Redact a PDF, and know that it worked

Drawing a black box over a PDF adds a drawing instruction on top of a text object that is still there. Select across it, copy, paste — and the text comes straight back out. It is how real documents leak, it has cost organisations court cases and it happens because the tool looked finished.

So a page with a match on it is not edited here. It is rendered to pixels, the boxes are painted onto those pixels, and the flattened picture is added to a new document in the original page’s place. The page object — its text, its fonts, its extractable characters — is never copied across, so there is nothing left to recover.

Flattening costs a page its text layer: it stops being selectable and searchable. So it happens only where there is something to remove. Every other page is copied object by object into the new document and is byte-for-byte what it was. You are told before you press the button how many pages will change and how many will not.

The whole promise of this page is that something is gone. That is checkable, so it is checked: the finished document is re-opened, every page of its text is extracted, and everything you asked to remove is searched for again. If any of it survived — sitting in an annotation or a form field rather than in the page text — you get a red warning and no download, because a file you were told was clean is worse than no file.

Clean a file redacts by dragging a box with the mouse, one page at a time, and it also strips the document’s metadata. That is the right tool when what has to go is a face, a signature or a logo — something with no text to search for.

This page is for the other case: an account number that appears eleven times across forty pages, where doing it by eye means finding ten of them. Both use the identical flattening method, so neither is a weaker kind of removal.

If a PDF is a photograph or a scan of a page, there is no text in it to find — only pixels that look like text. This page says so plainly rather than reporting nothing found, because “no matches” on a scan reads exactly like “your document is clean” and is not the same statement. Read it with OCR first, or drag a box over it.

Consider what a document being redacted normally is: a contract, a bank statement, a medical letter, a witness statement. Every mainstream redaction tool wants that file on its server, and several charge a subscription for the privilege. This page has no server: the parser and the renderer both run in your browser, and the page’s security policy blocks any connection to anywhere but this site. Here is how that is enforced.