Many well-known redaction failures — court filings, government releases, corporate disclosures — had the same cause. Someone drew a black rectangle over the sensitive text and shipped the file. The words were still there, one copy-and-paste away.
Understanding why that happens is the whole skill. Once you know where a PDF keeps its words, redacting properly is a short checklist.
A PDF page is a script, not a picture
Every page in a PDF is a list of drawing instructions: set this font, move here, draw these characters, move there, fill this rectangle. A viewer runs the list top to bottom and paints what it says.
When you use a drawing tool to cover text, the editor appends one more instruction — fill a black rectangle here — to the end of the list. The instruction that drew the text is still in the list, still executed, just painted over. That is why:
- Select-all and copy returns the hidden text: the selection follows the text instructions, not the pixels.
- Search finds the hidden words.
- Any editor can delete the rectangle and reveal the page exactly as it was.
- Screen readers and AI tools read the text straight out of the file.
Real redaction edits the list: it removes the text instructions (and any graphics) inside the marked area, then adds the box so a reader can see something was removed. After that, there is nothing underneath to find.
Blacking out text is not removing it
The phrase people reach for is “black out” — black out a name, black out an account number — and the phrase is the trap. It describes what the page should look like afterwards, and every editor can produce that look in seconds by drawing a filled rectangle.
But a drawn rectangle is one more instruction appended to the page’s script: fill this area with black. The instructions that placed the text are still there, earlier in the list, untouched. Select the region and the words highlight through the box. Search the file and they are found. Delete the shape and they are simply visible again. Print to PDF and some workflows drop the overlay entirely.
This is also the honest answer to “can a redacted PDF be un-redacted?” — if it was blacked out with a shape, yes, trivially, and that is how nearly every public redaction failure happened. If the text was actually removed from the page content and the file saved as a new copy, it cannot be read back from that page.
Two of the best-known cases show how little it takes. In December 2009 CBS News reported that the US Transportation Security Administration had posted an airport screening manual online whose redactions were black boxes that, in CBS’s words, anyone with basic knowledge of Adobe Acrobat could have removed (CBS News). In January 2019 BuzzFeed News showed that the blacked-out passages of a court filing by Paul Manafort’s lawyers could be read by copying the text and pasting it into another document; they revealed that prosecutors had accused him of lying about sharing 2016 campaign polling data with Konstantin Kilimnik (BuzzFeed News).
Redaction is deletion with a black mark left behind as a courtesy. The mark is the least important part.
The three places information hides
Redaction is only complete when all three are handled.
1. Page content
The text and graphics on the page itself — the part described above. This is where the redaction tool does its work.
2. The text layer of scans
A scanned document is a picture of a page. If it has ever been run through OCR — by a scanner, an office tool, or a previous “make searchable” step — it also carries an invisible text layer that mirrors the picture. Covering the pixels while leaving that layer intact is the scan version of the black-box mistake. Proper redaction must remove the marked pixels and the hidden text under them.
3. Everything outside the pages
Metadata (author, title, subject, keywords, creator, dates), bookmarks that quote headings, comments and sticky notes, form-field values, and attached files. None of these are on the page, so no page tool touches them. They need a separate pass.
How Filozy redacts
The Redact PDF tool works on your device, and it works by rebuilding pages rather than covering them:
- You mark by hand. Drag a box over each thing to remove on the page previews: a name, a number, a signature, a photo. Click a box to take it off again. The tool does not search the text or suggest matches, so find every occurrence first, in your usual viewer.
- Marked pages become images. Each page with a mark is redrawn at 150 dpi with the marked areas painted black, and that picture replaces the page. Nothing that was under a mark survives on that page, and neither does the rest of its text layer: the page can no longer be searched or selected.
- Unmarked pages keep their text. Pages without a mark are copied rather than redrawn, so they stay searchable, and anything sensitive you did not mark on them is still there. Two exceptions: if your PDF reader shows the file as secured or encrypted, even though it opens without a password, the copied pages come out blank, and content on a hidden layer becomes visible. Check every copied page in the result.
- Checked before you get it. The result is re-opened, and if any text can be extracted from a page you marked, you get an error instead of a file.
- What the check cannot see. Unmarked pages are copied with everything they refer to. If one of them links to a page you marked, as a contents page or a “see page 4” link does, or shares images with it, the marked page’s original text and images can be copied into the file where no viewer shows them, and this check does not look there. To be sure, put a small box in a blank margin of every page, so every page is rebuilt as an image and nothing from the original is copied. The cost is that no page keeps selectable or searchable text, and each margin box shows as a small black mark.
- A new file. The original’s title, author, subject, keywords and bookmarks are not copied into the result. Comments on unmarked pages are, and so is any file attached to one of them.
- Forms are refused. A PDF with fillable fields has to go through Flatten first.
The document never leaves your browser. For redaction that matters more than for any other job: uploading a file to redact it means the unredacted version has already been shared once.
If the file is going to an AI assistant rather than a person, the same steps apply, and consistent placeholders sometimes work better than black boxes. See Remove personal information from a PDF before uploading it to ChatGPT.
Step by step
- Work on a copy. Keep the untouched original somewhere safe and separate; you will not be able to un-redact.
- Find every occurrence first. Search the PDF in your usual viewer for each name, number and phrase, and note every page it appears on. A text search cannot see words inside photos, scans or signatures, so look through those pages too.
- Open Redact PDF, choose the copy, and drag a box over each item on its page.
- Check every box against your list, then save the redacted file.
- Test it. Open the result, press select-all and paste into a text editor; search for each term you removed. Nothing from a page you marked should come up. Pages you did not mark still have their text, so anything missed there will appear. A clean search does not rule out hidden copies of a marked page (see “What the check cannot see” above).
- Open the result in the Metadata tool and check what it says, and open any comments on the unmarked pages.
When redaction is the wrong tool
- You need the text gone but the layout kept for editing — redaction is permanent and visible; use an editor on the source document instead.
- The information is in an image you need to keep — a photograph with a licence plate, say. Redaction removes the region; if you need the rest of the image intact, edit the image first, then rebuild the PDF.
- Someone must be able to reverse it. Redaction is meant to be one-way, so keep the original for anyone who needs the full text.
The checklist, for the wall
Find every occurrence before you mark. Mark each one. Save. Select-all and search the result. Check the metadata and comments. Keep the original elsewhere. Only then send it.