The problem with doing it by hand
Opening forty PDFs and copying addresses out of them is slow, and it is unreliable in a specific way: you find the ones that look like addresses in the body text and miss the ones in the footer, the ones inside a screenshot, the ones in an attachment to an archived email, and the ones in a column of a spreadsheet you scrolled past.
Copying from a PDF adds its own failures — line breaks landing in the middle of an address, ligatures mangling characters, two-column layouts interleaving text that was never adjacent.
What extraction gives you
Extraction mode runs the same detection engine as redaction, but instead of rewriting the file it returns the findings. Results are grouped by entity type and itemised per file, so on a batch you can see immediately which document contributed what.
The detectors available include email addresses, URLs, phone numbers, person names, credit-card numbers, IBANs and IP addresses. Card candidates are tested against a Luhn checksum and IBANs against their mod-97 check digits before being reported, and common company names are filtered out of person-name results — so you spend less time discarding false positives.
It reads what a copy-paste cannot
- Scans and screenshots. OCR pulls addresses out of images, including images embedded inside Office documents and PDFs.
- Mail archives. EML files are read across headers, body and attachment text.
- Spreadsheets. Including sheets and columns that are hidden in the workbook.
- Markup and code. HTML, XML, JSON, YAML, CSV, logs and source files.
Common uses
- Auditing a document set before it goes out. See exactly what a batch exposes, then run redaction on the same selection.
- Checking a link inventory. Every URL across a set of documents, in one list — useful for finding internal or expired links in material about to be published.
- Recovering contact details from your own archived correspondence or invoices.
- Data-protection review. Establishing which files in a store contain personal data at all, before deciding what to do about them.
Please use it on documents you have the right to process. Harvesting addresses for unsolicited mail is both a bad idea and, in most jurisdictions, unlawful.
How to do it
- Upload one file or a batch.
- Select the entity types you want. Enabling fewer types gives a cleaner list.
- Run in extract mode — nothing is modified and no redacted copy is produced.
- Review the results per file, then switch to redaction on the same selection if that is where you were heading.
Limits
There is a per-document cap on how many matches of each type are returned — 50 on the Free plan, higher on Pro — and DokMine tells you when a document hits it rather than truncating quietly. See plans and quotas for the numbers. Detection is not exhaustive: obfuscated addresses, text in poor-quality scans and unusual formats can all be missed.