PDF PII Redactor — actually remove personal data, not just cover it

Finds emails, phone numbers, Social Security numbers, credit card numbers, IBANs, IP addresses and dates of birth in a PDF, deletes them from the file, then draws the black box.

The difference that matters

Most tools that call themselves redactors draw a black rectangle over the text and save the file. The words are still there, in the content stream, underneath the box.

Select-all and copy gets them back. So does pdftotext. So does any PDF parser written in the last twenty years. Real documents have been leaked exactly this way — including court filings, published with "redacted" names that anyone could recover by dragging a cursor across them.

This Actor uses PyMuPDF's redaction annotations, which remove the glyphs from the page before the box is painted.

Then it checks. After saving, the output is re-opened and re-parsed. If any of the text we claimed to remove is still extractable, the document is marked failed and you are not charged for it.

What it finds

Type	Notes
Email addresses
Phone numbers	Requires separators or a country code, so bare invoice numbers survive
US Social Security numbers	NNN-NN-NNNN
Credit card numbers	Luhn-checked — real cards pass, order numbers almost never do
IBAN	International bank account numbers
IP addresses	IPv4, correctly bounded so version strings are not eaten
Dates of birth	MM/DD/YYYY
Anything else	Supply your own regex — employee IDs, case numbers, internal references

The patterns are deliberately conservative. A false positive silently destroys real content in someone's document, which is worse than a miss they can catch by reading it.

Pricing

Charged per page that actually had something redacted. A 60-page contract with personal data on 3 pages costs 3 pages, not 60. Pages we did not touch are free — which means the work of finding the matches happens before anything is billable.

Nothing is charged for a document that fails, or for one whose redaction could not be verified.

Output

Each document returns a signed link to the redacted PDF, the page counts, a breakdown of what was found by type, and the verification result.

What it is not

It does not do face blurring, image redaction, or handwriting. It does not find names or street addresses — those need a language model and produce false positives that quietly delete the wrong words. If you need those, this is the wrong tool and I would rather say so here than take the run.

Scanned documents with no text layer have nothing to search. Run them through Searchable PDF OCR first, then through this.