Strip the identity out before the prompt leaves your machine
The fastest way to leak personal data into an AI provider’s logs is an over-shared prompt — a support ticket pasted whole, a CV, an email thread with names, addresses and phone numbers intact. Data minimisation says send only what the task needs. This tool helps you do exactly that: paste a prompt, and it detects identifiers and replaces them with consistent placeholders entirely in your browser, so the version you send carries the task but not the people.
How it works
The rewriter runs a set of pattern matchers locally and rewrites in place:
- Structured identifiers — emails, phone numbers, credit-card-like digit
sequences, IBANs, IP addresses and URLs are matched by shape and swapped for
typed placeholders like
[EMAIL_1]or[CARD_1]. - Names — runs of capitalised words that look like personal names become
[PERSON_1],[PERSON_2], and so on. - Consistency — the same real value always maps to the same placeholder, so if “Jane” appears five times the model still understands they are the same person. This preserves the structure of the task while removing the identity.
Because every occurrence is tokenised consistently, the AI can still reason about relationships (“[PERSON_1] emailed [EMAIL_1] from [EMAIL_2]”) and you can re-insert the real values into its reply afterward using the mapping the tool shows you.
Notes and tips
- Always review the output. Pattern matching is a strong first pass, but it cannot understand context — a date of birth or a rare medical term may slip through. You are the final filter.
- Keep the mapping local. The placeholder-to-real-value mapping never leaves your browser; use it to restore details in the response on your side only.
- Pair this with the AI Provider Data Retention Reference so you also know how long whatever you do send will be stored.
- For regulated or special-category data, redaction is necessary but not sufficient — confirm a lawful basis and a processor agreement first via the GDPR for AI checklist.