The short version

Drawing a black box over text adds a black box. The text remains in the content stream, selectable and copyable by anyone who tries.

Test every redacted document before it leaves. Open it, select all, paste into a text editor. If the hidden words appear, the redaction was decorative.

Redaction failures are not rare and they are not the preserve of the careless. Court filings, regulatory submissions and published reports have all leaked this way, often reviewed by several people who each saw a page that looked correctly blacked out.

The reason it keeps happening is that the failure is invisible. A document with fake redactions and a document with real ones are pixel-identical on screen. The difference only shows up if you go looking, and nobody goes looking at something that already looks right.

Why a black box does nothing

A PDF page is a list of drawing instructions executed in order. Place text, place an image, fill a rectangle. The renderer follows them from first to last, so later instructions paint over earlier ones.

Adding a black rectangle appends one more instruction to the end of that list. The instruction that draws the text is still there, earlier in the same stream, unchanged. Anything reading the document rather than looking at it walks the whole list and finds the text exactly where it always was.

Real redaction

The instruction is deleted

  • Text and image operators removed from the content stream
  • Select-all and copy returns nothing from that area
  • File size drops if the removed content was substantial
  • Text extraction tools find nothing to extract
  • Safe to publish
A drawn rectangle

One more instruction on top

  • The text operator is untouched, just painted over
  • Select-all and copy retrieves it immediately
  • File size unchanged or slightly larger
  • Any extraction tool reads it in full
  • Has caused repeated public disclosures
Highlighting in yellow has the same problem
So does covering text with a white box, placing an opaque image over it, or setting the text colour to match the background. All of these change what is painted and none of them change what is stored. If the operation felt like drawing rather than deleting, the content is still in the file.

The five second test

There is one check that catches nearly every failed redaction, and it needs no special software.

1

Open the redacted file

In any reader. The one you redacted in is fine, though a different one is a slightly better test.

2

Select all, then copy

Ctrl+A or Cmd+A, then copy. This asks the document for its text rather than its appearance.

3

Paste into a plain text editor

Notepad, TextEdit, anything without formatting. If the words you redacted appear in that paste, the redaction did not happen.

Make this a habit, not a decision
Run the test on every document that goes out with redactions, regardless of who did them or how confident anyone is. It takes less time than opening the email you are about to attach it to, and it is the only step that distinguishes a real redaction from a convincing one.

Redacting properly

Real redaction removes the underlying content and then, ideally, rewrites the file so nothing is recoverable from an earlier revision. There are two reliable routes.

Use a tool that removes rather than draws

Dedicated redaction features delete the text and image operators inside the marked area and then flatten the result. The important detail is that applying the redaction is a separate, deliberate step from marking it. If a tool lets you save while the marks are still marks, you have a document with annotations rather than a redacted one.

Remove the page entirely

Blunt, completely reliable, and often the right answer. If a page is mostly sensitive, extracting the pages around it produces a document from which that content is genuinely absent, with no operators left behind to recover. The cost is that you lose everything else on the page, which is why this suits appendices and whole sections better than a single paragraph.

Rasterising works and costs you the text layer
Converting pages to images and rebuilding a PDF from them guarantees the original operators are gone, because the page is now a photograph. It also destroys searchability, accessibility and text selection for the whole document, and inflates the file considerably. It is a legitimate last resort, not a default.

What redaction does not cover

Removing the words on the page is necessary and frequently not sufficient. Several other things in the file can give away what you just removed.

AnatomyWhat redaction does not reach
Form field layervalues typed into fieldsAnnotation layercomments, highlights, signaturesPage contenttext and images, always drawnreader draws
Removing page content leaves the layers above it, and the metadata beside it, untouched.
Also worth clearingThe rest of the trail
  • Document metadataTitle and author
  • Bookmarks and outlineOften name sections
  • Comments and annotationsDiscuss the content
  • Form field valuesStored separately
  • Attached filesTravel inside the PDF
  • Earlier saved revisionsRecoverable in the bytes
Incremental saves keep the unredacted version
PDFs can be updated by appending changes to the end of the file rather than rewriting it. If a document was redacted and saved that way, the original content may still be present in an earlier revision inside the same file. Save a fresh copy, or run the finished document through a tool that rewrites it completely, so only the current state survives.

Redaction that leaks by implication

A technically perfect redaction can still disclose. The shape of what was removed carries information.

A redacted name whose box is narrow rules out long names. A paragraph reduced to two redacted words in a sentence that otherwise survives can often be reconstructed from context. Redacting one instance of a name while missing another elsewhere in the document defeats the whole exercise. And a table with one redacted cell in a row that otherwise totals correctly gives up the missing number to arithmetic.

Read the document as an outsider
After redacting, read the whole thing as though you know nothing, and ask what the remaining text plus the gaps tells you. This catches the same name appearing three pages later, the total that reveals the redacted line item, and the sentence whose grammar narrows the missing word to one candidate.

Redact so the text is actually gone

Find it, mark it, and remove it from the content stream. All of it in your browser.

Open Redact PDF

Common questions

Verifying that it worked.

01Why is a black box not real redaction?

Because a PDF page is a list of drawing instructions, and adding a rectangle appends one more instruction on top. The instruction that draws the text is still in the file, so anything that reads the document rather than viewing it retrieves the text unchanged.

02How do I check whether my redaction worked?

Open the file, select all, copy, and paste into a plain text editor. If the words you redacted appear in the paste, the content is still in the document. This takes about five seconds and catches nearly every failure.

03Does highlighting text in black redact it?

No. Highlighting adds a coloured annotation over the text and leaves the text itself untouched. The same is true of white boxes, opaque images and setting the text colour to match the background.

04Is deleting the whole page safer than redacting part of it?

It is more reliable, because nothing is left behind to recover. The trade-off is that everything else on that page goes too, so it suits whole sections and appendices better than a single sentence in the middle of a page you need.

05Does redaction remove metadata as well?

Usually not. Redaction operates on page content, while the title, author, timestamps and any annotations live elsewhere in the file. Strip the metadata as a separate step before sending anything sensitive.

06Can redacted content be recovered from an earlier version inside the file?

Yes, if the file was saved incrementally. PDFs can be updated by appending changes rather than rewriting, which leaves previous revisions present in the bytes. Save a fresh copy so only the current state exists in the file you send.