Redaction removes content rather than covering it. Draw boxes over whatever must not be seen, and on save the affected pages are re-rendered as images with the boxed areas gone — so the text underneath is not merely hidden, it is absent from the file. That distinction is the entire point, and it is where most attempts at redaction fail: a black rectangle drawn in an editor sits on top of text that copies straight back out, which is how confidential names, figures and addresses reach the public in documents that looked redacted. Because the pages are flattened to images, the redacted pages stop being searchable and their text stops being selectable — a real cost, accepted deliberately, because it is what guarantees nothing survives underneath. Everything happens in your browser, so the unredacted document is never uploaded anywhere.
Why covering does not work
A PDF stores text and graphics as separate objects in a stack. Drawing a filled rectangle adds an object on top; it does not alter the text object beneath, which remains complete in the file. Select the region and copy, or run any text extractor, and the hidden text comes out intact.
This has produced a long series of public failures — court filings, government disclosures and corporate documents released with names, addresses and figures recoverable by anyone who pressed Ctrl-A. The visual result of covering and redacting is identical, which is exactly why the mistake keeps being made.
What happens when you save
Each page carrying a redaction box is rendered to a bitmap with the boxed regions painted out, and that bitmap replaces the page. There is no text layer left on it, no vector objects, and nothing underneath the black areas — because the page is now a picture in which those pixels are black.
Pages you did not touch are left as they are, so a hundred-page document with two redacted pages keeps ninety-eight pages of selectable text. The trade-off is contained rather than applied wholesale.
What redaction cannot reach
It works on what is visible on the page. Document metadata — author, title, creation software — sits outside the page content and is not touched, and neither are attachments or the contents of comment threads if the file carries them.
Nor does it know what is sensitive. Check the whole document rather than the paragraph you were told about: names repeated in headers, an email address in a footer, a filename in a screenshot on another page. Redaction is a precise instrument applied by an imprecise process, and the review is the part that matters.