The most common way people redact a document is also the one that doesn't work: open the PDF, draw a filled black rectangle over the sensitive part, save, send.
The text is still there. Anyone who receives that file can select it, copy it, and paste it out — or run a text extraction tool over it and get everything in seconds. The rectangle is drawn on top of the page; it doesn't remove anything underneath.
This isn't a rare mistake. It has produced real disclosure incidents in court filings, government publications, and corporate reports, and it keeps happening because the result looks completely convincing on screen.
Why the black box fails
A PDF isn't a picture of a page. It's a set of instructions describing what to draw — the text, its position, its font — plus optional annotations layered over the top.
When you draw a rectangle with a markup or annotation tool, you add one more instruction: fill this area with black. The instruction to draw the text underneath is untouched. Rendering the page paints the text, then paints your rectangle over it. The text remains in the file, in order, ready to be extracted.
Copy-and-paste is enough to reveal it. No specialist tooling required.
Things that also don't work
Flattening the file. Flattening merges annotations into the page content so they can't be moved or deleted. It doesn't delete what's underneath — both the text and the rectangle end up in the same layer, and the text is still extractable.
Cropping. Cropping changes which part of the page is displayed. The content outside the crop box stays in the file and can be recovered by changing the crop back.
Changing the text colour to white. The text is still text. It's just invisible against the background.
Deleting a page from a stack of scans. Reasonable — but check whether the file retains earlier revisions. Some editing workflows keep prior versions inside the document.
What actually works
A real redaction tool. Proper redaction removes the underlying content rather than covering it. eleyed does this automatically: it reads the page (including scans), takes what it finds off the page itself, and leaves nothing extractable underneath. Adobe Acrobat Pro has a Redact tool for marking boxes by hand; so does Apple's Preview on macOS Sonoma and later — its Redact tool genuinely removes content, unlike drawing a shape from the same markup toolbar. If you use Preview, make sure you're using Redact and not the rectangle.
Rasterising the page. Converting the page to an image and drawing the redaction into the pixels destroys the text layer by construction — there is no text left to extract, because the page is no longer text. That is also how eleyed's default PDF path works, with OCR so scanned pages are included. A searchable text layer can be added afterwards from the already-redacted image, so nothing that was removed comes back.
Don't forget the metadata
Even a correctly redacted page can leak through document properties. PDFs carry author names, the originating software, titles that were never meant to be seen, and sometimes the original filename — Smith-disciplinary-final.pdf tells a story before anyone opens it.
Check the document properties before sending, and rename the file. The rest of that surface — comments, attachments, form fields, invisible OCR text layers — is in what still leaks after you redact a PDF.
How to verify you've done it right
Don't trust how it looks. Test it:
- Open the finished file and try to select the text under the redaction. If it highlights, it's still there.
- Select the whole page, copy, and paste into a plain text editor. Read what comes out.
- Check the document properties for author, title, and software fields.
If step two returns anything you meant to remove, the redaction failed — regardless of how solid the black box looks.
Where the effort actually goes
Using the right tool solves the technical problem. It doesn't touch the slow part.
On a single page, finding what needs to come out is trivial — you can see it. Across a few hundred pages it becomes the whole job: every occurrence of a name, every variant spelling, every phone number buried in a signature block, every reference number in a footer. Miss one and the redaction is undone, because a name blacked out on page four and left visible on page forty tells the reader exactly what was hidden.
That mechanical sweep is what automated detection is for. It doesn't tire on page three hundred, and it doesn't skip the awkwardly formatted ones because it has already read five hundred similar lines.
The judgement — whether a particular third party should come out, whether context elsewhere gives the game away — stays with you. The software's job is the sweep: every occurrence, every variant, every footer. Reviewing a marked-up document is a different task from combing a clean one, and it's the one where human judgement is actually worth something.
If you're doing this for a subject access request, the judgement calls are the substance of the job — see SAR and DSAR redaction: what UK organisations must remove. For the working process — automated detection including scans, and desktop editors if you are marking known boxes — see how to redact a PDF.