" and "Find whole words only" to avoid replacing parts of other words.
4. Click "Replace All". Repeat for every name, company, email and number.

Tips:

• Search for first names and surnames separately as well; documents often use only one of them after the first mention.
• Search for common patterns: "@" for emails, "+" and "(0" for phone numbers, "IBAN", "Ref.", "No.".
• Check headers, footers and footnotes: Find and Replace covers them, but only if you search the whole document, not a selection.

Document Inspector

Word stores personal data outside the visible text: author names, last-modified-by, comments, tracked-change authors and custom properties. In Word go to File → Info → Check for Issues → Inspect Document, and remove document properties and personal information. Save under a new name.

Save a copy first

Always work on a copy. Once you replace and save, the original values are gone from that file, and Find and Replace has no memory of what was replaced.

Tips for consistent pseudonyms

Whether you use a tool or do it by hand, these habits keep the anonymized document useful:

• One label per entity. "Jane Doe", "J. Doe" and "Ms Doe" should all become "Person 1", not three different labels.
• Keep roles readable. If the document has a buyer and a seller, "Company A" and "Company B" are fine; adding "Buyer" and "Seller" in the first mention helps the model.
• Preserve formats. Replace an email with something that still looks like an email (email1@example.com), a date with a date, an IBAN with an IBAN-shaped placeholder. Models and downstream scripts behave better when the shape stays the same.
• Do not reuse labels across unrelated documents unless you intend to. "Person 1" in one file and "Person 1" in another should not be silently linked.
• Keep the mapping table. Without it you cannot answer "who was Person 3?" next week.

A sample mapping table

This is what the exported table looks like. Keep it local; it is the only thing that connects labels to real values.

Common mistakes

• Only replacing the first page. Names recur in signature blocks and annexes.
• Forgetting metadata. The author field alone can identify a firm.
• Redacting instead of labelling. Black bars remove context the model needs.
• Uploading the mapping table with the file. Then nothing was anonymized.

Do it now

Drop a Word file into the DOCX anonymizer. It runs in your browser, works with Wi‑Fi off, and returns a DOCX with the same layout plus a mapping table you control. If you want to see what happens under the hood first, read how it works.