What does AI Redact catch?

Names (people and organizations), emails, phone numbers (international formats), SSN patterns, credit card numbers (with checksum validation), addresses, dates of birth, IP addresses, and IBAN/account numbers. Coverage is broad but not exhaustive — manually verify for sensitive cases.

Can it redact images?

Yes — AI detects text inside images and offers to black it out. For photos containing PII (whiteboards, signed documents in photos), use the image-redaction toggle.

Is the redaction reversible?

No. That's the whole point. Save your unredacted master separately so you can re-redact differently later if needed.

Use case · 5 minutes

How to redact a PDF properly before sending it externally

Permanently remove names, salaries, account numbers, and PII — not 'cover with a black box'.

For: HR sharing offer letters as samples, lawyers preparing FOIA responses, anyone sharing internal docs externally

Start with step 1 Browse all tools

STEP BY STEP

The workflow

Auto-detect personally identifiable information

AI Redact scans the PDF for names, emails, phone numbers, SSN-shaped patterns, addresses, credit card numbers, and dates of birth. Each detection gets a confidence score; you accept or reject.

Open Redact Sensitive Info

Manually redact anything else

AI catches the obvious; you catch the contextual. Names of internal projects, codenames, vendor identifiers — search-and-redact each one across the whole document so you don't miss occurrences.

Open Redact PDF

Strip metadata in the same pass

Author names, edit history, original-file-path, and other metadata leak even when the visible page is clean. Toggle 'remove metadata' before exporting.

Open Redact PDF

WHY IT MATTERS

The case for doing this

Most 'redacted' documents in the wild aren't actually redacted — they're documents with black rectangles drawn on top. The text underneath is still readable by anyone who copy-pastes through the rectangle. Real redaction permanently removes the bytes. The difference is forensic: a real lawyer with a real PDF reader can recover non-redacted content from a fake redaction in under a minute. Don't be the source of that headline.

COMMON MISTAKES

What to avoid

Drawing a black rectangle instead of redacting

Annotation rectangles cover text visually but leave it intact in the file's data. Use the redact tool, not the highlight or shape tools.

Forgetting metadata

Names buried in 'Author' or 'Last Modified By' fields survive visual redaction. Always strip metadata in the same pass.

Trusting visual inspection on text-heavy redactions

Use search-and-redact for names and IDs. Manual scanning misses occurrences in headers, footers, and embedded annotations.

TIPS FOR THE BEST RESULTS

What works

Test by copy-paste before sending

Open the redacted file. Try to copy-paste from the redacted region. If you get the original text back, the redaction failed — don't ship.

Save with a clear filename

filename-redacted-2026-04-25.pdf signals 'this version is for sharing'. Keep the unredacted master under a different name.

Keep a redaction audit log

For regulated industries, log what was redacted and why. Our Redact tool exports a JSON log of every redaction with the page, region, and category.

RELATED WORKFLOWS

Other jobs people do with pdfcraft ai

How to redline a contract that came back from counterparty

Diff V1 vs V2 in 30 seconds. Material changes flagged, cosmetic ones filtered out.

How to OCR a folder of scanned PDFs so they become searchable

Make decades of scanned archives findable via Cmd+F, Spotlight, and Google Drive search.

How to extract every table from a financial PDF into one spreadsheet

10-K, 10-Q, annual report → clean Excel with one sheet per table, ready to chart.

FAQ

Questions about this workflow

Real redaction deletes the underlying text from the file. Black rectangles are annotations on top of text that's still there. Search 'BBC redaction failure' for examples of how often this gets shipped.

Try the workflow on a real PDF

Free tools run in your browser. No signup, no watermarks.

Start with step 1