AI Document Anonymizer — Anonymize Documents for ChatGPT, Then Restore the Answer
Replaces the names, companies, addresses, account numbers and IDs in a Word file, spreadsheet, PDF or pasted text with consistent placeholders such as [PERSON_01], so you can ask ChatGPT, Claude or Gemini about it without sharing who it is about. When the AI answers, paste the answer back and every placeholder turns into the real value again. Everything runs in your browser: the document and the key never leave your device.
How it works
- Detection: checksummed detectors find card numbers, IBANs, SSNs, Canadian SINs, UK National Insurance and NHS numbers, Australian TFNs and more. People are found from evidence (a title, a label such as “Client:”, a contract party clause, a known first name) and companies from their ending (Ltd, LLC, Inc). Switch on deep name detection and a small AI model (distilbert-NER, about 66 MB, served from this site and run on your device) also finds names with no clue around them in English text. Anything else that looks like a name is listed for you to add with one tap.
- One placeholder per person: “Sarah Mitchell”, “Ms Mitchell” and “Sarah” all become [PERSON_01], in every document of the batch, so the AI can still follow who did what. You can rename a placeholder ([BUYER]) or merge two that are the same person.
- Format kept: Word and Excel files come back as Word and Excel, with headers, footers, comments, tracked changes, link targets and document properties anonymized too, and author names cleared.
- Restore: the key that maps placeholders back to values stays in the tab, or is saved as a file locked with AES-256 and your passphrase. Restoring copes with the ways models rewrite placeholders (PERSON_01, Person 1, \[PERSON\_01\]) and flags any it does not recognise.
- Certificate: an optional PDF or JSON record lists what kinds of data were replaced, the settings and SHA-256 fingerprints of every file, without any of the values.
Limits
Names have no fixed pattern, so a name with no clue around it can be missed, even with deep name detection on: read the highlights, check the “Possibly missed” list and add your own words before sharing. Scans are read with OCR, which can misread a letter. PDFs and images come out as text. Anonymized text can still identify someone through context (a rare job title, a unique event), so remove what the AI does not need.
How to anonymize a document for ChatGPT
Step 1: Add the document or paste text
Drop a Word, Excel, PDF or text file, or paste the text straight in. Names, companies, addresses, account numbers and IDs are highlighted where they appear.
Step 2: Check the highlights
Read through what was found and the Possibly missed list, add any word that should go, and rename a placeholder such as [PERSON_01] if a role name helps the AI.
Step 3: Anonymize and copy it for the AI
Press Anonymize, then use Copy for AI or download the anonymized Word or Excel file. Save the key file with a passphrase if you will restore the answer later.
Step 4: Restore the answer
Paste the AI's answer into Restore and every placeholder turns back into the real name, company or number.
When to use the AI Document Anonymizer
- Asking an AI to review a contract
- The model reads placeholders instead of the parties' names, and its comments come back with the real names restored.
- Summarising a customer spreadsheet
- Names, emails and account numbers are swapped in every cell, and one customer keeps one placeholder in every row, so totals and patterns still make sense.
- Drafting a reply to a complaint
- Paste the complaint, ask for a draft reply, and restore the names in the draft before you send it.
- Sending a document to a translator or proofreader
- The anonymized Word file keeps its formatting, so the work can be done without seeing who the document is about.
Common mistakes
Copying the text out without reading it
A name with no title or label around it can be missed. Read the document and the Possibly missed list before anything goes to the AI.
Removing the names but keeping the story
A rare job title, a small town and a date can identify someone on their own. Cut the details the AI does not need for your question.
Anonymizing related files one at a time
Run them together. The same person then gets the same placeholder in every file, and the AI can connect what it reads.
AI Document Anonymizer compared with other ways to use AI on private documents
| AI Document Anonymizer | Pasting the document as it is | Editing names out by hand | Online anonymizer sites | |
|---|---|---|---|---|
| Where the document goes first | Stays in your browser | Straight to the AI service | Stays with you | Uploaded to the site's server |
| Same person, same placeholder | Yes, across every file in the batch | Not applicable | Only if you keep track | Varies |
| Word and Excel formatting | Kept | Not applicable | Kept | Often text only |
| Real names back in the answer | Restored for you | Not needed | By hand | Rarely |
| Cost | Pay per file with tokens, preview free | Free | Free, but takes your time | Free with limits, or a subscription |
Frequently asked questions
How do I anonymize a document before pasting it into ChatGPT?
