Redact a PDF online, on your own machine

Think about what the usual version of this asks of you. Send your most sensitive document to a website. It is processed on their machine, sits in their storage, is reachable by their staff and their backups, and is deleted later on their schedule, which you are asked to take on trust.

A document sensitive enough to redact is sensitive enough not to send anywhere. This one does the work in the browser tab on your own computer. Your PDF is read off your disk, the pages are rebuilt there, and the download is assembled there. There is no copy on anyone else's machine and no deletion window to believe in.

Draw a black rectangle over a name in a word processor, an annotator or most PDF tools, and you have added a shape above the text. The words are untouched in the page data underneath. Select across the box and paste it somewhere. Run any extraction tool. Out they come.

Government reports, court filings and corporate disclosures have leaked names exactly this way for twenty years, and they keep doing it, because the file looks perfect on screen. If you have ever hidden something with a black box and printed to PDF, assume the text is still in there.

On a page you drew a box on, the page is redrawn as a picture at roughly 216 dots per inch, the black is painted onto those pixels, and that flat picture replaces the page in a brand new file.

The characters that were under your box are never written into the new document at all. So copy, search, text extraction and a raw look at the file's bytes all come back with nothing. Pages you did not touch are copied across exactly as they were, with their text still selectable and searchable, so the file does not balloon over a document you barely edited.

Real removal costs you something and you should know what. A page you boxed becomes an image, so all of its other text stops being selectable, stops being searchable, and stops being readable by a screen reader. That page is also bigger than it was. The practical rule that follows: redact last. Do the editing, the signing and the page shuffling first, and treat the redacted download as the copy you hand out rather than the one you keep working from. Your original is still sitting untouched on your device.

This takes about a minute and it is the only proof that counts. Open the download, not the preview. Select across each black box and paste into a text field. Press Control F or Command F and search for the word you removed. Nothing should come back.

Then scroll the whole document rather than the pages you remember. The same name usually appears five or six times: in a header, a footer, a table of contents, a signature block, a footnote and again in an appendix. Look at the edges of each box too, since a sliver of a letter poking out is enough to guess the rest.

The filename. statement-john-smith-4471.pdf tells the reader what you just spent ten minutes removing, and no tool touches it. Rename the file.

The other copies. The original on your disk, the version in your sent folder, the one you forwarded to a colleague last week, the attachment still in a chat thread. Redaction produces a new clean file, it does not chase down the old ones.

And it does not stop inference. A redacted name in a document that names three people, one of whom is obviously not it, is not really redacted. Neither is a blacked out number in a table where the other rows and the total are visible. Think about what is left, not only about what you covered.

A bank statement for a landlord or a visa office, where they need the transactions and not your full account number. A medical letter going to an insurer with a date of birth and a policy number on it. A legal exhibit with an address in it. An invoice you want to show as a reference with the client's name out of it. A passport copy for a booking. All the same shape of job: the document has to travel, and two lines of it do not.

A scanned page is already pixels, so the black goes straight onto them and there is nothing underneath to recover. One thing worth knowing: if the scan was made searchable earlier, that recognised text layer is part of the page and is dropped when the page is rebuilt. Under a black box, that is exactly what you want.

Free. One PDF at a time, up to 50 MB, with 10 finished files a day across every tool on the site, reset at midnight UTC, and no watermark on the result. The honest limit is that nothing is automatic: you draw every box by hand, there is no search-and-redact across the document and nothing scans it for names or card numbers on your behalf. That means the thoroughness is yours to supply, which is why the checking step above is not optional.