How to Redact a PDF: Modes, Limits, and Output Checks
Anonymize PDF finds selected types of personal data, paints black rectangles over matches, and builds a new PDF from the resulting page images. It accepts one PDF or a group of images. It does not provide a preview in which you approve or reject individual matches.
That distinction matters when sharing a contract, receipt, or identification document. Removing the original text layer helps prevent copying text from underneath a black rectangle, but it does not make an overlooked name or visible account number safe to share.
Choose what to detect
The default categories are email, phone, credit card, US Social Security number, IBAN, and IP address. The broader ID / passport / long numbers category is optional and starts unchecked. These are pattern detectors; they do not recognize every country’s identifiers or every possible way of writing an account number.
Use the custom-terms field for known details such as a person’s name, company name, or reference number. Separate terms with commas, for example Alex Chen, PROJECT-4821. Terms must contain at least two characters. Matching ignores letter case, but a misspelled name or an OCR error can still prevent a match.
Decide where document content is processed
There are two independent AI controls. Turning off AI deep detection does not turn off AI OCR.
| Setting | How it works | Document content sent to AI |
|---|---|---|
| OCR Off, AI deep detection off | Uses the PDF’s readable text layer and local pattern matching | None |
| Local OCR, AI deep detection off | Recognizes image text in the browser before matching | None |
| AI deep detection on | Requests additional terms such as names and addresses | Extracted text, currently up to the first 60,000 characters |
| AI OCR selected | Uses AI recognition on pages that need OCR | Images of those pages |
Local OCR downloads its engine and language data when needed; that download is different from uploading your document. The English page uses English OCR, while other supported site languages select their corresponding language plus English. AI detection can miss information too. Choose an AI mode only when you are comfortable sending the indicated document content to that service.
Run the tool
- Open Anonymize PDF and select one PDF, or select a set of supported images. If a selection includes PDFs, the tool uses the first PDF.
- Choose the detection categories and enter known sensitive words or identifiers as custom terms.
- For a text PDF, OCR can stay off. For scanned PDFs or pages containing image text, select On — local (private) if you want recognition on your device. Image-only input automatically selects local OCR when OCR was off.
- Set AI OCR and AI deep detection according to the data flows above. Check both controls before processing.
- Click Redact & download. The tool scans and rebuilds the document, then downloads
redacted.pdf. There is no intermediate match-review screen. - Open the downloaded file and compare it with the original using the checks below.
Check the output before sharing
Check every page, not just the places with obvious black rectangles. Read headers, footers, stamps, signatures, account details, and repeated identifiers. Enlarge each redaction and inspect its edges for partially visible characters. A name inside a logo or a sideways label can need extra attention because text extraction and OCR may not recognize it correctly.
For example, if your document contains both Alex Chen and A. Chen, check both forms in the output. Adding only the first spelling is not evidence that the second was removed.
Trying to copy or search text is useful for checking whether a text layer remains, but a failed search is not proof that all personal data was removed. A missed value may still be readable in the page image and recoverable with OCR. A “No matches found” message is also not a privacy check.
If something remains, add the missing term and rerun the original file. If recognition still misses it, use a tool with manual redaction controls and verify that result before sharing. Keep the original so you can correct over-redaction as well.
What changes in the finished PDF
All output pages are rasterized images. Original selectable text, interactive form fields, and clickable links are not preserved as those interactive objects. This also means the result may be larger, and fine text should be checked at the size recipients will use.
The output is intended as a reviewed sharing copy. Keep your searchable original separately, and use OCR PDF only if you later need a searchable copy; that separate tool uploads the PDF for processing.
FAQ
Can I review highlighted matches before downloading?
No. The current tool scans, applies black rectangles, rebuilds the pages as images, and downloads redacted.pdf in one run. Review the downloaded file against your original before sharing it.
Does anonymizing a PDF send document content to a server?
Local detection and local OCR process document content in your browser. AI deep detection sends extracted text, while AI OCR sends page images that need recognition. To keep document content local, leave AI deep detection off and choose OCR Off or local.
Does a PDF with no selectable text prove that redaction succeeded?
No. The output is image-based, so even missed personal data is no longer selectable. You must visually check every page for sensitive information that remains readable, including at the edges of black rectangles.