PDF Redaction Checker
Drawing a black box over text is formatting, not redaction — the words are still in the file, one select-all away. Before a redacted document goes out, drop it here and see exactly what a text extractor would still get back.
🔒 Your document never leaves your browser. For a file sensitive enough to redact, local checking is not a convenience — it is the only trustworthy form.
Drop the redacted PDF here
One PDF at a time · nothing is uploaded
The mistake that keeps making the news
Court filings, government releases, corporate disclosures — the pattern repeats: someone drew rectangles over the sensitive parts, the file went out, and a reader pressed Ctrl-A Ctrl-C. This tool looks for that exact failure: dark filled rectangles in the page content with extractable text underneath, and shows the recovered text verbatim, because a count is easy to dismiss and your own words staring back at you are not.
Redaction leaks live in more places than the page
The document information dictionary keeps the author's name no matter what happens on the pages. Comments and sticky notes carry their own text. Bookmarks name the sections they point to. Attachments are whole separate files. And a PDF saved incrementally still contains the earlier versions of itself — content "deleted" in the current version may be intact one generation down. Each of these is a separate check here, each with its own rule ID.
What this tool can and cannot judge
It judges the text layer. A scanned page is an image: there is no text to recover, and whether a box on a scan truly hides the pixels underneath is a question about the image, not the file structure — stated honestly as its own finding rather than silently passed. Encrypted files need unlocking first (also local). And if the check finds real problems, this tool does not fix them: re-redact with software that applies redactions, or remove the affected pages entirely, then check again.
Every check this tool runs (8 rules)
RED-001 Text is still there under a dark rectangle
A dark filled rectangle covers text that is still present in the content stream. It looks redacted on screen, but select-all, copy, or any text extractor gets the original back. This is the mistake that keeps making the news.
Fix: Redact with a tool that APPLIES redactions (removes the text), then run this check again. Drawing a box over text is formatting, not redaction.
Source: yaktool engineering rule built on the PDF 32000 file structure; thresholds (dark = grey ≤ 0.25, covered = ≥ 60% of the word box) are our documented reading of "looks redacted".
RED-002 Redaction marked but never applied
The file carries /Redact annotations — the markup a redaction tool places BEFORE the "apply" step. The content underneath is fully intact; the marks even tell a reader exactly where to look.
Fix: Open the file in the tool that made the marks and run its "apply redactions" step, then re-check.
Source: yaktool engineering rule built on the PDF 32000 file structure; thresholds (dark = grey ≤ 0.25, covered = ≥ 60% of the word box) are our documented reading of "looks redacted".
RED-003 Annotations carry their own text
Comments, sticky notes and highlight annotations store text outside the page content. Redacting the page does not touch them — the annotation text travels with the file.
Fix: Delete or flatten annotations before publishing, then re-check.
Source: yaktool engineering rule built on the PDF 32000 file structure; thresholds (dark = grey ≤ 0.25, covered = ≥ 60% of the word box) are our documented reading of "looks redacted".
RED-004 Document metadata survives redaction
Title, Author, Creator and Producer live in the document information dictionary, not on any page. A file with the author's name in its metadata leaks it no matter how thoroughly the pages were redacted.
Fix: Clear the document properties (and XMP metadata) before publishing.
Source: yaktool engineering rule built on the PDF 32000 file structure; thresholds (dark = grey ≤ 0.25, covered = ≥ 60% of the word box) are our documented reading of "looks redacted".
RED-005 The file carries previous versions of itself
More than one startxref marker means the file was saved incrementally: earlier generations of its objects are still inside. Content "deleted" in the latest version may be intact in an earlier one.
Fix: Re-save with a full rewrite ("save as" in most tools, or a qpdf/print-to-PDF pass), which drops the history.
Source: yaktool engineering rule built on the PDF 32000 file structure; thresholds (dark = grey ≤ 0.25, covered = ≥ 60% of the word box) are our documented reading of "looks redacted".
RED-006 Embedded files ride along
The PDF carries attached files. They are separate documents — nothing done to the pages touches their content.
Fix: Remove attachments before publishing unless they are meant to be there.
Source: yaktool engineering rule built on the PDF 32000 file structure; thresholds (dark = grey ≤ 0.25, covered = ≥ 60% of the word box) are our documented reading of "looks redacted".
RED-007 Bookmark titles are outside the pages
Outline (bookmark) titles are document-level strings. If a bookmark names the thing you redacted, the name is still there.
Source: yaktool engineering rule built on the PDF 32000 file structure; thresholds (dark = grey ≤ 0.25, covered = ≥ 60% of the word box) are our documented reading of "looks redacted".
RED-008 Scanned pages: nothing to extract, nothing to prove
Pages with no text layer are images. There is no text to recover here — but this tool judges the text layer only, not the pixels: content under a box on a scan may still be readable to eyes or OCR if the box is semi-transparent or misplaced.
Source: yaktool engineering rule built on the PDF 32000 file structure; thresholds (dark = grey ≤ 0.25, covered = ≥ 60% of the word box) are our documented reading of "looks redacted".
Thresholds are published, not hidden: a rectangle counts as "dark" at grey ≤ 0.25, and a word counts as "covered" when the rectangle overlaps ≥ 60% of its box. Fill paths whose geometry cannot be parsed are counted and reported, never silently skipped.