How to Redact Sensitive Info from a PDF (Properly)

In 2011, the Transportation Security Administration released a redacted PDF of its airport screening procedures. The redaction method? Black rectangles drawn over sensitive text in a graphics program. The problem? The text underneath was still there. Anyone could copy-paste the "redacted" content right out of the document. The entire screening protocol was exposed.

This isn't ancient history. It keeps happening. The DOJ has done it. Law firms do it regularly. Corporations do it in SEC filings. In 2024, a court filing in a major antitrust case had "redacted" revenue figures that were trivially extractable. The attorneys used black highlight over the text — cosmetic, not structural.

If you need to redact a PDF, you need to do it properly. A black box that doesn't remove the underlying text isn't redaction. It's decoration.

Why Black Rectangles Don't Work

A PDF is not a flat image. It's a structured document with layers. Text exists as a separate data stream from the visual elements. When you draw a black rectangle over text using a PDF editor, annotation tool, or drawing program, you're adding a visual element on top of the text layer. The text is still there.

Here's what someone can do with a fake-redacted PDF:

• Copy and paste: Select all text (Ctrl+A), paste into Notepad. Every "redacted" word appears in plain text.
• Search: Ctrl+F finds text behind black rectangles instantly.
• Screen readers: Accessibility tools read the underlying text out loud, ignoring visual overlays.
• Text extraction: Any PDF-to-text tool strips out the visual layer and exports raw text, redaction boxes included — because they're not redaction.
• Reorder layers: Open in a PDF editor, send the black rectangle to the back, and the text is visible again.

This isn't hacking. These are basic operations any computer user can perform. If your redaction can be defeated by Ctrl+C, it's not redaction.

What Proper Redaction Actually Does

Genuine redaction permanently removes the text and image data from the PDF's internal structure. The content is deleted — not covered, not hidden, not moved to another layer. Deleted. The area where the content existed is replaced with a solid fill (usually black, but sometimes white or gray). There's nothing underneath to recover.

After proper redaction:

• Copy-paste returns nothing from the redacted area
• Search finds no hidden text
• The PDF's internal byte stream no longer contains the removed characters
• No tool, forensic or otherwise, can recover the original content

This is a destructive, one-way operation. That's the point.

What Needs to Be Redacted

Most people think about the obvious items — SSNs and credit card numbers. But PII (personally identifiable information) shows up in places you don't expect.

The obvious:

• Social Security numbers (full or partial — even last four digits combined with other info)
• Credit card and bank account numbers
• Driver's license and passport numbers
• Dates of birth
• Home addresses

The overlooked:

• Email addresses (especially personal emails in business documents)
• Phone numbers
• Salary figures and financial terms
• Medical diagnosis codes and treatment details
• Student records (grades, disciplinary actions, accommodations)
• IP addresses and device identifiers
• Biometric data references

The invisible (metadata):

• Author name in document properties
• Creation and modification timestamps
• Software version used to create the document
• GPS coordinates embedded in inserted photos
• Tracked changes and revision history
• Comments and annotations (even deleted ones — some remain in the file structure)
• Embedded fonts that reveal the original software environment

That last category is what most people miss entirely. You can redact every sensitive word in the visible document and still leak information through metadata. I'll cover metadata cleaning below.

Manual Redaction: Step by Step

If you're redacting a small number of items and you know exactly where they are, manual redaction works.

Using WhiskFile's Edit PDF

1. Open your document in Edit PDF
2. Identify each piece of sensitive information
3. Use the rectangle tool to draw a filled black rectangle over the text
4. Critical step: Flatten the PDF after adding all rectangles

Flattening merges all annotation layers into the base content, making the rectangles permanent. Without flattening, your "redaction" is just a removable annotation.

The catch with manual redaction: It's slow, error-prone, and you will miss things. In a 40-page contract, sensitive information appears in headers, footers, signature blocks, exhibit references, and cross-references. A SSN mentioned on page 3 might be referenced again on page 27 as "SSN ending in 4589." You have to find every instance.

AI-Powered Redaction: The Better Approach

For documents with more than a handful of sensitive items — or when you're not sure what sensitive data exists in the document — AI-powered detection is significantly more reliable than human scanning.

