PDF Tools

Redact a PDF so the text is really gone.

Black out names, numbers and whole pages in a PDF without uploading it. Every page is converted to a picture with the black painted in, so the saved file has no selectable text and no hidden text layer to recover, and the finished file is re-read to check.

What this redaction tool does

Most “redaction” fails because a black rectangle is drawn on top of the text and the text stays in the file underneath. Anyone can select it, copy it or delete the rectangle. This tool avoids that by never keeping the original page. For every page of the file it draws the page to a picture on your device, paints solid black over your boxes in that picture, and builds the new PDF from the pictures alone. Nothing sits behind the black.

How to use it

  1. Choose a PDF of up to 40 MB and 100 pages. A preview of page 1 appears.
  2. Drag boxes over what must go. Use Mark whole page for a page that should be blacked out entirely.
  3. Or type words or numbers, one per line, and press Find and mark matches to box every occurrence on every page. Step through the pages and check each one.
  4. Press Redact and build PDF. The tool reads the finished file back and only offers the download if the check passes.

What is removed and what is not

The file is read and rebuilt in your browser. The isolated tool frame loads no advertising or analytics scripts and is not allowed to open network connections.

Why a black box is not enough

A PDF page is a small program: a stream of drawing instructions that place text, lines and images. A rectangle added by a markup tool or a “highlighter set to black” is just one more instruction after the text. The text instructions are still in the file, and so is every character they carry. That is how journalists and lawyers have recovered names and figures from published documents that looked fully blacked out. Wikipedia’s article on redaction describes the failure plainly: text covered by overlaid graphics in a PDF remains in the file and can be uncovered by deleting the overlay.

The only reliable fixes are to delete the underlying instructions, or to replace the page with something that has no instructions to delete. Deleting instructions precisely means parsing every font encoding, every form XObject and every inline image, and a mistake leaves text behind. Replacing the page with a picture is blunt but easy to verify, which is why this tool does it.

How the removal works, step by step

The PDF is opened with Mozilla’s pdf.js renderer inside the tool frame. Every page is drawn at the resolution you chose (150 dpi by default) onto an off-screen canvas. Your boxes are then filled with pure black on that canvas, over the finished picture, so the pixels underneath are overwritten before the image is saved. The picture is encoded as JPEG or PNG, and pdf-lib builds a new PDF in which each page is exactly the original size in points with its picture drawn across it. No page, font, image or other object of the original file is copied; only the pictures go in.

Because the new file is made only of pictures, the old document information, outline, named destinations, embedded files, comments, links, form data, optional layers, structure tags and any earlier revisions kept by “incremental saves” never enter it. The page count and sizes come from what pdf.js showed you, so a file that two readers interpret differently is judged by what you saw. The file name is the one thing the tool cannot change: the download is named after your source file.

The check that runs before you can download

After building the file, the tool opens it again and extracts the text of every page. Every page must come back with no text at all; if even one character is found the tool refuses to save and tells you which page. For each word you typed, it also searches the raw bytes and every decompressed stream of the new file in several encodings (plain, UTF-16 in both byte orders, hexadecimal strings, and quoted strings with their escapes undone), ignoring upper and lower case unless you ticked match case, exactly as the search does. If the word is found anywhere, that counts as a failure.

Be clear about what this proves. It proves the new file has no text layer on any page and no copy of the words you typed, written out exactly. It is a second check on top of the design (nothing but pictures is saved), not the reason to trust it. The word search cannot see a word the PDF stores in separate pieces (for example “Jo” and “hn” in two text commands), so “not found” means “not found as exact text”; for a name that matters, also draw a box over it and look at every page. It cannot read a photo or scan: if a page is a picture of text, you must draw boxes over the picture by eye. And a picture that is not covered can be read by a person or by text-recognition software.

Worked example

Suppose a 12-page rental application shows a Social Insurance Number on pages 2 and 9 and a bank account number on page 5. You type both numbers on separate lines and press find. The tool boxes each match, including numbers that the PDF stored as several separate pieces of text that sit next to each other on one line. You step through the pages to check them. On page 9 the number sits inside a scanned signature block that has no text layer, so you draw a box over it by hand. You press redact. All twelve pages become 150 dpi pictures, with boxes painted over the numbers on pages 2, 5 and 9. The result reports that no page came back with a text layer and that neither number was found as exact text in the file. The file will be larger than before, because every page is now an image.

