Is the text really gone, or just hidden?

Gone. It is removed from the page content stream, not covered with a shape. You can confirm it yourself: export the file, reopen it here, and run the Find hidden text tool over it.

Email addresses, US Social Security numbers, payment card numbers, phone numbers and IBANs, or any exact phrase you type. Search terms are treated as literal text, so a name with a full stop in it does not behave like a wildcard.

Can I draw a box over an area instead?

Not yet. Pattern and phrase search cover the case that actually causes disclosures — the same identifier repeated across a long document — and freehand region selection is the next addition. The engine already accepts arbitrary rectangles.

Does it redact images or scanned pages?

It removes matched text. A scanned page has no text layer until it has been through OCR, so there is nothing to match on. Do not rely on this for a scan.

How many pages does it search?

The first two hundred, so a very long document does not lock up the browser while you are still typing. The panel says when the search was capped.

Redact a PDF properly — the text is removed, not covered

How it works

01Open the PDF.
02Choose a pattern — email addresses, card numbers, national IDs — or type an exact phrase.
03Review every match before you commit. Each one is listed with its page.
04Export. The matched content is destroyed and the result is checked.

Find every one, not just the ones you spotted

The reason redactions fail is rarely that someone used the wrong tool. It is that a four-hundred-page bundle contained the same phone number eleven times and somebody found ten. Searching by pattern is the part that actually prevents disclosure: choose "payment card numbers" and every one in the document is listed, with its page, before anything is removed.

Checked before it is flagged

A pattern that matches too much is as bad as one that matches too little, because a tool that removes the wrong things gets switched off. Card numbers are validated with the Luhn checksum, so invoice references of the same length are left alone. Phone numbers are filtered by digit count, so years and page numbers are not swept up.

The page stays a page

Many redaction tools flatten the whole page to an image so nothing can be recovered from it. That works, and it also destroys search, copy and screen-reader access for all the content that was never sensitive, while multiplying the file size. Here only the redacted content is removed. Everything else remains real text.

Verified, or you get nothing

After the content is removed the exported bytes are re-read and searched for everything that was supposed to be gone. If any of it is still there the export fails with an error naming what survived, rather than handing you a file you would reasonably assume was safe. That check cannot be turned off.

Questions

Is the text really gone, or just hidden?: Gone. It is removed from the page content stream, not covered with a shape. You can confirm it yourself: export the file, reopen it here, and run the Find hidden text tool over it.
What can it find?: Email addresses, US Social Security numbers, payment card numbers, phone numbers and IBANs, or any exact phrase you type. Search terms are treated as literal text, so a name with a full stop in it does not behave like a wildcard.
Can I draw a box over an area instead?: Not yet. Pattern and phrase search cover the case that actually causes disclosures — the same identifier repeated across a long document — and freehand region selection is the next addition. The engine already accepts arbitrary rectangles.
Does it redact images or scanned pages?: It removes matched text. A scanned page has no text layer until it has been through OCR, so there is nothing to match on. Do not rely on this for a scan.
How many pages does it search?: The first two hundred, so a very long document does not lock up the browser while you are still typing. The panel says when the search was capped.