A Word file carries more than the words on the page: comments, tracked changes, headers, text boxes and the author name. Here is how to clean one by hand, where people go wrong, and when it is worth letting a tool do it.
The manual way
- Work on a copy. Save the file under a neutral name before you touch anything. The file name is part of the data.
- Write down the list. Every person and company, every address, phone number, email address, account number and ID number that appears. Include the people who are only mentioned.
- List the spellings. "Maria Lopez", "Ms. Lopez", "M. Lopez", "Lopez", "Maria", possessives such as "Lopez's", and any nickname or typo that appears in the file.
- Replace one at a time. Press Ctrl+H. Search for the longest form first ("Maria Lopez"), then the shorter ones. Use "Replace All" and read the count to see how many were changed.
- Use the same placeholder for the same person. Write the pairs in a table as you go. That table is the key. Keep it outside the chat.
- Clean the rest of the file. See the next section.
- Read the whole document once. Use the search box for each original name again. A zero count is what you want.
The places Find and Replace misses
- Headers, footers and text boxes. Do not assume the replace reached them. Open each section's header and footer and look at every text box.
- Comments. They carry reviewers' names and often quote the text.
- Tracked changes. The deleted text is still in the document. Accept or reject all changes first.
- Footnotes and endnotes. Easy to forget in a legal or financial document.
- Document properties. Author, company and title are stored in the file. In Word, Info, then Check for Issues, then Inspect Document, can remove document properties and personal information. It does not touch names in the body.
- Images. A scanned page pasted into the file is a picture, and Find does not read it. See personal data in scans and screenshots.
- Hyperlinks. A link can contain an email address or a file path with a name in it.
Keep a key table as you go
While you replace, build a simple table in a separate document:
| Placeholder | Real value | Spellings found |
|---|---|---|
| Person_1 | Maria Lopez | Maria, Ms. Lopez, M. Lopez, Lopez's |
| Company_1 | Northwind Ltd | Northwind, Northwind Limited |
This table is what lets you restore the real names in the AI's answer. It is as sensitive as the original, so keep it out of the chat and out of any folder that syncs to an AI tool. The round-trip workflow uses it.
Do not forget the people who are only mentioned
The parties are easy. The people in the margins are what get missed: a witness named in one clause, the previous owner in a recital, the accountant in a footnote, a relative listed as a contact. They are people too, and their names belong in the key table. Read the document once with only one question in mind, "who else is named here?", before you start replacing.
Paste text, not the file
If you upload a file, its name, properties and anything left inside come with it. Pasting the cleaned text into the chat sends only what you can see. Copy from the cleaned document, and read what you paste.
A last check before you paste
- Search for each original name and each first name on its own. The count should be zero.
- Search for "@" to find email addresses and for long digit strings to find account and ID numbers.
- Look at the dates. A date of birth is an identifier, and so is a date that only one event could fall on.
- Read the first and last paragraph again. Signature blocks and addresses sit there.
- Ask whether the content still identifies the client without any name. If it does, cut it.
Where the manual way stops working
For a one-page letter with two names, find-and-replace is fine. It stops working as the document grows:
- A forty-page agreement with a dozen parties and several spellings each takes a long time and still leaves misses.
- A missed "Lopez" on page 31 exposes the client, and you will not see it unless you search for it.
- Several files for the same client need the same placeholders. A table in a separate document and a lot of attention is the only safeguard.
- Inflected languages have forms that a search does not match.
With a tool
A tool finds names, companies, addresses and numbers across the whole document, gives each the same placeholder everywhere, and keeps the key for you. The good ones show what they found and let you correct it. That review step matters: detection is not perfect, and a tool that does not let you check its work is asking for trust it has not earned.
In ClientSafe you add the Word file to a case. It finds names, companies, addresses and other personal data, and highlights them in the original view. You untick anything that is not personal data and mark text it missed. The safe copy is then ready to copy for the AI. The original file is not changed.
Frequently asked questions
Does Word's Inspect Document anonymize the text?
No. It removes properties, comments and similar hidden content. Names written in the body stay.
Is it enough to rename the file?
No. A neutral file name is necessary but covers only one place.
Can I just print to PDF?
A PDF keeps the same text, and often more metadata. It does not remove anything.
Which is better for a short note?
By hand. For anything that runs to several pages or several files, a tool saves time and misses less.
ClientSafe reads Word documents together with PDFs, spreadsheets, presentations, emails and other formats, and makes one safe copy per document. Review what it found before you copy.