How to Redact Something So It Stays Redacted
Black boxes in PDFs and documents are famously recoverable. Pixelated text has been brute-forced. Here's what actually removes information.
A black rectangle over text feels like redaction. In most file formats it isn’t. It’s a rectangle, placed on top, with the text still sitting underneath it in the file.
Select the region and copy. Open it in a different reader. Extract the text layer. The “redacted” content comes straight back.
This has happened to governments, law firms, courts and newspapers, repeatedly, and in cases where the exposed material was the identities of protected witnesses.
Why it happens
Documents and PDFs are made of objects layered in a stack. Drawing a shape adds an object above the text. It changes what renders; it changes nothing about what the file contains.
Every method of getting at the underlying content is trivial:
- Select the area and copy. The text comes with it
- Open the file in a different application that renders layers differently
- Run any text-extraction tool over it
- Delete the shape object
None of this requires skill or special software.
What actually works
Destroy the pixels, then flatten. The information has to stop existing in the file, not merely stop being visible.
For an image or screenshot, that means redrawing it with those areas replaced and re-encoding. There’s no layer to peel back because there is no layer. That’s what the redaction tool does. And it runs in your browser, which for this particular job matters more than usual. Uploading the un-redacted original to a stranger’s server is a strange way to begin protecting it.
For a PDF, use software with a genuine redaction feature, one that removes content rather than covering it, and always verify the output afterwards. Or, more reliably, export the page as an image, redact the image, and rebuild the PDF from that. You lose selectable text, which for a document you’re releasing is usually a feature.
Blur and pixelation are not equal
Solid block removes the information completely. It’s the only one of the three with no argument against it, and it’s what you should use for anything that matters: account numbers, addresses, names, signatures, licence plates, medical details.
Blur keeps shape and colour, which looks less brutal and preserves a photo’s composition. Heavy blur is irreversible in practice. Light blur is not. The underlying data is still partly present, and reversing a weak blur is a solved problem.
Pixelation is the riskiest, and by some distance the most popular. That combination is unfortunate.
Pixelated text has been recovered by brute force. The attack is straightforward: if you know the font and the pixelation block size, render every plausible string, pixelate each one identically, and find the one that matches. For anything drawn from a limited set, card numbers, dates of birth, postcodes, licence plates, phone numbers. The search space is small enough to exhaust.
Researchers have demonstrated this convincingly enough that the advice is now unambiguous:
Blur a face. Block a number.
Never pixelate anything with a predictable format.
What people forget to cover
Redacting the obvious thing and missing everything else is the standard failure. Before sharing a screenshot, check:
- The address bar. URLs carry session tokens, account IDs, internal hostnames, search queries.
- Browser tabs. Titles reveal what else you were doing.
- Notification banners that appeared mid-capture.
- The taskbar or dock. Application names say a lot about your employer and your setup.
- Autocomplete dropdowns showing previous entries.
- Reflections in glasses, monitors and windows.
- Email signatures and letterheads further down the page.
- Partially visible text behind a window you assumed was covering it.
- Scrollbar position, which indicates how much more document exists.
And the metadata
A perfectly redacted photo that still announces where it was taken hasn’t achieved much.
Phone photos routinely embed GPS coordinates accurate to a few metres, alongside the device model and timestamp. The EXIF viewer will show you what’s in a file.
The good news: redrawing an image on a canvas discards all of it as a side effect. Redacting properly strips the metadata for free.
Verify before you send
Download the finished file. Open it. Zoom in on every redacted area.
Two things to confirm: that everything you meant to cover actually is, and that nothing legible survives at the edge of a box drawn slightly too small. Partial characters are often enough to reconstruct the rest.
For anything with legal or safety consequences, have someone else check it. You know what you intended to hide, which makes you the worst available person to notice what you missed.
The one-line version
If the original content is still in the file, it isn’t redacted, it’s covered. Those are different things, and only one of them survives contact with someone who cares.
Try it yourself
Free, instant, and nothing you enter leaves your browser.