Redacting a PDF by hand is painstaking work: rereading every page, spotting each IBAN, each phone number, each email address by eye — and hoping none was missed. The promise of automatic detection is to turn that chore into a simple verification. Here is how it actually works in Document Flare’s redaction tool, and — just as important — what it does not claim to do.
What the detection finds: structured identifiers
The detection relies on patterns and checksums, not on “intelligent” recognition. Nine families of data are spotted when the document loads:
- IBANs — international format + mod-97 check digit: a randomly composed fake number does not pass;
- French social security numbers (NIR) — 13 digits + key, canonical or compact format;
- Bank cards — 16-digit sequences validated by the Luhn algorithm;
- SIREN and SIRET — 9 and 14 digits, spaced or not, spotted with their context;
- Email addresses;
- Phone numbers — French and international formats;
- Birth dates — spotted by context (“born on…”, “date of birth…”);
- IPv4 addresses.
The checksum is what separates serious detection from a plain regular expression: for IBANs, cards and NIR, the tool verifies the mathematical consistency of the number. This drastically cuts false positives — a serial number that looks like a bank card but fails Luhn is not offered for redaction.
Fellowes 6M — P-4 micro-cutP-4 micro-cut: right for ID documents and financial papers.affiliate
What the detection does not find — and why that is deliberate
Personal names, postal addresses, case references, context-specific confidential mentions: none of this data is structured, and none has a checksum. A detector that spotted them would be probabilistic — with missed names (false negatives) and zones redacted for no reason (false positives).
That is why the tool offers a keyword mode instead: you enter the terms to target — a name, an address, an internal reference — and every occurrence found in the text is offered for redaction. The selection responsibility stays with you; the tool guarantees that everything you target is found.
The honest limit to keep in mind: automatic detection is a review accelerator, not a substitute. The redacted document always deserves a human pass before being sent.
What happens after the selection
Once the zones are chosen, the redaction itself is destructive: black rectangles are drawn, then each affected page is rasterized — converted into an image — before being reinserted into a rebuilt PDF. The text under the zones no longer exists in the produced file: no copy-paste, no extraction, no leaking through glyph positions.
The details of that method — and why security research recommends it over reversible redactions — are in our article on the science of PDF redaction. For the step-by-step walkthrough, see redacting personal data in a PDF.
Why all of this runs in your browser
Redacting a document means handling precisely the files you do not want to entrust to anyone. Routing a contract or a medical file through a third-party server to redact it would be self-defeating: the upload itself creates a copy on someone else’s machine.
Here, text extraction, pattern matching, checksum verification and rasterization all run entirely on your device. You can verify it: open the Network tab of the developer tools during the operation — no request carries the document. The verification method is detailed in our local processing guide.
The complete chain before sharing
Automatic detection is the first step of a protection chain:
- Detect — let the scan spot the structured identifiers;
- Complete — add your keywords for names and internal references;
- Redact — the zones are rasterized, the removal is permanent;
- Clean — erase metadata and attachments with the cleanup tool;
- Verify — reopen the produced PDF and try copying under the rectangles.
That last gesture is the only test that matters: nothing should be extractable under the black zones.
Fellowes 6M — P-4 micro-cutP-4 micro-cut: right for ID documents and financial papers.affiliate
Fellowes 60Cs — 10-sheet shredderTen sheets per pass and a 22 L bin: sized for regular office use.affiliate
