Redacting a PDF properly, start to finish

Get the file ready before you mark anything

Take an HR coordinator who has to send grievance-bundle-2026.pdf to an outside mediator by 5 pm. It is 31 pages: printed emails, a meeting summary and a two-page scanned witness statement. Three colleagues' names, one home address and a personal mobile number must not reach the mediator. Ten minutes of preparation saves a second round later.

• Finish every other change first. Combine the parts with Merge PDFs, reorder, compress or sign now, because redaction should be the last edit before sending.
• Save an untouched copy of the original somewhere separate. The redacted file cannot be reversed, and you will need the source if you missed something.
• If the PDF asks for a password to open, remove it with Unlock PDF first (only for files you are entitled to unlock). The redaction tool does not open protected files.
• Write down every string to remove, with its variants: "Petra Novak", "P. Novak", "Ms Novak", "Novak's", and the email address p.novak@ that contains the name.
• Note which pages are scans. Their words are pictures, so they will need hand-drawn boxes.

Redacting in PDFGym, step by step

1. Open Redact PDF and choose the prepared file. It is read inside this browser tab and never uploaded.
2. Type each string from your list into the find box and mark all matches. Repeat for every variant.
3. Page through the preview. Draw boxes over scanned text, signatures, photos and anything search missed; switch to text selection to mark a single block.
4. Optionally press Find suggestions to get a list of likely names, street addresses and identifiers, and keep only the ones that should go.
5. Choose the output: keep unmarked text selectable, or flatten the marked pages.
6. Press Redact PDF, read the verification and the before-and-after view of a marked page, then download the new file.

Picking a marking method

Mixing methods is normal. In the grievance bundle, search would catch the names in the emails, while the witness statement needs boxes.

Selectable or flattened: what the new file keeps

Every page with a mark is rendered fresh as a high-resolution image with the black areas painted in, so the original drawing instructions for that page are not carried over. In the default mode the tool then lays an invisible text layer over that image containing only the words outside your marks, which keeps search, copy and screen readers working for the rest of the page. Flattening skips that layer: nothing on a marked page can be selected at all. Pick flattening when the leftover words on those pages never need to be searched or copied, and the default for long reports and filings.

Pages without marks are copied across as they were. Some things are removed from the whole document, though: comments, links, form fields, bookmarks, attachments, document properties and XMP metadata. If the recipient needs a clickable table of contents or a fillable form, tell them it will not survive. Because marked pages become images, the file can also grow.

What gets tested before the download appears

After writing the new PDF, the tool opens it again and checks it. The page count and page sizes must match the original. Each marked area must render almost entirely black. No selectable character may overlap a mark, and the remaining text must be exactly what was expected. Removed fragments of four characters or more are searched for in the raw bytes, both as plain text and in hex encoding. Supported metadata and attachments must be gone. If any test fails, you get an error instead of a file.

Those tests describe this output. For an independent look, run the download through the redaction checker; the guide on reading text under black boxes explains what that report can and cannot rule out.

Slips that leave private details behind

Redacting first, merging later. Any edit after redaction produces a file you have not verified. Fix: redact last, and if you must change something, redo the redaction from the original.

Searching only the full name. Initials, titles, possessives and email addresses slip through. Fix: search each variant and read every page once at normal size.

Leaving the context. Blacking out a name beside "the only night-shift supervisor" still identifies her. Fix: mark the description too, or the job title.

Expecting suggestions on scans. Image-only pages have no selectable text, so the detector skips them. Fix: box them by hand.

A revealing file name. novak-complaint.pdf undoes careful work. Fix: rename the download before sending.

When to hand the job to something else

The boxes are plain black. If a court, a records office or a freedom-of-information process expects exemption codes printed on each box, a redaction log or numbered pages, use software built for that workflow or follow the office's own instructions. The same applies when your employer mandates approved redaction software for classified or regulated records.

Hundreds of files a week is a job for desktop or server software with batch processing, not one tab at a time. A scan with no text layer can be redacted here with boxes, but if you need to search it to find every mention, it has to go through OCR first. And for documents in languages other than English, do not lean on suggestions; use search and a careful read.