Highlighting text in black is not redaction
The usual approach in Word is to select a name and apply black highlighting, or drop a filled shape over it. Both are formatting. The text underneath is unchanged: turn the highlight off, or send the file to anyone who opens it in a different application, and the name is there. Exporting that document to PDF preserves the problem exactly.
Find-and-replace is closer to correct but only reaches what you thought to search for, and it leaves every other copy of the value inside the file untouched — and a .docx contains a great many places to keep a copy.
The parts of a .docx that keep a copy
A .docx is a ZIP archive of XML parts. Cleaning the visible body text leaves plenty behind:
- Tracked changes. Deleted text is not gone; it is stored as a deletion revision. Accepting changes is a separate step people routinely forget, and the original wording is fully readable in the XML until they do.
- Comments. Reviewer names and comment bodies live in their own part and are invisible in a printed copy.
- Headers, footers and footnotes. Separate XML parts again, and typically the place a client name or matter reference is repeated on every page.
- Document properties. Author, last-modified-by, company and manager fields, filled in automatically and rarely looked at.
- Embedded images. A pasted screenshot of an email thread is a picture as far as text search is concerned. DokMine runs OCR over it and blacks out matches in the image itself.
- Text boxes, shapes and SmartArt. Stored outside the main paragraph flow.
What comes back
DokMine rewrites the document XML in place and returns a valid .docx.
Paragraph styles, tables, lists, numbering, section breaks and fonts
survive; nothing is flattened to plain text or converted to PDF. Matches
are either blacked out or replaced with a consistent typed token such as
[NAME_1], so the document still reads properly and repeated
mentions of the same person stay traceable to each other.
How to do it
- Drag the .docx onto the workbench — or a folder of them, since results are itemised per file.
- Select the entity types you need: person names, emails, phone numbers, URLs, card numbers, IBANs, IP addresses.
- Run extraction first to see every match listed by type before you commit to anything.
- Redact, download, then open the file in Word and read it.
Where it will struggle
Person names are the hardest category in any document and the one most likely to be missed or over-matched — a surname that is also a common word, a name in a language the model handles poorly, an initial standing alone. Text rendered inside a low-resolution or rotated screenshot may not OCR cleanly. Unusual constructions, deeply nested content controls and heavily customised templates can all hide text from a pass.
DokMine does not guarantee complete detection or removal. Review the returned document — including its comments, tracked changes and properties — before you send it.