What PDF Redact does not do
PDF Redact is a new tool built for the PDFs people actually share, not for every file ever written. This page lists what it leaves out, what it removes whole to stay safe, and what to do instead. We keep it current.
Last reviewed .
Most of these are deliberate
Every PDF feature a tool supports is more code that can hide a mistake, and redaction is the one job where a mistake cannot be taken back. So PDF Redact supports the common cases well and refuses the rest openly: when it meets content it cannot remove precisely, it removes the whole element and lists it in the result, rather than leaving something behind or spending years chasing edge cases and backwards compatibility. Each limit below carries one of three labels.
- By design
- Left out on purpose so the tool stays small, new and safe. We do not plan to add it.
- Removed whole
- The file opens, but this part cannot be edited precisely, so it is removed whole and listed in the result. Nothing is quietly kept.
- Not yet
- A gap we know about. It may change; this page will say when it does.
Opening files
The tool reads PDF 1.0 to 2.0, including damaged files, which it repairs by scanning for their objects. It stops at these.
-
By design
PDFs encrypted with a certificate (public-key encryption) or any security handler other than the standard password one do not open.
Instead: Ask the sender for a copy protected with a password, or with no protection.
-
By design
Files with no recoverable cross-reference table, trailer or catalog do not open, even after the repair scan.
Instead: Re-save the file from the program that made it, then redact the new copy.
-
Removed whole
Streams compressed with anything outside the common set (Flate, LZW, ASCII85, ASCIIHex, RunLength and JPEG for images) cannot be read. On a page you redact, such a stream is dropped whole and the result tells you.
Instead: Re-save the file with a mainstream PDF tool, which rewrites the streams with standard compression.
-
By design
The password and the permission flags of the original (no printing, no copying) are neither enforced while you work nor kept in the saved copy, which is always unencrypted.
Instead: Add a password or restrictions to the redacted copy with your PDF viewer if the recipient needs them.
Images and scans
Pixels under a redaction are blackened in the image data itself for the formats most PDFs use: unfiltered, Flate, LZW and baseline or progressive JPEG, in gray, RGB, CMYK, indexed, Lab, ICC, Separation and DeviceN color. Everything else is removed whole when a redaction touches it.
-
Removed whole
Images stored as JBIG2, JPEG 2000 or CCITT fax, which some scanners and archive tools produce, are not decoded. A redaction that touches one removes the whole image, and the preview shows it blank.
Instead: Check the result for the missing picture. If the rest of the image must stay, re-save the scan as a regular JPEG or PNG based PDF first.
-
Removed whole
Unusual JPEGs (lossless, arithmetic-coded, hierarchical, 12-bit or 16-bit) and images whose data does not match their description are removed whole.
Instead: Re-export the PDF from its source so the images are standard 8-bit JPEGs.
-
Removed whole
Images drawn inline in the page content, images in a Pattern or malformed color space, and images over about 268 megapixels are removed whole.
Instead: These are rare; review the result and re-export the page if the picture must stay.
-
By design
A redacted JPEG is stored again losslessly, so the saved file can be larger than the original. It is never re-encoded as a JPEG.
Instead: Nothing to do. Lossless storage keeps every unredacted pixel exact, which matters more than file size.
-
By design
There is no text recognition (OCR). A scanned page is a picture, so search finds nothing on it.
Instead: Draw boxes over what to hide. The pixels are removed from the scan itself.
Fonts and search
Search works on the text a PDF reader could copy: fonts with a Unicode mapping, standard encodings, Identity and Unicode CMaps, Type 3 fonts and ActualText, including vertical writing. Case is folded and ligatures are expanded. Beyond that:
-
By design
Text in a font with no Unicode mapping cannot be searched, because the file does not say which characters the glyphs are.
Instead: Select the text by hand or draw a box over it. It is removed like any other text.
-
Removed whole
Fonts that rely on a regional CMap (Shift-JIS, GB, Big5 or KS encodings) rather than Unicode have unknown character widths. A redaction in such text removes the whole string.
Instead: Review the result; the removed string is listed. Files from modern tools use Unicode CMaps and are unaffected.
-
Removed whole
Text in a font of an unknown type, or whose glyph widths cannot be determined, is removed whole where a redaction touches it, so the remaining text cannot shift.
Instead: Review the result; each removed text object is listed.
-
Not yet
Search is a plain phrase match. There are no regular expressions, no wildcard patterns, and accents are not folded (searching "Jose" does not find "José").
Instead: Search for each spelling, or select the text by hand.
-
Not yet
Right-to-left scripts are searched in the order the file stores them, which may differ from the order you read them.
Instead: Select right-to-left text by hand rather than by search, and check the preview.
Forms, annotations and interactive content
Interactive features are removed or scrubbed rather than edited, which is the safe choice for the content they can hide.
