ScrubOpen Scrub

GUIDE · DOCX

How to redact a Word document before you give it to an AI

Deleting names from the visible text is not enough. A .docx file also carries comments, tracked changes, field codes, hyperlink targets and author metadata. Here is where to look, and how to scrub all of it in one pass.

Where personal data hides in a .docx file

A Word document is a ZIP archive of XML files. The page you see is only part of it. Before you upload a contract or a report to ChatGPT or Claude, check these places:

WhereWhat can be there
Body text, headers and footersNames, addresses, account numbers, signatures
CommentsReviewer names and internal remarks
Tracked changesDeleted text that is still in the file, plus who changed it
FieldsMail-merge results and hyperlink text that look like normal text but are stored separately
Hyperlink targetsmailto: links and URLs with an address in the query string
Footnotes, endnotes, glossaryAnything the author put there
Document propertiesAuthor, last modified by, company
Images and embedded objectsScanned signatures, screenshots, embedded spreadsheets

What Word's own tools do

File › Info › Check for Issues › Inspect Document removes comments, revisions, document properties and some hidden content. It is worth running. It does not touch names, emails or numbers in the body, and it removes comments and tracked changes rather than keeping them readable, which may not be what you want when you are asking an AI to work on a draft.

Scrubbing a .docx with Scrub

  1. Open Scrub (web app or desktop app, Documents tab) and drop the .docx file in.
  2. Review the replacements. Add client and project names to the dictionary once; they are replaced everywhere, including headers, footnotes and comments.
  3. Save the scrubbed copy and the mapping file. Treat the mapping as confidential: it holds the originals.
  4. Give the scrubbed copy to the AI and ask for the edits you need.
  5. Load the edited file back with the mapping. Scrub replaces the placeholders and keeps the formatting.

Scrub rewrites text inside the document's runs, so formatting such as bold or a split run in the middle of a name does not break the match. It also replaces the visible results of fields, addresses inside field codes and hyperlink targets, and clears the author fields in the document properties.

What it tells you it could not check

Before you save, Scrub lists what it did not inspect: images, embedded objects and any link content it could not interpret. Documents with unusual internal layouts are rejected instead of being half processed. Check the listed items yourself; a scanned signature or a screenshot of an email is still readable to an AI.

Before you upload

Questions

Does Word's Inspect Document remove names from the text?

No. Inspect Document removes comments, revisions, document properties and some hidden content, but it does not touch names, emails or numbers in the body. You still need to replace those.

Can I restore the AI-edited document?

Yes, if the AI keeps the placeholders. Scrub saves a mapping; feed the edited .docx back and it replaces [PERSON_1] and the rest with the originals, keeping the formatting.

What about images and scanned pages?

Text inside images is not read. Scrub lists images and embedded objects it could not inspect so you can check them yourself before sharing.