Redacting a PDF Properly (a Black Rectangle Isn't Redaction)

A black box drawn over a paragraph in a PDF doesn't remove the text underneath it, and that mistake has leaked real documents before.

By PulsarKit · Published September 8, 2026

Why covering text is not the same as removing it

Drawing a black rectangle over a paragraph in a PDF editor changes one thing: what is visible on screen. Running the file through Flatten PDF afterwards does not help either: it bakes the rectangle into the page but leaves the text objects underneath it untouched. The words themselves stay where they were in the page's underlying content stream. Anyone who opens the file, selects that area and copies it gets the supposedly hidden text in their clipboard in about a second, with no special tools, just the same reader everyone already has. This isn't a hypothetical risk. Court filings, police reports and government contracts have leaked this way, publicly, more than once, because someone reasonably assumed a black box meant the information was gone.

The only approach that removes the text

Real redaction in a PDF means the page containing sensitive text is rebuilt as a raster image, a picture, with the redaction boxes filled onto those pixels before the image is created, not painted over an existing page afterwards. The new page carries a picture instead of the original text objects, so there is no text left in the file to select, copy or extract. The words aren't hidden; they are gone. A properly built redaction tool, and Redact PDF is built this way, re-reads its own output afterwards and refuses to hand back a file if any text is still detected on a page that was supposed to be cleaned. It checks its own work rather than assuming the process succeeded. That check is the difference between a tool you can trust with a court filing and one you have to verify by hand every time.

Only the pages you touch should change

Rasterising an entire document to hide two lines on one page would inflate the whole file several times over and destroy text search on every page that had nothing to hide. The pages you don't mark stay as they were, copied across byte for byte, still selectable and searchable and still their original small size. The trade-off is worth expecting rather than being surprised by: the redacted pages will get noticeably heavier, because a photograph of a page of text is a few hundred kilobytes where the text itself was a few kilobytes of instructions. That weight is the cost of the words really being gone.

A name hidden on one page can still sit in the document properties

Redacting the visible text solves half the problem. The title, author, subject, keywords and the software that produced the file all live in the document's metadata, separate from anything drawn on a page. A name blacked out on page four is worth nothing if that same name still appears the moment someone opens the file's properties panel. Metadata should be cleared as part of the same pass that removes the visible text, or straight afterwards with Remove PDF Metadata, not as an afterthought, and not left for whoever receives the file to discover on their own.

Finding every occurrence, not just the one you noticed

A name, an account number or a date format rarely appears only once in a real document. A search across the whole file, rather than a manual scan page by page, is what catches every instance instead of the two or three someone happened to spot while skimming. Searching by pattern rather than by exact phrase is what catches something like a social security number or a phone number in a consistent format wherever it appears, with every match listed by page so nothing gets redacted by accident that should have stayed visible.

Why a scanned page needs a different approach entirely

A search only works on real, selectable text, and a scanned document has none. It is a picture of a page from the start, so a name typed into a search box will find nothing on it no matter how many times that name appears in the image. The fix there is to select the areas by hand instead of searching for them. The result is identical either way, since the page gets rasterised in both paths, whether or not the original had selectable text.