Anyone who receives the file can get the name back in a few seconds: select the area and copy, run pdftotext on the file, or open it in an editor and move the rectangle. Search engines and AI tools that read PDFs extract the text layer and never look at the rectangle.

What real redaction changes in the file

A redaction that holds removes the glyphs from the content stream, removes or rewrites images that contain the same value, and then draws the black area. After that, the text layer has nothing to extract at that position. A quick test on any redacted PDF:

pdftotext redacted.pdf - | grep -i "rossi"

If the name comes back, the redaction is cosmetic.

The places people forget

• Document properties: author, title and subject often repeat a client name.
• Annotations and comments, which keep their author.
• Bookmarks and the outline, built from headings that may contain names.
• Embedded files and form field values.
• Scanned pages, where the name is in the image and a text search finds nothing until OCR runs.
• Earlier revisions kept by incremental saves, which a viewer does not show but a parser can read.

Word and Excel have their own versions of the same problem: tracked changes, comments, hidden sheets, cell notes and custom properties.

Masking for AI is a different job

When the file goes to a language model, deleting the names makes the text hard to reason about. Replacing each value with a numbered label (PERSON_1, IBAN_2) keeps the relationships readable, and a map kept on your side puts the real values back into the answer. Masked text is still personal data under the GDPR for whoever holds the map, so it suits analysis inside an organisation, while files that leave for good are blacked out.

Redactic does both for PDF, Word, Excel and scanned documents: it removes the values from the file, cleans comments, tracked changes, hidden sheets, document properties and embedded files, and shows the result for review before download. The feature list is on the Redactic site.