RedactDrop › Guides › What “verified redaction” means

What “verified redaction” means — how RedactDrop checks its own file

Published 27 September 2026 · by Tekiba

Why check at all

Redaction fails quietly: a word can be drawn in pieces, stored twice or kept in an earlier version of the file while the page looks right. So after removing what it finds, RedactDrop reads the new file as if it had never seen it.

What the check reads

The check runs in your browser, on the new file, before anything is offered for download:

• The text as a PDF reader sees it. A second, separate copy of pdf.js — not the one that read your original — opens the new file and extracts the text of every page, the annotations, form fields, bookmarks, attachments and metadata.
• The text as RedactDrop's own reader sees it. RedactDrop interprets the drawing instructions of every page, form, annotation appearance and Type 3 glyph itself, so a word that pdf.js would read differently is still caught.
• The file's own strings and names. Every string, name and key of every object — including objects no page refers to — is searched for the words.
• The bytes. The file and its decoded streams are searched in several encodings, and in each font's own codes for the words. Image data such as JPEG is not searched this way.
• The structure. The file has to match what RedactDrop writes — only what is needed to draw the pages, plus any links or bookmarks you chose to keep — with no metadata, no annotations other than kept links, and no earlier versions.
• Under each box you drew: no character, no image sample that differs from the others there, and no shape inside the box.

In our own test suite, the output is also read by a third, independent text extractor, pdftotext from Xpdf, and every test file has to come out with zero matches. That extractor is part of how RedactDrop is tested, not part of the check your browser runs.

When RedactDrop refuses

If a page does not pass, the file is not written. The most common reasons:

• A shape under a drawn box that cannot be cut. A curve that runs from under the box to outside it, a pattern or soft-mask fill, a clipping path crossing the box: RedactDrop cannot remove just the part under the box and say for certain what remains, so it does not try.
• A word in a font's name. If a word you are removing is part of the name of a font the page needs to draw its text, removing it would break the page, so the page is refused.
• The check still finds something. Anything the removal missed and the check found stops the file.

Turning a page into an image

When the problems found are tied to particular pages, you can choose to turn those pages into images; all the pages that did not pass are converted together. RedactDrop draws each page, paints over the words and boxes on the pixels, puts the image in place of the page, and checks the whole file again. The other pages stay as they were. The option is not offered when a problem belongs to the whole file rather than a page, and it is refused when RedactDrop cannot tell where on a page the remaining words are, since it could not paint over them.

What it costs: an image page has no selectable or searchable text, and its pixels are not checked for words — the check can confirm the page has no text left in it, but not what the picture shows. So look at an image page before you share the file. RedactDrop does not make this choice for you; the default is to refuse.

Where the check stops

The check covers what RedactDrop can read. It does not read text inside images or scanned pages, because it does no OCR. It does not search image data such as JPEG for your words. The area a box clears around each character is worked out from the character's width plus a margin, not from its exact outline, so a character in an embedded font that reaches well beyond its width can be kept beside the box with a sliver under the black. And PDF is a large format: a file can be put together in ways nobody anticipated. That is why “Verified” is written as a count of what was found, and why the terms ask you to look at the result yourself, and at every page of anything highly confidential.

How to check a redacted PDF yourself

1. Search it in your viewer for each word you removed, and for parts of them: a surname alone, the last four digits of a number.
2. Select across the black areas, copy, and paste into a plain text editor.
3. Look at the document properties and at the attachments, bookmarks and comments panels.
4. Open it in a second viewer, or run a text extractor over it. With Xpdf installed:

pdftotext -layout redacted.pdf -

If any of these finds something, do not send the file.

Questions