Quick answer
If the original text object remains underneath, copying, searching, extraction, or removing the overlay can expose it. A proper workflow applies a redaction that removes the selected content, saves a new file, sanitizes other hidden information when appropriate, and independently checks the result.
No PDF upload
Check whether hidden text remains
Ready. Add optional phrases, then choose a PDF to inspect.
PDF redaction report
Per-page evidence
| Page | Extracted characters | Text items | Covered candidates | Redact annotations |
|---|
Covered extractable text candidates
| Page | Extracted text | Reason | Rendered evidence |
|---|
Known-phrase search
| Phrase | Occurrences | Pages | Result |
|---|
Review all text returned by extraction
Interpretation: A finding is evidence to investigate. No finding is not certification: image pixels, hidden layers, metadata, attachments, revisions, alternate files, unusual encodings, and parser limitations remain outside this result.
Three redaction states that look similar
Black-box overlay
An opaque shape is drawn after the text. The page looks covered, but the original characters remain in the content stream. This is a common failure that extraction can expose.
Marked but not applied
A PDF /Redact annotation identifies content intended for removal. Until supporting software applies it and saves the result, the marked content can remain present.
Applied content removal
The selected text or image data is removed and a replacement appearance is drawn. Other hidden information—metadata, comments, attachments, layers, scripts, and alternate copies—still needs separate review.
What a reliable verification workflow checks
| Check | What it can reveal | What it cannot prove |
|---|---|---|
| Extract all text | Words and values that remain in standard text objects, including under some overlays. | Image-only text, unusual encodings, or content a parser cannot decode. |
| Compare text positions with rendering | Extractable text whose bounding region renders as a mostly dark block. | Every design: legitimate white-on-black text can resemble a cover, while irregular covers can evade a threshold. |
| Inspect Redact annotations | Content marked for redaction that may still await application. | That an annotation-free PDF is clean; applied redactions normally remove the annotation. |
| Search known phrases | Exact names, identifiers, addresses, or values that still survive extraction. | Unknown sensitive content, spelling variants, images, or fragments. |
| Inspect hidden information | Metadata, comments, attachments, forms, links, scripts, layers, and other payloads. | Handled separately by the PDF privacy inspector; it also is not a universal security verdict. |
Official guidance: mark, apply, sanitize, verify
Adobe’s current redaction workflow distinguishes marking content from applying redactions, then offers sanitization for hidden information. Adobe’s sanitization guidance lists metadata, comments, hidden layers, and other information that may remain outside the visible page.
The PDF 1.7 specification defines Redact annotations as regions to be removed when redactions are applied. Their existence is therefore not equivalent to the underlying content already being gone. This checker treats the annotation as a review signal and separately examines what PDF.js can still extract.
Reproducible redaction fixtures
All identifiers are synthetic. The unsafe fixture draws ALPHA-ACCOUNT-7842 and then draws an opaque black rectangle over it. The safe control draws the same rectangle but never writes that marker. The pending fixture leaves PENDING-REDACTION-9917 visible beneath a Redact annotation that is deliberately not applied.
Unsafe black-box PDF · Content-removed control · Pending annotation PDF · Expected results and SHA-256
Unsafe Poppler extraction · Safe Poppler extraction · Pending Poppler extraction · Rendered image inventory
How this checker detects a likely black-box failure
- Render without annotations. Each page is rendered at 1.5× scale so an annotation appearance does not itself become the evidence.
- Extract positioned text. PDF.js returns each text item, transform, width, and page position.
- Sample the corresponding pixels. A candidate is reported when at least 72% of pixels in the text region are very dark and the mean luminance is at most 62, or when text overlaps a Redact annotation by at least 25%.
- Run exact phrase checks. Optional phrases are searched case-insensitively across the complete extracted text and reported by page.
- Publish raw evidence. JSON retains thresholds, boxes, pixel statistics, annotations, page text, and limitations; CSV provides a concise review list.
Limitations and responsible interpretation
This is a heuristic preflight, not a redaction certificate or forensic guarantee. Legitimate white text on a black background can be flagged. A cover that leaves much of a text item on a light background can escape the pixel threshold. Image-only pages, clipped or layered content, embedded files, prior revisions, malformed objects, unsupported encryption, viewer-specific behavior, and parser defects require other methods.
Do not paste extracted private text into a remote service merely to investigate it. If a consequential document fails any check—or if the stakes are legal, regulatory, journalistic, or personal—return to the original source, apply redaction using maintained software, save a separate output, sanitize hidden information, and have another qualified person or approved system verify the exact deliverable.
Frequently asked questions
Does drawing a black box over PDF text redact it?
Not necessarily. If the text object remains underneath, it may still be selectable, searchable, copyable, or recoverable by removing the overlay. Proper redaction removes the underlying content and then applies the replacement appearance.
How can I check whether a PDF is properly redacted?
Search known values, extract all text, compare text positions with rendered covers, confirm Redact annotations were applied, and inspect metadata, comments, attachments, layers, scripts, revisions, and alternate copies separately. No single browser check certifies safe redaction.
Does a Redact annotation mean content was removed?
No. It can be a pending mark. Supporting software must apply the redaction and save the resulting file. If the underlying text is still extractable, removal is not complete.
Can a properly redacted black box still trigger this checker?
A black rectangle with no text beneath it should not trigger the covered-text rule. However, legitimate white-on-black text, nearby text whose bounding box overlaps the cover, or unusual rendering can create a false positive. Review the extracted snippet and page context.