Tool · Local detection
Clean the text before you hand it to an AI
Paste what you were about to send an assistant and see what personal data it carries. Each match is swapped for a label so the model keeps the context without receiving the data.
Nothing leaves the browser
100% Local. AIs use your data to train, don't give away your privacy.
Automatic pattern matching: it may miss things. Review the result before sending it.
What the tool looks for
The personal data that slips into a working document usually has a recognisable shape, and that is what makes automatic detection possible:
- Email addresses
- Phone numbers, national and international formats
- Payment card numbers
- IBAN account numbers
- National identity numbers
- IP addresses
Each match is replaced with a label such as [EMAIL] or[PHONE]. The structure of the text stays intact, so the assistant still understands that there was an email there and can write accordingly.
What it does not catch — and why that matters
No automatic detector catches everything, and trusting that it does is worse than using none at all. By design, it misses:
- Proper nouns. Telling "Marta Ruiz" apart from any other pair of words would require understanding the text, not recognising a pattern.
- Postal addresses written in free form.
- Data that is only sensitive in your context: a project codename, a revenue figure, a client's name.
- Anything inside attachments or screenshots you have not pasted as text.
So the output is a draft to review, not a certificate. Read it before you copy it.
The complete method, in three steps
- Replace, do not delete. Leave the labels where the data was. If several people are involved, number them:
[CLIENT_1],[CLIENT_2]. - Review by hand what the tool cannot know: names, internal nicknames, figures that would expose you.
- Undo the swap in the answer. When the AI returns the text, put the real data back in your own editor. It takes ten seconds.
When none of this is necessary
Most real work needs no personal data at all: drafting, summarising, translating public content, explaining a concept, reviewing an argument or writing code with no credentials inside. For all of that, pasting directly is perfectly reasonable.
Caution starts the moment the text contains an identifiable person or a business secret. And there is an important difference between the two: with your own data you decide; with someone else's, you do not.
Frequently asked
Common questions
Is the text I paste here sent anywhere?
No. Pattern detection happens entirely in your browser and nothing is transmitted or stored. It is the exact opposite of what happens when you paste that same text into a chatbot.
Does it catch every piece of personal data?
No, and that is worth saying plainly. It reliably catches things with a fixed format: emails, phone numbers, cards, IBANs, national ID numbers, IP addresses. An unusual proper noun or an internal project codename will slip through. The final review is yours.
Why replace with labels instead of deleting?
Because deleting the data outright leaves the text incoherent, the AI answers worse, and you end up pasting the original anyway. With consistent labels the model keeps the context and receives no real data.
Does this make me legally compliant when handling client data?
It helps, but it does not replace a review. If you process other people’s personal data at work, responsibility for where you send it remains yours and depends on the agreements you have with the provider.
Where to go next