How to redact a PDF in your browser (AI-detected PII, permanently removed)

Redact finds PII in a PDF with AI, lets you choose what to cover, and permanently removes it — all in your browser. The PDF file itself never leaves your device; only its text content is sent out for detection.

Before you start

• What goes over the network: to detect PII, the text spans extracted from your PDF are sent to a third-party AI provider — text only, never the file. A badge next to the file name shows exactly what was sent (for example "214 text spans sent to AI").
• AI allowance: each scan counts as one AI action against your monthly allowance (25 per month on Free, 200 on Pro, 500 on Max — the cap applies even during the private beta). A small meter in the toolbar shows what's left. Without an account there's also a tight hourly scan limit (8 per hour by default), so sign in if you're redacting several documents.
• File size: the drop zone caps files at 100 MB. Split bigger files with Pages first.
• Passwords: a PDF locked only with a permissions (owner) password is unlocked on-device automatically — you'll see a heads-up that the redacted copy won't be password-protected, so re-protect it afterward if you need that. A PDF that needs a password to open must be unlocked first.

Steps

1. Open Redact and drop your PDF, or click Try with a sample to test on a sample full of fictitious PII.
2. If a banner says the PDF has filled form fields, click Flatten form fields. Typed field values live in form widgets, not the page text, so the scan can't see them until the form is flattened — Redact blocks the download until you do.
3. Click Auto-find PII. Text is extracted and sent to the AI; after a few seconds you'll see something like "Detected 12 PII spans across 4 categories."
4. If the PDF has no embedded text, you'll see "This PDF has no embedded text — it looks like a scan. Run OCR to extract text first." Pick the document's language from the dropdown (24 languages, English by default) and click Scan with OCR (slow). OCR runs page by page inside your browser with a live progress count; Cancel OCR stops it.
5. Review the Detected PII panel. Each detected category has a checkbox and a count (the possible categories are SSN, Address, Phone, Email, Name, Date of birth, Account number, License, Passport, and Other PII). Everything detected starts checked; untick what you want to keep. The page preview shades exactly what will be covered.
6. Pick a Redaction style:

1. Leave Forensic scrub (recommended) checked. The style only controls how the mask looks; scrub controls whether the text underneath is actually removed. With scrub on, each redacted page is flattened to an image right in your browser, so the covered text is permanently gone. With it off, a black bar only covers the text — the original characters stay in the file and remain selectable — and the stub styles are locked to Black bar for exactly that reason.
2. Click Apply redactions and download.

Result

You get a copy named after the original with a -redacted suffix (for example return-redacted.pdf). Besides covering the selected text, Redact strips PDF metadata (author, title, producer) and removes annotations — links, comments, sticky notes — on any redacted page, so PII hiding in them can't leak. Scrubbed pages lose their selectable text; re-add a text layer with Searchable if you need it. Pages without redactions stay untouched, and your original file is never modified.

After the download, a card offers Sign this PDF →, which opens the redacted copy straight in Sign — handed over in memory, no re-upload, still never leaving your browser.

Save a scan, or redact many files

• Save draft encrypts the PDF plus the OCR and AI results to this browser only, so you can resume later without re-running (or re-paying for) the scan. Drafts appear under "Resume a draft" on the tool's start screen.
• Bulk redact takes many PDFs at once: AI flags every PII category in each file, redaction (and optional forensic scrub) is applied, and you download a ZIP of the redacted copies. Files with filled form fields are skipped — flatten and redact those individually.

Related

• Redact troubleshooting — encrypted files, scans with no text layer, masks that stay selectable
• How to password-protect a PDF — re-protect the redacted copy
• How to make a scanned PDF searchable — re-add a text layer to scrubbed pages