Here is an uncomfortable fact. Some of the most careful organisations on earth, law firms, government agencies, newspapers, have published documents where the blacked-out words could be copied and pasted straight out of the file. The box was sitting on top of the text like a sticker, and the text underneath was still there, fully intact. Anyone who knew to try could read every word.

If trained professionals with compliance teams get this wrong, the rest of us do not stand a chance by accident. And the stakes are not abstract. A blurred house number that can be sharpened back. A pixelated face that is still recognisable. A screenshot shared to show off a purchase, with a home address sitting in the corner that nobody noticed. I have done a version of that last one myself, which is part of why I care about getting it right.

So let us actually pull this apart. Not the ten-second version, the real one, because redaction is one of those things that looks trivial and quietly is not.

Hiding is not the same as removing

This is the whole game, so I want to put it first. There are two completely different things you can do to a sensitive part of an image, and they look identical on screen:

  • You can hide it, by putting something over the top or softening it.
  • You can remove it, by replacing those pixels with new ones that carry none of the original information.

Hiding feels like enough because your eyes cannot see the secret any more. But "my eyes cannot see it" and "the information is gone" are not the same sentence. A faint blur still contains a mathematical echo of what was underneath. A semi-transparent box still lets some of the original through. A black rectangle pasted onto a PDF page does nothing at all to the text living beneath it. Only the second approach, genuinely destroying the pixels, is redaction. Everything else is a costume.

Watch a weak blur fail in real time

Let me show you rather than tell you. Here is a mock order confirmation, the kind of thing people screenshot and share without a second thought. On the left, I covered the address and card line with a normal, gentle blur, the sort most people reach for. On the right, I dropped a solid block over the same area.

Side by side: a confirmation with the address under a light blur where the text shape is still visible, versus the same address under a solid black block
Left: a light blur. You can still make out the line lengths, the structure, roughly where the numbers sit. Right: a solid block. There is nothing to recover because there is nothing there.

Look at the left side for a moment. Even at a glance you can see the shape of the text, the gaps between words, the rough position of the digits. Blur does not delete information, it spreads it around. A determined person with free sharpening tools, or just patience and context, can often walk that backwards far enough to read it. For a street address with a known format and a postcode, "far enough" is not very far at all.

Why pixelation is not the safe option people think it is

Pixelation feels more aggressive than blur, more final. It is not, at least not at the strength most people use. Pixelation works by averaging a square of pixels into one colour. If the squares are small, the averaging is gentle, and gentle averaging is reversible in exactly the way blur is. Researchers have rebuilt readable text from pixelated screenshots and recovered recognisable faces from lightly mosaicked photos. There are tools built specifically to do it.

Strength is everything. Here is the same portrait at three levels:

A portrait shown three ways: original, weak pixelation where the face is still clearly recognizable, and strong pixelation where it is not
The middle one is what most people think is enough. You can still tell who that is. Only the right one, where the blocks are large enough that whole features collapse into a single tile, is genuinely safe.

The reason this is recoverable comes down to one word: predictability. Blur and weak pixelation are fixed mathematical operations, and anything done by a fixed formula can be attacked. For text especially, the attacker has a huge advantage, because they know roughly what they are looking for. A credit card number is sixteen digits in a known font. A street address follows a pattern, often ending in a postcode they can already guess from context. Software can render every plausible answer, blur each one the same way you did, and find the version that matches your blurred patch. That is not a hypothetical. People have rebuilt blurred figures from financial PDFs and recovered faces from mosaicked photos using exactly this approach.

My rule, and I am being deliberately blunt here: if you can still tell what the thing is, so can the wrong person. For a face or a number plate, reduce it until a feature is a single flat block. For anything made of text, do not pixelate at all. Cover it with something solid instead, which brings us to the part nobody talks about.

The PDF trap that catches even the careful

This is the one that bites professionals, and it is so common it has its own genre of news story. When you open a PDF and draw a black box with most editors, you have added a shape to the page. You have not touched the text. The characters underneath are still selectable, still searchable, still copyable. Export it, send it, and the recipient highlights the "redacted" region, copies it, and pastes your secret into a document. No hacking, no special tools, just Ctrl+C.

A confidential memo with a black box over a figure, and below it the same text pasted out of the PDF revealing the hidden amount and company name
The box looks final. But because it is sitting on top of live text, anyone can select the area and paste the "hidden" line straight out. This exact mistake has exposed settlement figures, witness names, and salary data in real filings.

The fix is almost stupidly simple, and it is the same fix that solves half the problems on this page. Flatten the document to a plain image first. Take a screenshot of it, or export it as a JPG or PNG. A flat image has no hidden text layer, no selectable characters, nothing living beneath the surface. Then you redact the image, where a solid block genuinely covers solid pixels. Flatten first, redact second. In that order it is hard to get wrong.

Two close cousins of this trap deserve a mention, because they fool people the same way. The first is the cropping illusion: in some apps, "cropping" a screenshot just hides the edges rather than deleting them, and the cropped-away part can be restored. The second is the layer trap in design tools, where hiding or lowering the opacity of a layer leaves the original sitting in the file for anyone who opens it properly. Same lesson every time. If the original data is still in there somewhere, you have hidden it, not removed it.

The blind spot: the data you cannot see at all

You can do everything above perfectly and still hand over the exact thing you were hiding. Photos carry metadata, a hidden layer called EXIF, and it routinely includes the GPS coordinates of where the shot was taken, the date and time, and the device. You can blur a face flawlessly, post the picture, and the file quietly announces the address it was taken at.

This is worth a habit. Before a sensitive photo leaves your device, look at what it is carrying. You can check and clear it with our image metadata viewer, which shows you the EXIF data, including any location, so you can strip it. There is a quiet bonus here too: when you flatten and re-export an image as a fresh JPG (the trick from the PDF section), most of that metadata gets dropped automatically. Doing redaction the right way cleans up the invisible data as a side effect.

So, how do you actually do it

Putting it together, the whole process is short once the principles click. Flatten whatever you are working with into a plain image. Cover the sensitive parts the right way for what they are: a solid shape over any text, strong pixelation or a solid shape over faces and plates. Export a brand new file so the redaction is baked into the pixels and cannot be peeled back off. Glance at the metadata. Done.

For the covering step, our blur and pixelate tool does the work entirely in your browser. That detail matters more than it sounds. A lot of online editors upload your image to a server to process it, which means the private picture you are trying to protect takes a trip through someone else's computer on the way to being protected. Browser-based means the file never leaves your tab. For this particular job, that is not a nice-to-have, it is the point.

A small thing that helps: after you export the redacted copy, open it fresh and try to select or zoom into the hidden area yourself. Thirty seconds of trying to break your own redaction is the cheapest insurance there is.

When none of this is worth the effort

I am not going to pretend every image is a state secret. If you are smudging a stranger in the background of a holiday photo out of basic courtesy, a normal blur is completely fine and worrying about reconstruction would be silly. The careful version is for when recovery would actually hurt someone: legal and medical paperwork, anything involving a child, financial details, a home address, a face that did not consent to being online. Match the effort to the consequences. Most of the time a light blur is fine. The trouble is that people use the light-blur level of effort for the home-address level of risk, and that mismatch is where the leaks happen.

Get the principle and you never have to memorise a checklist. Are you hiding this, or removing it? If the honest answer is hiding, you are not done yet.