Why a black box on a PDF hides nothing

A PDF page is two things at once

A PDF page holds a list of drawing instructions: place this font at this size, draw these characters at these coordinates, fill this rectangle with this colour. The instructions are executed in order, and the last one wins on screen.

That is all a black box is — one more instruction, appended after the text. It changes which pixels you see. It changes nothing about the instructions that came before it, and those still spell out every character underneath. Selecting the area and copying it returns the text; so does any tool that reads a PDF's text layer, which includes every search engine, every e-discovery platform and the command-line utilities that ship with most operating systems.

This is not an exotic failure. It has put unredacted names into court filings, settlement figures into public dockets and personal data into published government documents — in each case from a file that looked correctly redacted to everyone who handled it.

The tools that do this by default

Anything whose job is annotation rather than redaction: the highlighter and shape tools in a PDF reader, drawing a filled rectangle in a graphics editor, adding a black-filled box in a word processor before exporting to PDF. None of them claims to remove text. They are drawing tools, and they draw.

Printing to PDF does not fix it either, unless the printing path rasterises the page — many keep the text layer intact precisely because keeping text searchable is considered a feature.

How to check a file you already sent

Two minutes, no tools:

1. Open the PDF in any reader.
2. Select the whole page — Ctrl+A or ⌘+A — and copy.
3. Paste into a plain text editor.
4. Search that text for what you thought you had removed.

If it is there, it is there for everyone. Note the reverse is not a clean bill of health: text can also survive in places a copy-paste does not reach, such as document metadata, file attachments, or an unflattened form field's stored value.

What actually removes text

Only one thing: the characters have to stop being in the file. In practice that means either editing the page's instructions to delete them, or converting the page to an image and painting over the image — after which the page has no text layer left to recover anything from.

The second approach is blunter and safer, and it is what Caviardeur does: a page carrying a redaction is rasterised, the zones are painted onto the bitmap, and the bitmap replaces the page. Pages you did not touch keep their text. Before the file is handed over, it is re-read and searched for every string that was supposed to be gone; if any of it is still extractable, the export is refused rather than delivered.

What still needs your judgement

A scanned document has no text layer to detect anything in — nothing is found automatically, and the passages to remove have to be drawn by hand. Text baked into a photograph or a signature image is invisible to any text-based tool, yours included. And a rectangle drawn in the wrong place is still a rectangle drawn in the wrong place: no automatic check can tell that the box you placed covers the name you meant.

The short version

A black box is a picture of a redaction. If the text can still be selected, it was never removed — and the only way to know is to look, on the file you are about to send.