How to redact a PDF

A cover is not a redaction

PDF pages are lists of drawing instructions. Adding a black rectangle puts one more instruction above the name, number or image you meant to remove. The covered material can remain in the content stream, text layer or an annotation below it. A person may still select it, search for it or extract it without removing the rectangle.

A real redaction makes the source content absent from the result. Redact PDF rebuilds each marked page from a fresh render and applies the black region to that new page. It does not copy the original marked page content into the output. That is the difference between using a PDF redaction tool and colouring over a page.

Finish the document before redacting it

Work on the final copy. Merging can reorder pages, compression can rebuild a document, and a later signature or stamp can write new objects and metadata. If any of those happens after redaction, the file you checked is not the file you send.

Keep an untouched source in a separate place. The redacted copy is for distribution; the source is for correction if you missed a region. Do not use the redacted file as the only record, because removal is meant to be irreversible in that copy. Rename the finished file too. A revealing file name sits outside the PDF and no redaction engine can change it.

Find page content and data outside the page

Read every page once and mark names, addresses, identifiers, signatures, faces, barcodes and any surrounding words that reveal the same fact. A label can be as sensitive as its value. Removing an account number while leaving “retirement account” beside the mark may still disclose more than you intend.

Then inspect what is not printed. A PDF can carry an author, title, creator, producer, creation date, modification date, XMP packet, attachments, annotation text, form values, outlines and open actions. Page redaction and PDF metadata removal are different jobs. The metadata guide covers the second pass.

Mark selectable text and scanned text differently

Selectable text gives you three useful shortcuts. Select the whole text block, search for a repeated value across pages, or let the in-tab detector suggest people, street addresses and validated identifiers. Suggestions are a review queue, not an all-clear. They can miss private information, so redacting a PDF still ends with a page-by-page read.

A scan is a picture. Searching for text will not find words that exist only as pixels, and a text detector cannot mark them. Draw a box around the printed area instead. Include the full height of letters and a small margin on every side. If the same scan also has an OCR layer, the mark must cover both what you see and the text coordinates below it.

Decide what should remain selectable

The default mode keeps unmarked text selectable. It rebuilds the marked page as a render, then adds back a text layer containing only characters outside the redaction marks. This is the practical choice for reports, filings and long documents that still need search and copy after private lines are removed.

The strict mode flattens every marked page and leaves no selectable text on that page. It is the narrower safety choice when nobody needs to search the surviving text. Unmarked pages remain unchanged in either mode. The trade is explicit because turning a whole document into images would break accessibility and make clean text harder to use.

Check the file rather than the preview

A page preview proves only that black pixels are visible. The output needs several separate checks: each marked region renders black, no selectable text overlaps a mark, removed text fragments are absent from the raw PDF bytes, and supported private metadata is gone. The tool reopens the output and runs those checks before it shows a download.

Open the downloaded file again in a different reader. Search for every exact name and number you removed, try to select through each mark, and inspect the document properties. For a second view, Redaction check looks for supported opaque cover boxes with live text underneath and reports the text it can recover.

Know what the checker cannot promise

No automatic scan sees every possible leak. Text burned into an image has no selectable layer to compare. A cover made from an unusual clipping path or many small shapes may not look like one rectangle to the parser. A hidden fact can also survive in a filename, a previous email attachment or a cloud version history outside the PDF.

That is why the last pass belongs to a person who knows what the document contains. Review every page at normal size and at high zoom. Read the checker's limits. Send only the copy you inspected. The recovery guide explains how fake redactions, crops and hidden layers differ from content that was actually removed.

A short final checklist

• Finish every edit before the privacy pass and keep the original separate.
• Mark printed text, scanned pixels, repeated values, labels and identifying images.
• Remove metadata and attachments, then rename the file outside the PDF.
• Search, select, inspect properties and run the redaction checker on the downloaded copy.
• Send the exact file you checked, not an earlier export from the same folder.

If you need to redact information in a PDF today, use the tool first and this list as the review pass. If you need to redact text in a PDF that is only a scan, use drawn boxes rather than search. If the job is described as “black out a PDF,” the same rule holds: the content has to be destroyed, not hidden behind black ink.