Redaction mistakes that leak
The same small set of mistakes accounts for nearly every unredacted document that has escaped into the world. None of them are careless in the ordinary sense: each one produces a document that looks correct on screen. That is exactly why they keep happening.
Covering instead of removing
A black rectangle, a highlight in black, a white-filled shape, an image pasted over the area — all of these paint on top of the page and leave the text in the file. So does setting the text colour to white, which is the same mistake with extra steps: invisible is not absent, and selecting the region reveals it immediately. If your method is "put something over it", the text is still there.
Blur and pixelation
Blurring or pixelating a region of an image is genuinely destructive — the original pixels are gone — but it is far weaker than it looks. Pixelation applied to text is often reversible in practice, because the number of possible strings that could produce a given block pattern is small and can be searched; the same is true of some blurs. Short, structured data like a card number, a postcode or a licence plate is the worst case. If the content matters, cover it with a solid fill rather than obscuring it.
Cropping, hiding and flattening that did not
Cropping the page hides content without deleting it and is undone from a menu. Hiding a layer leaves the layer in the file. Deleting a page from a viewer sometimes only unlinks it while the objects remain, especially when the file is saved incrementally. And "flatten" means different things in different tools: in some it merges annotations into page content, in others it rasterises, in others it merely locks form fields. Assuming flatten equals redact is how documents that went through a professional workflow still leaked.
Inconsistent redaction — the one that survives good tools
Even with a perfect technical redaction, a document leaks when the same detail is removed in one place and left in another: the name blacked out in the body but present in the signature, the footer, an email address or the file's own metadata. Related is the arithmetic leak — every line of a table redacted except the total, from which the missing figure is recoverable — and the identification leak, where a redacted name is still obvious from a job title, a date and a location. Search for every occurrence, and read what is left as if you were trying to work out what was removed.
Frequently asked questions
Is blurring safe for redacting text?
Not really. Blur and pixelation applied to short, structured text can often be reversed, because there are few enough possible originals to test. Use a solid fill.
Is white text safe?
No. It is invisible, not absent — selecting the area or extracting the text reveals it immediately.
Does cropping remove the content outside the page?
No. It changes what is displayed. The content stays in the file and the crop can be undone.
Does flattening count as redaction?
It depends entirely on what your tool means by flatten, which is why it is not a reliable instruction. Verify the output instead of trusting the word.
I redacted the name everywhere in the body — is that enough?
Check the signature block, the footers, any email addresses, the file name and the document properties. Partial redaction is the most common failure with a correct tool.
Can a redacted figure be worked out from the rest of the table?
Often, yes. If the total and every other line survives, the redacted line is arithmetic. Redact the total too.
Remove the text, then prove it is gone. Redact takes the words out of the page rather than covering them, then reopens the file it just made and searches it twice — by text extraction and by raw stream scan — before letting you download. The file never leaves your browser.
Redact a PDF in the browser