Common mistakes

  • Trusting the preview alone. Search finds text only where the PDF has a text layer. A scanned page has none, so a search returns nothing. Always page through the file before you send it.
  • Redacting one occurrence. A name often appears in headers, footers and metadata. Search for it, then also look at every page.
  • Forgetting the file name. The output keeps your file name plus -redacted. If the file name itself is sensitive, rename it before you share.
  • Boxes that are too tight. Descenders (g, y) and ascenders can poke out. Search boxes are positioned by estimate and add a margin of at least 2 points; for rotated or skewed text the whole piece of text is boxed. Check every match, and give hand-drawn boxes some room too.
  • Sending a lower dpi than you can read. At 100 dpi small print may be unreadable. Use 150 or 200 for forms.

What this tool does not do

  • It does not find sensitive data by itself. It will not guess that a number is a SIN or an address. You mark it or search for it.
  • It does not make a legal determination of what must be redacted for a court, an access-to-information request or a privacy law. That depends on your rules.
  • It does not read text out of images or scans (no OCR).
  • It does not edit the original file. Your source PDF is untouched; the download is a new file.
  • It does not hide anything you leave uncovered. Every page is saved as a picture of what it looks like, so uncovered content is still readable by people and by text-recognition software.
  • It cannot find a word the PDF splits across several internal text commands. The check says “not found as exact text” and can miss such a word.
  • It cannot open password-protected PDFs. Remove the password first with the remove password tool.

Limits

Up to 40 MB and 100 pages. Each page is limited to 32 megapixels at the resolution you choose, so a very large drawing sheet at 300 dpi may be refused; pick a lower resolution. Because every page becomes a picture, the saved file is usually much larger than the original: roughly 100 to 500 KB per page at 150 dpi as JPEG, several times more as PNG or at 300 dpi. The tool shows an estimate once you choose a file and refuses a result it expects to exceed about 250 MB. Pages are processed one at a time to keep memory use down, but a 100-page file at 300 dpi can still take a while on a phone.

Related tools

After redacting, clear leftover author fields with the PDF metadata remover, delete review comments with the comment remover, fix page rotation with the rotate tool, and lock the file with password protection. To shrink a file that grew because pages became images, use the PDF compressor.

Frequently asked questions

Is the text really removed, or just covered?

It is removed. Every page of the file is replaced by a picture with the black boxes painted into the pixels, and the saved file contains only those pictures: no text, fonts, comments, forms, layers or metadata from the original. The tool also re-reads the finished file to check that no page has a text layer before it lets you download. Anything you did not cover is still visible in the picture, and a person or text-recognition software can read it, so look at every page.

Is my PDF uploaded anywhere?

No. The file is opened, drawn, redacted and saved on your device. The tool runs in a sealed frame that has no advertising or analytics code and is blocked from making network requests, so you can turn off your connection after the page loads and it still works.

Why is the whole file no longer searchable?

Because every page is a picture. That is the point: a file with no text layer has no hidden text to recover. If you need a searchable copy, keep the original and share only the redacted file.

Can I redact only some pages and keep the rest as text?

No. Every page is converted to a picture, including pages you did not mark. Copying pages as text would carry their fonts, hidden layers and other internal data into the new file, and the claim that nothing from the original survives would no longer hold. You only need to mark the pages that contain what must go.

What does the check after redaction actually prove?

That no page of the new file has extractable text, and that the words you typed do not appear as exact text (any case, unless you ticked match case, and in plain, hex, UTF-16 and escaped-string forms) anywhere in the new file. It cannot prove anything about text inside a photo or scan, about words you did not type, or about a word the PDF stores in separate pieces.

Will it find text in a scanned PDF?

No. A scan is an image with no text layer, so search finds nothing. Draw boxes over the scan by hand. There is no OCR step.

What happens to comments, bookmarks, attachments and metadata?

They are not copied into the new file, because nothing but the page pictures is. That includes comments, links, form fields, optional layers, bookmarks, attachments, document title and author. The download file name is the one thing that is carried over, so rename it if it is sensitive.

How large will the redacted file be, and which resolution should I choose?

Each redacted page becomes an image, so expect roughly 100 to 500 KB per page at 150 dpi as JPEG. Choose 150 dpi for ordinary documents, 200 for small print and 100 only when size matters more than legibility. PNG is lossless but much larger.

Sources

Checked

  • Redaction (Wikipedia) — describes why text covered by overlaid graphics in PDF and word-processor files remains in the file (retrieved 2026-10)
  • Adobe PDF reference, PDF 32000-1 (ISO 32000-1) — a page is a content stream of drawing instructions, section 7.8; images are image XObjects, section 8.9 (retrieved 2026-10)
  • pdf.js (Mozilla) — the renderer and text extractor that draws pages and re-reads the finished file (retrieved 2026-10)
  • pdf-lib — the library that builds the new PDF from the page images (retrieved 2026-10)

Runs entirely in your browser. Nothing you choose is sent or stored.

Common mix-ups

  1. ISO 32000-1 (PDF 1.7) specification, Adobe-hosted copy

Checked against these sources on .

Advertisement
Advertisement