AI PII Redact scans your entire document and automatically identifies:

• Social Security numbers (all common formats: XXX-XX-XXXX, XXXXXXXXX, XXX XX XXXX)
• Email addresses
• Phone numbers (US and international formats)
• Physical addresses
• Names (with context — it distinguishes between a person's name and a company name)
• Financial account numbers
• Dates of birth
• Medical record numbers

I tested it on a 12-page mock employee onboarding packet that I seeded with 34 pieces of PII across various locations — inline text, table cells, headers, footers, and form fields.

| PII Type | Instances | Detected | Missed | |---|---|---|---| | SSN | 4 | 4 | 0 | | Email | 7 | 7 | 0 | | Phone | 5 | 5 | 0 | | Address | 6 | 5 | 1 (partial address in a paragraph) | | Full name | 8 | 7 | 1 (name used as a file reference) | | DOB | 3 | 3 | 0 | | Account number | 1 | 1 | 0 |

33 out of 34 detected. The missed address was a city and state without a street number, embedded in a sentence — "grew up in Springfield, Illinois" — which is arguably not PII in isolation. The missed name was used as a filename reference ("see Smith_report.docx"), which is a genuinely tricky edge case.

The tool flags each detected item and lets you review before applying redaction. You can approve, reject, or add items it missed. This review step matters — no automated system should redact without human confirmation.

Metadata: The Redaction Everyone Forgets

You've redacted every SSN, every name, every phone number in the visible content. You export the PDF and send it. The recipient opens document properties and sees:

• Author: Janet M. Richardson
• Created: 2026-03-14 at 9:47 AM
• Modified: 2026-09-20 at 2:13 PM
• Producer: Microsoft Word 2024
• Keywords: confidential, HR, termination

The author's full name. The exact timeline of document creation. Descriptive keywords. All sitting in the metadata, completely unredacted.

Before distributing any redacted document, clean the metadata. Remove the author, clear the title and keywords, strip the creation date if it's sensitive. Some tools call this "sanitizing" or "scrubbing" the document.

Also check for:

• Embedded thumbnails — older PDFs sometimes contain thumbnail previews that show the unredacted version
• Incremental saves — PDFs that were edited and saved multiple times may contain previous versions in the file structure
• XMP metadata — an XML-based metadata standard that duplicates (and sometimes extends) standard PDF metadata

Legal Requirements for Redaction

Different contexts have different redaction obligations:

FOIA (Freedom of Information Act): Federal agencies must redact exempt information (national security, personal privacy, trade secrets) before releasing documents. Improper redaction here has led to leaked classified information.

HIPAA: Medical records shared for research, legal proceedings, or between covered entities must have PHI (Protected Health Information) redacted. The Safe Harbor method requires removing 18 specific identifier types.

Court filings: Federal courts require redaction of SSNs (last four only), birth dates (year only), financial account numbers (last four only), and minors' names in public filings. Many state courts have similar rules.

GDPR: Requires "pseudonymization" or "anonymization" of personal data when sharing for secondary purposes. Redaction is one acceptable method.

Messing up redaction in any of these contexts isn't just embarrassing — it can result in sanctions, fines, or malpractice claims.

The Redaction Checklist

Before sending any redacted document, verify:

1. Text removal: Select all (Ctrl+A), copy, paste into a text editor. Search for any sensitive terms. If you find them, your redaction failed.
2. Search test: Use Ctrl+F to search for known sensitive values (specific SSN digits, names, phone numbers).
3. Metadata cleaned: Check Document Properties for author, title, keywords, creation date.
4. Flattened: Open in a PDF editor and try to select or move the redaction rectangles. If they're selectable, the PDF isn't flattened.
5. No hidden layers: Check for hidden layers, optional content groups, or annotations that might contain original text.
6. Review a printed copy: Sometimes things are visible in print that aren't obvious on screen, and vice versa.

This takes five minutes. A redaction failure takes months to recover from — if recovery is possible at all.

When to Use Each Method

Manual redaction (Edit PDF + flatten): Small documents, fewer than 10 items to redact, you know exactly where everything is. Suitable for a quick one-off.

AI-powered redaction (AI PII Redact): Anything longer than a few pages, documents where you're not certain what PII exists, batch processing, legal or compliance contexts where thoroughness matters.

Password protection (Protect PDF): Not a substitute for redaction, but useful as an additional layer when sharing redacted documents with a limited audience. Redact first, then protect.

The tools are available. The knowledge is straightforward. The only thing standing between you and a properly redacted document is taking the extra two minutes to do it right. Given the alternative — your client's SSN sitting in a court's public filing system, extractable by anyone with a mouse — those two minutes are worth it.