How to Remove Personal Info Before Pasting Into AI

Step by step

1. Open the sanitizerGo to the Prompt Sanitizer — it scans and cleans text in your browser, before anything is sent anywhere.
2. Paste your textDrop in the email, log, contract, spreadsheet row, or support ticket you were about to hand to an AI.
3. Choose what to redactPick the categories to remove — emails, phone numbers, cards, national IDs, IP addresses, URLs, secret tokens — and whether to replace them with clear labels like [EMAIL] or solid blocks.
4. Copy the clean textReview the result, copy the sanitised version, and paste that into your AI instead of the original.

Why sanitise before sending to AI

Every day, people paste real material into chatbots with names, home addresses, card numbers, health details, and API keys still in it. The moment that text reaches a cloud model it has left your control: depending on the product and its settings, it may be stored, used to improve models, or seen by a human reviewer. For regulated data — customer records, health information, anything under GDPR or similar — that can be a compliance problem, not just a privacy one. Removing the sensitive parts first, locally, neutralises the risk while keeping the content that actually needs the AI’s help.

What it detects

The tool targets structured, high-risk data that is both dangerous to leak and reliably catchable by pattern: email addresses, phone numbers, credit-card numbers (validated with the Luhn checksum to cut false positives), national-ID numbers, IP addresses, URLs, physical addresses, and long secret tokens such as API keys. Each match is replaced with a clear placeholder like [EMAIL] or [CARD], so the sentence still reads naturally and the model still understands the structure of what you asked.

Labels or solid blocks

Labelled placeholders keep the text useful — the model can see that “email [EMAIL] on [PHONE]” is a contact detail without ever seeing the real values, which is usually what you want when you need a helpful answer. Solid blocks remove even the hint of what type of data was there, which is better when you are sharing the text with people rather than a model, or when the category itself is sensitive.

Why pattern matching, and where it stops

Structured data has predictable shapes, so pattern matching catches it reliably and — crucially — does it entirely on your device, with no second AI needed and nothing uploaded. What patterns cannot know is context: an ordinary first name, a project codename, an unusual internal reference, or a sensitive fact phrased in plain prose has no fixed shape to match. That is a limit of any local detector, and it is exactly why the next step matters.

Always give the result a final read

Treat the sanitiser as a strong first pass, not a guarantee. Glance over the cleaned text before you paste it and remove anything the detectors could not know was sensitive — a name, a one-off identifier, a confidential detail written out in words. A few seconds of review turns “mostly clean” into “safe to send.”