Blog ›

How to check whether a redacted PDF is really redacted (two-minute test)

30 September 2026 · 5 min read

A PDF with black bars across the names looks done. Whether it is done depends on something you can't see: whether the text is still in the file under the bars. It very often is. Here are three tests that take two minutes, the ways redaction fails, and how to do it so there is nothing to find.

Why black boxes fail

A PDF page is a list of drawing instructions: put this text here, this image there, this rectangle on top. When someone "redacts" by drawing a filled rectangle, the instruction for the text is still in the list; the rectangle is simply drawn afterwards. Anything that reads the instructions rather than looking at the picture, which is every text extractor, search engine, screen reader and AI tool, sees the text as if the box weren't there.

This is not theoretical. Court filings, government reports and corporate disclosures have leaked names, prices and whole paragraphs this way, usually discovered by a journalist pressing Ctrl+A. The same failure happens with text set in white, text hidden behind an image, and tracked changes left in a Word file that was exported to PDF.

Test 1: select and copy (30 seconds)

Open the PDF in any viewer. Press Ctrl+A (⌘A on Mac) to select all, copy, and paste into a plain text editor. Read the paste. If the redacted words appear, the redaction failed. Also drag-select directly over a black box: if a highlight appears inside the box, there is text there.

Test 2: search (20 seconds)

Press Ctrl+F and search for a word you know was redacted, or for a fragment such as the first three letters of a name or a digit sequence from a number. A hit means the text is present. Do this for at least three different redacted items; sometimes some were done properly and others were only covered.

Test 3: extract the text (one minute, the definitive one)

Viewers can hide things; an extractor cannot. On a Mac or Linux terminal: pdftotext file.pdf - | less. On Windows, open the PDF in Word (File → Open, Word converts it) or in Edge and use Read aloud: the voice reads exactly what is in the file, including anything under a box. If you use Python, pdfplumber or pypdf will show every text run. What comes out is what any recipient can get out.

Extra credit: check the metadata (File → Properties, or exiftool file.pdf). Author names, original filenames and comments survive redaction more often than the body text does. Also check for attachments and layers in the viewer's side panel.

What a passing result looks like

• Select-all copies nothing from the masked areas, and dragging over a box highlights nothing.
• Search finds none of the removed words or number fragments.
• Text extraction shows either no text at all in those spots, or the replacement label you chose ("[REDACTED]"), and nothing else.
• Metadata contains nothing you didn't intend to share.

How to redact so it passes

Two approaches work, and they are the only two:

1. Remove the content from the file. Acrobat Pro's Redact with Apply, macOS Preview's Redact tool, and a few other editors with a feature explicitly named "redact" delete the text and image data under the mark. The rest of the page stays live and searchable.
2. Flatten the page to an image. Render each page to a picture with the boxes painted on, then rebuild the PDF from those pictures. There is no text left anywhere to extract. The result is larger and no longer searchable, which is the trade-off for certainty. This is what PDF24 does and what the pdfagents.ai Redactor does.

Everything else, including drawing shapes, setting text to white, covering with an image, or "redaction" tools that only add an annotation, fails test 1. The steps for each tool are in how to redact a PDF.

A note on scans and photos

If the document is a scan, the text under the box is a picture, so the copy test passes trivially. But two other failures appear: OCR text layers (a scan that has been made searchable contains an invisible text layer, which extractors read), and low-opacity or misaligned boxes that leave letters partly visible. Test 3 catches the first; zooming to 400% catches the second.