How to remove names from a document before pasting it into ChatGPT

What should I take out?

Anything that points straight to one person. Singapore’s Personal Data Protection Commission (PDPC) gives these as common examples of direct identifiers: name, email address, mobile number, social media username, and NRIC, FIN, passport, birth certificate, account and work permit numbers.

Then check details that identify someone when put together, such as date of birth, address, postal code, job title and company name. Take out any that would give the person away, or make them vaguer: “in her 40s” instead of a date of birth.

How do I do it in Word on Windows?

1. Save a copy of the file and work only on the copy. On the Review tab, choose Accept > Accept All Changes and Stop Tracking. Otherwise each name you replace stays in the file, crossed out (Microsoft).
2. List every name in every form: the full name and each part of it (“Jason Tan Ah Kow”, “Jason”, “Ah Kow”, “Tan”), “Mr Tan” or “Mdm Lim”, nicknames, initials and email addresses. A search for the full name won’t catch the other forms (US court guide). Check the signature block, headers and footers too.
3. Give each person one tag and use it every time: [Person 1], [Person 2], and so on. Numbers get their own, such as [ID 1], [Phone 1], [Email 1] or [Account 1]. Don’t use initials: PDPC says a replacement should not be something others can work out by guessing.
4. Replace the longest forms first: email addresses, then full names, then single names. Otherwise “Tan Ah Kow” can turn into “[Person 1] Ah Kow”.
5. Press Ctrl+H. Type a name in Find what and its tag in Replace with. For a single word, such as a surname, click More and tick Find whole words only, so “Tan” is not changed inside “Tanjong”.
6. Click Find Next, then Replace, one at a time. Don’t use Replace All: it also changes words you didn’t mean (Microsoft). If two people share a surname, click Find Next to skip the other one, then do a second round with their tag.
7. For NRIC and FIN numbers, click More, tick Use wildcards (Microsoft) and search for [STFGM][0-9]{7}[A-Z]: S or T for citizens and PRs, F, G or M for foreigners, then seven digits and a check letter (ICA, PDPC).
8. For phone numbers, search for [3689][0-9]{7} and then [3689][0-9]{3} [0-9]{4}, which catches numbers written with a space. Singapore numbers have eight digits and start with 3, 6, 8 or 9 (IMDA).
9. Click Find Next to step through each search, and add every number it finds to your list. Then untick Use wildcards, replace each number with its tag the same way as the names, and search for @ to catch any email address you missed.

Got a whole contract to anonymise? See How to anonymise a contract without uploading it anywhere.

What if I use Google Docs or Word on the web?

Word for the web only finds and replaces basic text, so the number searches need desktop Word (Microsoft). Google Docs can run them: use Edit > Find and replace and tick Match using regular expressions (Google). Docs has no whole-word box, so type a single name between \b marks, such as \bTan\b, to leave “Tanjong” alone (RE2).

What if I only have a PDF?

Open it in Word. Word makes an editable copy and leaves your PDF as it was. If a page is mostly pictures, Word may bring it in as an image, and then the text can’t be edited (Microsoft).

Why a tag, and not a black box?

A tag keeps each person separate, so the text still makes sense: [Person 1] wrote to [Person 2]. A black box drawn over the words, or white text, only hides them. A US court’s redaction guide says white text makes the words look like they disappear, “but they don’t!”

How do I check it before I paste?

• Press Ctrl+F and search the copy for each original name and number. Nothing should come up.
• Run the number searches again, and reread the headers, footers and signature block.
• Paste the text into Notepad, so you read exactly the words you are about to send. Then clear the formatting (the Edit menu or the formatting toolbar), or turn formatting off in Notepad’s settings first. A link can hide an email address behind its text (Microsoft). Then copy it from Notepad into ChatGPT.

Should I paste the text or upload the file?

Paste the text. A Word file can carry comments, tracked changes and names in its properties. If you must upload it, run File > Info > Check for Issues > Inspect Document on the copy and remove what it finds (Microsoft). Give the file a name with no one’s name in it, and delete any picture of a signature, an IC or a name card.

How do I put the names back?

Keep your key in a separate file on your own computer. PDPC says the table linking replacements to real names “should be kept securely as it permits re-identification.” When ChatGPT answers with [Person 1], paste its answer into a Word file on your computer and use Find and Replace to swap each tag back for the name. Don’t paste that version back into the chat.

Is the document anonymous now?

Not completely. PDPC calls removing names and NRIC numbers de-identification, “only the first step of anonymisation”. In its example, a food app removes customers’ names, and one customer can still be identified from his date of birth, gender and company. So ask: could someone who knows the business tell who this is? If so, take that detail out too.

Do ChatGPT’s settings help?

They help, but they don’t replace taking the names out. Turning off Improve the model for everyone in Settings > Data controls means new chats won’t be used to train OpenAI’s models, though they can still appear in your chat history (OpenAI). A Temporary chat stays out of your history and doesn’t create or update ChatGPT’s memory, but OpenAI may keep a copy for up to 30 days for safety purposes (OpenAI).

• Don’t rate the answer. A thumbs up or down lets OpenAI use the whole conversation for training, even after you turn training off (OpenAI).
• Don’t save a Temporary chat. Once saved, it becomes a regular chat in your history.

What if my staff use ChatGPT at work?

PDPC’s guidelines on personal data in generative AI (20 July 2026) say a business that uses an AI tool, even one it subscribes to, carries the main responsibility for making sure the tool lets it meet its duties under Singapore’s personal data law, the PDPA. They also say the business should tell its users, staff included, what kinds of personal data may go into the tool.