Draw boxes over sensitive text — the export destroys what's under them, it doesn't just cover it
| Method | Actually safe? | Why |
|---|---|---|
| Black box / shape drawn over text (Word, preview apps, most "edit PDF" tools) | No | The text object still exists underneath; select, copy, or extract text and the "redacted" words come back |
| Highlighter or white text | No | Same problem — formatting, not removal |
| Password or "no-copy" permissions flag | No | Encrypts or restricts access but removes nothing; any tool that ignores the flag reads everything |
| Delete-and-resave in a text editor | Usually not | Content streams are compressed and cross-linked; "deleted" objects can linger and be recovered |
| Print to paper and rescan | Yes | Genuinely destroys the digital text — at the cost of quality, effort, and ending up with a scan |
| Professional redaction (Acrobat's Redact tool, etc.) | Yes | Removes the underlying content objects and scrubs metadata server-side or in-app |
| This tool: rasterize and rebuild | Yes | Every page becomes an image with boxes burned in — the only method that's actually safe fully client-side |
A PDF page is a program that draws text, vectors, and images. Most "redaction" features just add another drawing on top — the instructions for the sensitive text are still right there in the file, which is why they surface again the moment anyone extracts text. The fix is to stop shipping the original page at all.
Open the file with pdf.js and draw boxes on rendered pages. Each box is stored in PDF page coordinates, so it stays anchored to the content no matter the zoom. On export, every page is re-rendered at your chosen DPI (150, 200, or 300), your boxes are filled solid in that raster, and pdf-lib assembles a brand-new PDF from those images. What was under a box never makes it into the output; what wasn't is still plainly readable, just no longer selectable or searchable, because the rebuilt file contains no text layer at all.
Say a US Letter page (612 × 792 points) shows a bank account number, and you drag a box over it at screen position 120, 300 pixels, 240 by 60 in size, while the viewer is rendering at 1.5× scale. The tool converts those corners into PDF page space — x from 80 to 240 points, y from 552 to 592 — and stores that. At export time the page rasterizes at 200 DPI to 1700 × 2200 pixels, the box lands on exactly the same content, and it's filled solid before the image is encoded. Scale the export to 300 DPI and the raster becomes 2550 × 3300 pixels — 8.4 megapixels per Letter page — with the box still locked to the same words.
Afterward, the export has no text layer: searching the document for the account number finds nothing, copying is impossible, and the original file's author, title, and producer metadata are absent because the output is a new document, not an edited copy of the old one.
Because the box is just a mark on top of the page. The words underneath are still in the file, one copy-and-paste or text-extract away — there have been real court and government filings where "redacted" names were recovered this way. Real redaction removes or destroys the underlying content; this tool rasterizes every page and rebuilds the PDF so nothing under a box survives.
Yes. The export contains no text objects at all: each page becomes an image with your boxes burned in, assembled into a fresh PDF. Covered pixels are filled with a solid color before the image is encoded, so the original glyphs under a box are never in the output. Verify it yourself — open the result and try to select or search the covered words; there is no text layer to find them in.
Yes. The output is a brand-new document assembled from page images, so the original's document properties (author, title, producer, revision history) are not carried over. Bookmarks, embedded attachments, comments, and form fields are dropped as well — anything you need to keep should be considered before you export.
The text layer, everywhere — not just under the boxes. Text that wasn't redacted is still perfectly readable on screen and in print, but it can no longer be selected, searched, or read aloud, and file size usually grows because pixels cost more than characters. That's the honest price of the only client-side method that's actually safe.
The method — destroying covered content — is the right one, and it's what professional tools do under the hood. But suitability for a specific filing or disclosure is your call to verify: check the output yourself before sharing, and follow the rules that apply to your situation. This tool is not legal advice.
No. The PDF is parsed with pdf.js and rebuilt with pdf-lib inside your browser, and it works offline once the page has loaded. No server, no copy kept, no trace.