How Redaction Studio works

The problem this is built for

When most tools “black out” a passage, they draw a black rectangle on top of the words and leave the words in the file underneath. The page looks redacted. Select the area, copy, paste, and the text comes back. This has caused real leaks in court filings and published government documents, and it is why our free checker exists.

Removing the words properly is better, but it is not the end of the story — see the next section.

Rasterising: what the tool actually does

“Rasterising” means turning something into a picture made of dots. That is the whole method here, and it is worth understanding because it explains both the strength and the cost.

When you press Apply, for every page:

• The page is drawn onto a blank canvas at the resolution you chose, exactly as it would look if you printed it.
• The redaction boxes are painted onto that picture, in black. At this point the covered dots are simply black dots. Whatever was underneath is gone, because a picture has no underneath.
• A brand-new PDF is built, with that picture as the page.

Nothing is copied across from the original page. Not the text, not the fonts, not the notes, not the attachments, not the earlier versions. The new file is not your document with things taken out — it is a new document containing photographs of your pages.

That is why so much disappears at once. Comments, form contents, hidden layers, embedded files, bookmarks, page thumbnails and document properties are not individually stripped. They are simply never carried over.

Why this defeats the spacing attack

In 2023 researchers showed something uncomfortable about redaction done the “proper” way. Even when the hidden words are genuinely deleted from the file, the letters on either side of the gap stay exactly where they were. Letters are not all the same width, so the size of the gap, and the precise positions of the surviving letters around it, carry information about what used to be there.

With a list of likely candidates — a set of names in a case, say — that is often enough to work out the removed word. The researchers assessed eleven widely used redaction tools, including the best-known commercial one, and found that all of them leaked this way. They were able to recover redacted names from real published documents.

Turning the page into a picture removes the thing being measured. There are no letter positions in a picture, because there are no letters — only dots. The measurement cannot be taken, so the attack has nothing to work on.

This is the unusual claim of this tool: the simple-looking approach is the more secure one. Tools that keep your text searchable are keeping the very thing that leaks.

The three box styles

A black box still says something about what it hides: how long it was. These three settings trade that off against how much of the page gets covered.

Uniform blocks — the default

Every box in the document is drawn to the same width, whatever the length of the thing it covers. A three-letter name and a thirty-character account number get identical boxes, so nobody reading the output can tell one from the other, or guess a short word from a short box. This is the safest setting and the reason it is the default.

The cost: if you mark one long passage, every box in the document becomes that wide. The tool tells you when this is happening and suggests whole-line mode instead.

Whole line

The entire line of text is covered, from the start of the writing to the end of it. Nothing at all can be inferred from the shape of the box, because the box is simply the line. It is the safest option and the most disruptive, since it takes the surrounding words with it.

Exact fit

The box matches what you marked. It is the tidiest to read and the leakiest: anyone can measure the box and learn how long the hidden text was. The tool shows a warning when you choose it. Use it when length is not sensitive — covering a signature, say, rather than a name.

Why boxes are slightly larger than the words

Every box is drawn a little wider and taller than the text it covers.

Letters do not stay inside their own boundaries. The tails of g, j, p, q and y hang below the line. Italic letters lean out past where they begin. Accents sit above. A box cut to the exact width and height of the words would clip those parts, and a clipped letter is a visible letter.

So the padding is deliberate, and it means a word sitting immediately beside the one you marked may end up partly covered. That is the intended behaviour, not a fault. Covering slightly too much is recoverable — you can always redact again with a smaller selection. Covering slightly too little is not.

Finding text to redact, and when that is not possible

When you search for a term, the tool does not guess where the words are on the page. It reads the page’s own drawing instructions — the list of commands that tells a screen or a printer where to put each letter — and takes each letter’s position and width from there. It is working from the same information the renderer uses, which is why the box lands where the letters are.

Some pages carry no text at all. A scan is a photograph: you can read the words on screen, but as far as any software is concerned there is nothing there but a picture. Searching such a page cannot find anything, however obvious the words look to you.

The tool names those pages on screen when you open the file, and repeats the warning with your search results, because a search that quietly returns nothing on a page full of sensitive words is the most dangerous thing a tool like this could do. On those pages you must draw the boxes by hand. The QA report records which pages could not be searched, so the limitation travels with the document rather than staying in your memory.

Checking its own work

Before you are offered a file, the tool re-opens the file it has just written and inspects it as if it were a stranger’s document. It checks that:

• no text can be extracted from anywhere in it;
• the page count matches the original;
• there is exactly one version of the document inside the file, with no earlier version attached;
• there are no comments, form fields, attachments, scripts, bookmarks, hidden layers or page thumbnails;
• there is no author, title, subject, keyword, creator or producer recorded;
• every redaction box is solid black across its whole area — not merely present, but complete;
• for anything found by searching, the area the target words occupied is completely covered.

That last check exists because of a fault found during testing. A box can be perfectly solid and still be in the wrong place, leaving the words it was meant to hide sitting beside it. Checking the box alone could not see that. Checking the place where the words were can.

If any check fails, no file is produced. You get an explanation of what failed instead of a download. A refusal is recoverable; a file that looks redacted and is not, is not.

The QA report

Alongside the redacted PDF you get a plain-text report recording the file name, the page count, the date and time, how many redactions were applied and on which pages, which box style was used, which pages could not be searched, and the result of every check listed above.

The report deliberately contains none of your document. Not the redacted words, not the terms you searched for, not a line of the text. It is written to be attachable to a file note without becoming a second copy of the thing you were trying to protect.

What this tool does not do

• It does not keep your document searchable. The output is a picture. You cannot select or search the text in it, and screen-reader software cannot read it aloud. If you are required to produce accessible documents, this output does not meet that requirement. This is the real cost of the method and we would rather you knew before buying than after.
• It does not hide that something was removed. Black boxes are visible. This removes content; it does not disguise the fact of removal.
• It does not stop anyone reading what is still on the page. Anyone can run text recognition over the parts you left visible, exactly as they could with a photocopy.
• It does not open encrypted or password-protected files. Remove the password in whatever created the document first.
• It does not decide what is sensitive. The pattern presets match shapes — something that looks like an email address, or a card-style number. They are a starting point for your own review, never a substitute for it, and nothing is ever applied without you confirming it.
• It does not keep a digital signature. Redaction breaks signatures by definition. The signature is removed and the report says so.
• It does not make you compliant with anything. Compliance is a property of your whole process, not of a downloaded file. The only claim this tool makes about itself is architectural: your file is not transmitted.

Checking the “no upload” claim yourself

Do not take our word for it. Load the tool, turn off your wifi, reload the page, and redact a document. It works with no connection because there is nothing for it to connect to.

If you want to go further: the source of the tool is readable — open the file in a text editor. The page carries a browser instruction that forbids network connections outright, and you can watch for yourself in your browser’s developer tools that a full redaction makes no requests at all.