-
Removed whole
Annotations and form fields that touch a redaction, or that show the searched phrase anywhere, are removed whole, hidden and print-only ones included. Their appearance is never partly edited.
Instead: Review the result: each removed annotation is listed. Flatten the form first if its fields must stay.
-
By design
XFA forms are not supported. The XFA data is removed whenever document text is scrubbed, leaving the plain PDF form.
Instead: Export the form as a standard PDF before redacting if the XFA version is the one that matters.
-
Not yet
A searched phrase is scrubbed from bookmarks, form fields, annotations, links, page labels, layer names, document information and XMP, but not from file attachments embedded in the PDF, document-level JavaScript, the open action, or field event scripts.
Instead: Remove attachments and scripts before sharing if they may hold the same information. The tool never runs scripts.
-
By design
Content drawn through tiling patterns, shadings and Type 3 glyph procedures is not searched. A shape painted with a pattern is removed, but text inside the pattern definition itself stays.
Instead: These are rare in documents people share. Draw a box over the area and check the result.
-
By design
Multimedia, 3D, rich media and PDF portfolios are not supported.
Instead: Extract the documents from a portfolio and redact each one.
-
By design
Digital signatures do not survive redaction, because the file is rewritten in full rather than updated in place.
Instead: Redact first, then sign the redacted copy.
Document structure and the saved file
Saving writes a clean new file: only the objects the document still uses, numbered from one, with a classic cross-reference table and standard compression. Older revisions of pages and images never survive.
-
By design
Accessibility tags (the structure tree) are removed from redacted documents, because they can hold hidden copies of the redacted text. Tagged PDF and PDF/UA conformance are lost.
Instead: Re-tag the redacted copy in your authoring tool if accessibility conformance is required.
-
By design
PDF/A and other conformance claims are not checked or preserved. The saved file is a regular PDF.
Instead: Convert the redacted copy to PDF/A afterwards if an archive requires it.
-
By design
The saved file is not linearized for fast web view, not incrementally updated, and never encrypted. Its version is at least PDF 1.4.
Instead: Nothing to do for most uses. Run the file through an optimizer if you need fast web view.
-
By design
The tool redacts; it does not edit. It does not merge, split, reorder or OCR pages, convert formats, or fill forms.
Instead: Do those in your usual PDF tool, before or after redacting.
The browser tool
Everything runs in your browser on your own device, which is why your files never leave it. It also sets these limits.
-
By design
Needs a current desktop browser with WebAssembly. Internet Explorer and other legacy browsers are not supported, and there is no app to install.
Instead: Use a recent Chrome, Edge, Firefox or Safari.
-
By design
Large files are bound by your device's memory, and the page is busy while a redaction runs. There is no server-side processing to fall back on.
Instead: Close other tabs for very large files, or split the document first.
-
By design
The preview draws fonts the PDF does not embed with a built-in substitute, so some characters (symbols, dingbats, CJK) show as boxes, and mesh shadings, some blend modes and hidden layers are not drawn. The redaction itself is exact; only the picture on screen is approximate.
Instead: The sidebar lists everything not drawn faithfully. Check the saved file in your PDF viewer.
-
Not yet
One document at a time. Opening another file clears the current selection.
Instead: Finish and save one file before opening the next.
-
Not yet
There is no undo. Selections can be removed one at a time or all at once, and nothing is changed until you choose Redact.
Instead: Review the selection list before redacting. The original file on your disk is never touched.
-
Not yet
Pointer only: no keyboard shortcuts, and a text selection cannot span more than one page.
Instead: Select on each page in turn, or search to redact every match at once.
-
Not yet
Output options are the box color and whether to remove metadata. There is no overlay label such as REDACTED, no option to keep annotations or accessibility tags, and no printing.
Instead: Save the redacted PDF and print or label it in your PDF viewer.
-
Not yet
Built for a desktop screen. Touch works for selecting and drawing boxes, but small screens only get a stacked layout.
Instead: Use a laptop or desktop for long documents.
Supported, in case you wondered
The limits above read best in proportion. These are things a small tool is often assumed to lack, and this one has them.
- Password-protected PDFs, including AES-256, opened with the user or the owner password
- Damaged files, repaired by scanning for their objects when the cross-reference table is broken
- Progressive and baseline JPEGs, with the covered pixels blackened in the image data itself
- Hidden copies of text: accessibility tags, ActualText, soft mask groups, and hidden or print-only annotations
- Vertical and mixed-direction CJK text when the font carries a Unicode mapping
- Every match of a phrase across the whole document, in bookmarks, form fields and links as well as on the page
- The editing history: earlier revisions of changed pages and images are dropped when the file is saved
Found a PDF that behaves in a way this page does not describe? Tell us at [email protected]. The security page explains how redaction works, and the redaction guide how to check a result.
Redact your next PDF properly
PDF Redact is free during invite-only early access. Ask for an invitation and start redacting.