Select the information to redact, then drop your document. The first page is redacted in seconds, no account needed.
Drop your file here
PDF, JPG, or PNG (max 5MB)
*During this demo, your files are securely processed and deleted immediately. Create an account for full redaction and to manage your files and settings.
To anonymize a PDF and keep it readable, upload it to Redact PDF AI, choose Pseudonymize, review the entities found and download. Each person, company or address becomes a coloured box with a consistent tag ([Person 1], [Org 1]) on every page, the layout stays the same, and the output is flattened so the original names cannot be recovered.
Anna Keller, residing at 12 Rue du Lac, Geneva, sold the vehicle to Bella Moreau on 3 March. Ms Keller informed Moreau SA the next day.
[Person 1], residing at [Address 1], sold the vehicle to [Person 2] on [Date 1]. [Person 1] informed [Org 1] the next day.
The reader can still tell who sold, who bought and who was told. Nothing in the file identifies them.
Drop a PDF, JPG or PNG in the box above. Scanned files work: built-in OCR reads them first. Files travel over TLS and are processed on EU and Swiss-hosted Azure infrastructure.
Pick the categories to replace (people, organizations, addresses, emails, phone numbers, dates, IBANs, card numbers). The AI groups every mention of the same entity, so "Anna Keller", "Ms Keller" and "A. Keller" all become [Person 1].
In the Studio editor you can rename any tag: [Person 1] can become [Seller], [Client] or a fictitious name, and the change applies to every occurrence. You can also remove a false positive, add a term the AI missed, or switch a box back to a black redaction for data the reader should never see, such as an IBAN.
Export a flattened copy with the tags burned in, plus the correspondence table (tag → original) to keep on your side if you need to restore the names later.
Both remove the personal data from the file. The difference is what the reader is left with: a black box tells them nothing, a tag tells them it is the same person as on page 2.
Because by then the document has already left your hands with the names in it. Pseudonymizing first means the AI only ever sees [Person 1] and [Org 1], while the context it needs (who owes whom, which clause applies to which party) stays intact. Redact PDF AI processes files in the EU and Switzerland, deletes them automatically and never uses them to train AI models.
Every category has its own tag word, in the language of the document, numbered in order of first appearance across the whole document, not page by page. Each person or company also gets its own colour.
There are four common approaches. They differ on two things lawyers care about: whether the document stays readable, and whether the names ever leave your hands.
| Redact PDF AI (pseudonymize) | Black-box PDF redactor | Text anonymizer (copy-paste) | Asking the chatbot | |
|---|---|---|---|---|
| Reader can follow who is who | Yes, consistent tags | No | Yes | Usually |
| Your own labels ([Seller], fictitious names) | Yes, rename any tag | No | Depends on the tool | If you ask, not always consistent |
| Original layout kept | Yes, same PDF pages | Yes | No, plain text | No |
| Scanned PDFs and photos | Yes, OCR built in | Depends on the tool | Rarely | Depends |
| Names leave your hands before anonymization | No, only to an EU/Swiss processor that deletes them | Depends on the tool | Depends on the tool | Yes, the chatbot sees them |
| Names recoverable from the output file | No, flattened | Only if not properly flattened | No | No |
Not under the GDPR. Pseudonymized data can be linked back to a person with extra information, such as the correspondence table, so it is still personal data for whoever holds that table. Anonymized data cannot be linked back by anyone. The Court of Justice has accepted (C-413/23 P, 2025) that for a recipient with no reasonable means of re-identifying anyone, it may not be personal data. A pseudonymized PDF sent without its table can therefore be anonymous for its recipient, provided nothing else in the text identifies the people.
Initials leak information and collide: Anna and Alex both become A. A numbered tag is unique per entity and reveals nothing, and it stays the same on every page, which is what an AI or a reader needs to follow the document.
Yes. Rename any tag in the editor and every occurrence follows: use roles ([Seller], [Buyer], [Landlord]), fictitious names, or your own convention for companies ([Bank], [Employer]). Role labels give an AI the most context. If you use names, pick obviously fictitious ones and never the person’s real first name, which would reveal part of the identity.
Yes, with the correspondence table you download alongside the PDF. It stays with you; we do not need it and it is deleted with the document.
Yes. Built-in OCR reads scans, faxes and phone photos in more than 100 languages before the AI looks for names, so a scanned judgment is pseudonymized like a digital one.
Export it to PDF first (File → Save as PDF), then upload it. You get a pseudonymized PDF with the same layout as your Word file.
It removes the direct identifiers, which is what professional secrecy and the GDPR care about most. Still read it once: a rare job title or a unique event described in the text can identify someone on its own, and you can black out those passages in the same pass.