1. What PromptScrubber is
A free web tool that removes personal data from text before you paste it into ChatGPT, Claude, Gemini or any other AI assistant. It replaces each value with a stable placeholder — [PERSON_1], [EMAIL_1], [IBAN_1] — and can put the original values back when the answer arrives.
2. Why it exists
People paste real emails, contracts and support tickets into AI tools every day, often without noticing the name, the IBAN or the phone number inside. Redacting by hand is slow and easy to get wrong. A local tool that finds the most common identifiers in one pass removes most of that risk without asking anyone to trust a server.
3. Design principles
- Local by construction. There is no backend, no account and no upload. The page even ships a Content Security Policy that blocks outbound requests, so a future mistake cannot quietly send data somewhere.
- No AI in the privacy tool. Detection is deterministic: regular expressions, checksum validation (IBAN mod-97, Luhn, Italian tax code control character) and small bundled dictionaries. You can read the code and predict the behaviour.
- Honest limits. The interface says which categories are heuristic. Scanned PDFs are refused instead of being passed through with a reassuring message.
- No dark patterns. Nothing is locked behind a fake paywall, the dismissive button is as visible as the accepting one, and no optional category is pre-selected.
4. How it is built
Vanilla JavaScript and CSS, no framework and no build step. The whole application is a handful of readable files:
| File | Responsibility |
|---|---|
detector.js | Finds candidate personal data and resolves overlaps. |
anonymizer.js | Assigns placeholders and rewrites the text. |
restore.js | Puts the original values back from the local map. |
file-handler.js | Extracts text from TXT, PDF and DOCX, locally. |
app.js | Interface, state and keyboard flow. |
PDF.js and Mammoth.js are bundled in vendor/ and loaded only when you open a PDF or DOCX. No CDN is contacted at runtime.
5. Status and roadmap
The free core is complete: detection, anonymization, restore, TXT/PDF/DOCX text import, copy and download.
Not implemented, and therefore not advertised as available:
- PDF or DOCX export that preserves the original layout.
- OCR for scanned documents.
- Batch processing and a paid tier.
If a paid tier is introduced, the free text features described on this site will remain free.
6. Feedback
Detection misses are the most useful reports. Describe the pattern, not the person: see the contact page.