Key point. PromptScrubber runs in your browser tab. There is no upload, no API call and no AI model; the page security policy blocks outbound requests.
The concrete problem
A document can look harmless while still carrying names, contact details, account numbers or small clues that identify someone. The risk grows when text is pasted into a chat, attached to a ticket or sent to a colleague who does not need the real identity.
Deleting identifiers by hand is slow and inconsistent. One pass catches the email address, another leaves the phone number in a signature, and a third misses the same person mentioned again in a quoted reply.
Automated detection is useful as a first pass, but it is not a promise of completeness. Structured values are easier to recognise than names or organisations described in ordinary prose.
Practical steps
- Define who will see the text and why. Keep only the context that recipient genuinely needs.
- Paste the text into the editor, or open a .txt, .pdf or .docx file. PDFs are read only when they contain an embedded text layer.
- Review the highlighted findings. Uncheck anything you deliberately want to keep, such as a public company name.
- Select any missed value in the editor and add it manually as a person, company or custom item.
- Copy the anonymized text or download it as .txt. If you exported the mapping, store it as carefully as the original document.
How PromptScrubber helps
PromptScrubber checks for structured identifiers such as email addresses, phone numbers, IBANs, tax codes, payment cards, IP addresses and recognisable keys or tokens. Person and company names come from local dictionaries and heuristics, so they still need your review.
Repeated values receive the same placeholder. That keeps a conversation readable: [PERSON_1] remains the same person in five different lines, without exposing who it is.
Limits and responsibilities
- No automated pass can guarantee that every personal detail has been found. Read the preview before sharing.
- Scanned or image-only PDFs are refused because OCR is not implemented.
- DOCX files are read as plain text; formatting, table order and images are not preserved.
- The restore map contains the original values. It is not anonymous data and must be protected.
- Placeholder-based output is closer to pseudonymization than to irreversible anonymization.
Frequently asked questions
What counts as personal data in a piece of text?
Anything that identifies or could reasonably single out a person: a name, email, phone number, account number, IP address, date of birth combined with context, or a detailed local reference. The exact answer depends on the surrounding text and on who is reading it.
Does PromptScrubber delete the original values?
It replaces them in the output text. While the tab is open it keeps a local mapping so you can restore them. Closing the tab removes that in-memory mapping, but this is not a guarantee of physical erasure from every layer of your device.
Can I use it for compliance decisions?
No. It is a text-preparation tool, not a legal assessment. Use it alongside your own privacy process and review the result before any disclosure.
What if the tool misses something?
Select the text in the editor and add it manually. Manual review is part of the workflow, not an optional extra.
Prepare your text
Open the tool: paste, review and anonymize text without uploading it.