Drop the Word, Excel, PDF or text file on the AI Document Anonymizer, check the highlighted names and numbers, then press Anonymize. Every name, company, account number and ID becomes a placeholder such as [PERSON_01]. Copy the result into ChatGPT, Claude or Gemini. The file is processed in your browser and never uploaded.
Can I get the real names back into the AI's answer?
Yes. Paste the AI's answer into Restore and every placeholder is replaced with the original value from your key, even when the model rewrites [PERSON_01] as "Person 1" or escapes it in Markdown. A Word or Excel answer is restored in its own format. Restoring is free and runs on your device.
Is it safe to use ChatGPT with client or patient documents?
Sending identifiable client or patient data to an AI service can breach HIPAA, GDPR, UK GDPR, PIPEDA or ABA Model Rule 1.6 duties. Replacing identifiers before the text leaves your device removes most of that risk. This tool does the replacement locally, with presets for HIPAA Safe Harbor, GDPR and legal work, and nothing is uploaded.
Does the AI Document Anonymizer upload my files?
No. Detection, replacement, OCR of scans and the encryption of your key all run in your browser with JavaScript and WebAssembly. Your documents, the anonymized copies and the key that maps placeholders to real values stay on your device. Only the token charge for copying or downloading reaches our server.
Will the same person get the same placeholder everywhere?
Yes. "Sarah Mitchell", "Ms Mitchell" and "Sarah" all become [PERSON_01], and a company's short name joins its full name. The numbering is shared across every document in the batch, so the AI can follow who did what across a contract, its schedules and the related emails. You can merge or rename placeholders in the review.
Does it keep the formatting of Word and Excel files?
Yes. A Word file comes back as Word with its styles, tables, headers, footers, comments and tracked changes, and an Excel file comes back as Excel with its sheets and formulas. Hyperlink targets, document titles and author names are anonymized or cleared too. Pictures and embedded files are kept as they are and listed for you to check.
Does it work for HIPAA de-identification?
The HIPAA Safe Harbor preset replaces names, addresses, phone and email, record, plan, account and licence numbers, and cuts every date to its year, which covers most of the 18 Safe Harbor identifiers. Ages over 89 and identifiers written in free prose still need a human check, which the review screen is built for.
How much does the AI Document Anonymizer cost?
Running the anonymizer, reviewing every hit and restoring the AI's answer are free. Copying or downloading the result costs 30 tokens per document, with lower rates on larger batches, or 10 tokens for a pasted text. One charge covers both copy and download. The anonymization certificate is 15 tokens on its own.
What happens if I lose the key file?
Without the key, the placeholders in an AI answer cannot be turned back into real values, which is the point: the key is the only link between them. Save it locked with a passphrase, which is encrypted with AES-256 on your device. A lost passphrase cannot be recovered, so keep the original documents as well.
AI Document Anonymizer at a glance
| Where it runs | Entirely in your browser. Files are never uploaded. |
|---|---|
| Cost | 30 tokens per file (about $0.30 at the base rate); larger batches cost less per file: 6–10 take 10% off the total, rising to 30% off at 41–50 |
| Accepts | Word (.docx) and Excel (.xlsx) files, PDFs (scans read by OCR), text, Markdown, CSV, JSON, HTML and email files, screenshots, or pasted text |
| Produces | The same Word or Excel file with names, companies, addresses, account numbers and IDs replaced by consistent placeholders, a text version ready to paste into ChatGPT, Claude or Gemini, a key file locked with your passphrase, and the AI's answer restored with the real values. |
| Batch | Up to 50 files per run. |
| Account needed | Only to spend tokens. Browsing and previewing are free. |
| What it does not do | People are found from evidence such as a title, a label or a known first name; the optional deep name detection model (about 66 MB, run on your device) also finds bare names in English text. A name either one misses is listed as a suggestion. PDFs and images come out as text. Context, such as a rare job title, can still identify someone. |