Redact PDF

The most consequential misunderstanding in document handling is that a black rectangle hides something. It does not. A PDF stores its text as characters with coordinates, and drawing a filled shape at those coordinates adds an object on top while leaving every character exactly where it was. Select the area, copy, paste — the text appears.

This has gone wrong in public more times than anyone can reasonably count: court filings, defence documents, corporate disclosures and regulatory submissions where the covered names were extracted within minutes of publication. The people who made those files were not careless in the ordinary sense. They used a tool that looked like it was redacting and was not.

Real redaction destroys the content. Here, that means the pages carrying a redaction are flattened into images with the marks burned in, so the characters underneath cease to exist rather than being hidden behind something. Nothing can be selected, copied, searched or recovered, because there is nothing left to recover.

Flattening a page has a real price: the text layer goes with it. A flattened page cannot be searched, its content cannot be copied, and a screen reader can no longer read it aloud. That is not a side effect to be minimised — it is the mechanism by which the hidden text is destroyed.

So the cost is charged per page rather than per document. Pages containing a redaction are flattened; pages that do not are left completely untouched, keeping their original text. A long report with redactions on three pages therefore stays searchable across the other pages, and only loses accessibility where it genuinely had to.

If the finished document needs to be fully searchable again — a common requirement for disclosure bundles — running OCR PDF over it reads the flattened pages back into a text layer. Crucially, it reads what is now visible, so the redacted content stays gone.

The hard part of redaction is not the marking, it is the missing. Sensitive information repeats, and it repeats in the places nobody rereads: a footnote, a header, a table of contents entry, an appendix, the third page of a schedule. Redacting the six obvious instances and overlooking the seventh defeats the whole exercise.

Search-driven marking is the answer for anything with a consistent form — a person's name, an account number, an email address, a reference code. It sweeps every page at once and shows you the matches to confirm, which turns a proofreading problem into a review problem.

Then check the variants by hand. Names appear with and without initials, numbers with and without spaces or hyphens, addresses in full and abbreviated. And read the document afterwards as a reader rather than as an editor: information can be reconstructible from context even when every direct mention is gone, and only reading it as a whole reveals that.

Redacting page content leaves everything outside the page untouched, and documents carry more of that than most people expect. Document properties record the author's name, the organisation, the software used and the creation and modification times. Remove Metadata clears those, and it is worth doing on anything published rather than sent to one person.

The filename is the one most often forgotten, because it is not part of the document at all. A carefully redacted file called smith-settlement-draft-final.pdf has already said a good deal before anyone opens it.

Finally, decide who is allowed to have the redacted version at all. Protect PDF adds a password for controlled circulation, and keeping the unredacted original clearly separated — different folder, different name — is what prevents the truly bad outcome of sending the wrong file.