Why drawing a black box is not redaction
This is the single most consequential misunderstanding in document handling, and it has embarrassed governments, law firms and newspapers repeatedly. A PDF stores text as text and graphics as graphics, in separate layers. Drawing a black rectangle over a name adds a graphic on top; it does not remove the text underneath.
The result looks perfectly redacted on screen and is trivially defeated. Select the area and copy it, and the hidden text lands on your clipboard. Extract the document's text and it appears in full. Open it in almost any editor and the rectangle can simply be deleted. Anyone who wants the information can have it in seconds.
Real redaction removes the underlying content, not just the view of it. That is the difference between a document that is safe to publish and one that merely appears to be, and it is worth checking which kind your tools produce before you rely on either.
Checking your own redactions
There is a simple test that catches the overwhelming majority of failed redactions, and it takes about ten seconds. Open the finished document, select all the text on the redacted page, copy it, and paste it into a plain text editor. If the redacted words appear, the redaction is cosmetic and the document is not safe to send.
A second check is worth doing for anything that will be published: extract the document's text and read it. This catches content that is present in the file but not rendered visibly, which no amount of looking at the page will reveal.
Finally, remember that metadata sits outside the page content entirely. A properly redacted page can still ship with an author name, an organisation, a filename in the document title, or a revision history. Cleaning metadata is a separate step and it is easy to forget when you are focused on the visible text.
Redacting sensibly, not just thoroughly
Redact the smallest thing that achieves the goal. Blacking out entire pages when three lines are sensitive makes the document harder to use and invites the reader to assume the worst about what is missing. Precision reads as considered; wholesale removal reads as evasive.
Think about what remains as well as what is removed. Redacting a name but leaving a job title, a department and a date can identify someone as effectively as the name did — this is how many well-intentioned redactions fail. If the goal is that a person cannot be identified, look at the combination of details, not each detail in isolation.
And work from the original every time. Redacting a document, then later needing a different set of redactions, is a job for a fresh copy of the source. Redacting an already-redacted file compounds any error in the first pass and gives you no way to check what was originally there.
Redaction in practice: the cases that go wrong
The failures that make the news follow a small number of patterns. The most common is the black box over live text, which is a display change rather than a removal. The second is redacting the visible page while leaving the same information in an embedded attachment, a comment, or an earlier revision stored in the file.
A third is redacting a scanned image and forgetting the document has been through OCR, so an invisible text layer still sits beneath the picture containing every word on the page. The image looks redacted, the extractable text is not, and nothing about the document's appearance reveals the problem.
A fourth is more subtle: redacting consistently enough that the shape of what is missing gives it away. Redacted names of differing lengths, in a document with a known list of candidates, can be narrowed down by measurement alone. Where this matters, replacing content with a fixed-width marker is safer than blacking out the exact extent of the original.