A black rectangle is not PDF redaction: a reproducible test

A PDF can look safely redacted while still containing the original email address, phone number, or identifier in its text layer.

That happens when someone draws a black rectangle over text instead of creating a copy where the content is no longer present. The visual result may look identical, but the risk is not: the covered text can sometimes still be selected, copied, searched, or extracted.

This post documents a small reproducible test using only synthetic values. It is not a claim that automatic detection makes every PDF safe. It is a checklist for deciding whether a downloaded file has actually stopped exposing the markers you intended to remove.

The test file

The three-page sample contains deliberately fake markers:

• [e-mail]
• [phone]
• 4111 1111 1111 1111
• 00000000T
• ES82 0000 0000 0000 0000 0000

Never use real personal data in a public demo. Keep the original protected and work on a copy.

Visual hiding vs. effective redaction

The difference is simple:

1. Visual hiding adds a shape over the existing page content.
2. Effective redaction produces a result in which the sensitive content is no longer exposed in the final document.

You can test the first case without any special tooling. Try selecting text under the black rectangle, copy it into an editor, or search for the original value. If the value remains available, the PDF was only covered.

For an additional independent check on macOS or Linux, extract the text layer from the file you plan to share:

pdftotext final-copy.pdf - | rg 'ana\.santos@example\.invalid|202-555-0147|4111 1111 1111 1111'

No output for these markers is one useful signal. It is not a substitute for reviewing every page, images, annotations, form fields, attachments, and metadata.

The browser-based workflow

For the test I used PriviTools PDF Auto-Redactor. The document is handled in the browser during the normal workflow; the tool makes two limitations explicit:

• pattern-based detection can miss data or flag harmless content;
• the downloaded copy should be reviewed before it is shared.

I loaded the synthetic file and enabled emails, phones, card numbers, IBANs, national IDs, and U.S. Social Security patterns. The scan identified ten findings across the sample.

The useful step is not pressing “download” immediately. Review the matching list first. A detector does not know the full business context of a document, and a false positive can hide information you needed to retain.

After review, the generated copy is rasterized, removing its original text layer. I opened that downloaded copy separately and checked the sample markers again. In this specific test, the markers above were no longer available as selectable or extracted text.

The checklist below is designed to be reproduced with any controlled test file.

Watch the recorded walkthrough, using the same synthetic test file:

A practical verification checklist

Before distributing a redacted PDF, verify the final downloaded file, not just the preview:

1. Search for each value you intended to remove in a second PDF viewer.
2. Try selecting and copying from the redacted area.
3. Extract its text layer and search for known markers.
4. Inspect metadata, comments, form values, embedded files, and links separately.
5. Review every rendered page, especially scans and image-only pages.
6. Keep the original securely stored; distribute only the checked copy.

Image-only PDFs need extra care. A text-layer search may return nothing even when a value is still visible as pixels. If OCR was involved, validate both the visual result and the OCR output.

The important limitation

Automated redaction is assistance, not a guarantee. Pattern variations, OCR mistakes, unusual fonts, complex forms, and values without a recognizable format can be missed. The safe workflow is:

detect → review → generate a new copy → verify the final file → share.

If you work with legal, regulated, or high-impact records, follow the review process required by your organization. The reproducible demonstration above verifies only the listed synthetic markers in one sample file.

For a fuller guide to the problem and the checks, see How to redact a PDF without leaving the hidden text accessible.