Auto Redact PDF
Local only — nothing is uploaded unless you save or share it.
Auto redact PDF files: it finds the personal information, you confirm, the words are deleted.
- Drop in the PDF you want cleaned up.
- It reads the text and lists what looks personal — tick anything that should go.
- Press Auto redact, then download the file with those words removed.
Names and addresses need AI
The list above was found by pattern — emails, phone numbers, cards, CPR numbers. Names and postal addresses have no pattern to match, so finding those takes a model reading the document.
Your document is not uploaded — only its text is sent, and only when you press this.
Save this tool
Keep Auto Redact PDF handy
Use your browser's Share menu, then “Add to Home Screen”.
About this tool
It is a pdf redaction tool with the searching built in. To automatically redact pdf files you first have to find what is in them, and redacting by hand means reading every page looking for the things you would rather not send. This does that reading for you and shows you what it found, so the job becomes ticking a list instead of hunting. It works in two halves — only one of which costs anything. Email addresses, phone numbers, card numbers, IBANs and Danish CPR numbers all have a recognisable shape, so those are found by pattern — free, offline, in your own browser, and the same answer every time you run it. That is most of what people mean when they say redact pii. Names and postal addresses have no shape at all; recognising "Havnegade 12" as an address takes a model, and that half uses one AI credit. To remove personal information from pdf pages you still confirm each one — nothing is removed until you say so. Every find is a suggestion with the text shown beside it, because both halves are wrong in opposite directions: a pattern will offer an invoice number that happens to look like a phone number, and a model will miss a surname it has never seen. A tool that quietly deleted whatever it guessed would be worse than one that deletes nothing, because you would not find out until the document had already been sent. It will anonymise pdf output properly, which is where most tools fail. When you confirm, the removal is the real thing. The page is flattened and the words are gone from the file — not a black rectangle drawn over text that is still underneath, which is how law firms and government departments have leaked documents for twenty years. Select over a normal redaction, paste into a text editor, and the name comes back out. Not here.
Frequently asked questions
What counts as personal information here?
By pattern, and for free: email addresses, phone numbers, payment card numbers (Luhn-checked so invoice numbers are not offered), IBANs (checked with their own mod-97 digits), Danish CPR numbers (the date is validated), UK National Insurance and US Social Security numbers, IP addresses, UK and Danish postcodes, and web addresses. It also guesses at two things a rule normally cannot: a street name followed by a house number, and a name introduced by a title like Dr or Fru. With the AI pass: names of people and postal addresses generally, which is how it can remove names from pdf documents at all — no rule can find an untitled name. It deliberately does not offer company names, job titles, dates or amounts.
Is my document uploaded?
The pattern half never leaves your browser at all — it is a set of rules applied to text on your own machine. The AI half sends the document's **text**, not the file, because the words are all a model needs to spot a name. Nothing is stored either way, and if you only use the free half nothing is transmitted at all.
Why do I have to confirm every find?
Because to find personal data in document text is a judgement, and both halves make mistakes, and in opposite directions. Patterns are literal: a long reference number can look like a phone number. A model is not literal: it can miss an unusual surname or offer a street that is actually a company name. Neither is safe to act on unsupervised, and the failure would be silent — you would only discover it after sending the document. Ticking a list takes twenty seconds and removes that risk entirely.
Does it really delete the text, or just cover it?
Deletes it. The page is rendered and flattened, so the characters are no longer in the file — you can check by opening the result and trying to select the redacted area, and there is nothing there to select. This matters more than it sounds: the standard failure is a black rectangle drawn on top of text that stays in the document underneath, and it has embarrassed law firms, newspapers and government departments repeatedly.
What about scanned PDFs?
A scan is a picture of a page with no text layer, so there is nothing to read and nothing will be found. The tool says so rather than reporting a clean document, which would be the dangerous answer. For a scan, run it through OCR first, or use the manual redaction tool and draw the boxes yourself.
Why does the AI half cost a credit?
Because it is the part with a real cost per run — a model reads the document's text every time. The pattern half costs us nothing, so it costs you nothing, and it finds most of what people actually want removed from invoices and forms. The credit buys the names and addresses, which is the part no rule can find.
Help improve this tool
Found a bug, want a change, or need a tool we don't have?
You have an unsent draft.
Thanks — that's with us. You'll find the reply in your inbox.Go to my messages