Working With a Redacted PDF You Can't Search or Copy
Almost every guide about redaction is written for the person holding the original. This one is for the person at the other end, handed a flattened file and asked to review it, quote it or respond to it by Friday.
The redaction problem has two sides and only one of them gets written about. On one side is the producer, deciding what to cover and worrying about whether the black actually removed anything. On the other side is you: a PDF has landed in your inbox, parts of it are blacked out, and before you can do anything useful with it you discover that Ctrl+F finds nothing, the text will not highlight, the footnote numbers will not copy and the cross-references are no longer clickable.
This is a genuinely different problem, and it is worth saying clearly that it is usually not a sign that anything has gone wrong. You have most likely been given a correctly redacted document. The thing that makes the redaction permanent is the same thing that broke your workflow, and it was not done to inconvenience you. But nobody has told you that, and in the meantime you still have to read the file, pull quotes out of it, cite it accurately, hand it to a colleague who uses a screen reader, and work out whether the parts you cannot see matter.
What follows is how to tell what kind of file you are holding, what you can realistically get out of it, and what to ask for when the answer is not enough.
First, work out which of three files you have
Three quite different things arrive looking identical on screen, and the right next step depends entirely on which one you have. The test takes under a minute.
Open the file and pick a word you can clearly read on the page, somewhere nowhere near a black box. A heading works well. Press Ctrl+F or Cmd+F and search for it. Then try select-all, copy, and paste into a blank text document.
Nothing is found and nothing pastes. The page is an image. Every page has been re-rendered as a picture and the PDF rebuilt around those pictures, so there is no text anywhere in the file, redacted or otherwise. This is the most common outcome and the one the rest of this page is about. The tool on this page produces exactly this: it re-renders every page to a canvas, paints the black rectangles onto that rendering, encodes the result as a JPEG and builds a new page from it, so no text object survives on any page, including pages nobody drew on.
Visible words are found, and the text pastes out cleanly. The text layer is intact. Now the important question flips: try searching for something you would expect to be under a black box, and try selecting across one. If hidden text comes out, the redaction is cosmetic and the document has a defect. That is a finding, not a shortcut, and the right response is to tell the sender rather than to keep reading. There is a whole method for checking a redacted PDF someone sent you, and the same checks that protect them protect you from acting on material you were never meant to have.
Visible words are found, but nothing comes out from under the boxes. This is the best case for everyone: a tool removed the underlying text objects and left the rest of the page as real text. You can search, quote and cite normally. The discussion of whether a redacted PDF stays searchable covers why the three outcomes exist and which tools produce which.
Getting work done with an image-only file
Assume from here that you are in the first case. The file reads fine and behaves like a scan. Here is the order that wastes the least time.
- Read it before you process it. A surprising share of the reason people reach for OCR is that they wanted to find one figure on one page. If the document is under about thirty pages, paging through at full width is faster than any conversion, and it is the only approach that is guaranteed accurate.
- Zoom generously and stop worrying about blur. At typical flattening settings the page is around 144 pixels per inch, which is fine at page size and visibly soft at 400% zoom. Softness at high magnification tells you about the rasterisation, not about tampering, and not about the quality of the original.
- OCR it if you need to search, with your expectations set correctly. Running optical character recognition over the file gives you a searchable copy and a text transcript. It is genuinely useful for locating things. It is not a substitute for reading the page. The Tesseract project's own documentation says the engine works best on images which have a DPI of at least 300 dpi, and a flattened PDF page is typically well below that: PDF pages are measured in points at 72 to the inch, so a renderer working at twice page scale lands at about 144 pixels per inch, meaning a US Letter page becomes roughly 1224 by 1584 pixels. Add JPEG compression and the character edges soften further. Expect ordinary body text to come through usable and expect trouble in small print, dense tables, footnotes and anything handwritten.
- Never quote from OCR output. Find the passage with the search index, then read the words off the image and type the quote yourself. OCR substitutions in numbers are the dangerous ones, because a transposed digit in a figure looks completely plausible and nothing flags it.
- Pin down your page references before you cite anything. A rebuilt file carries no page labels and no bookmark outline, so your viewer numbers pages by position, starting at one. If the original had front matter numbered separately, or if whole pages were removed rather than covered, the number printed on the page face and the number in your viewer will disagree. Cite both: "page 14 of the PDF (marked page 9)". It costs five words and prevents an argument.
- Expect every link to be dead. A page rebuilt from an image has no link annotations, no form fields and no comment objects, because none of the original annotation layer is carried across. A table of contents still looks like a table of contents and does nothing when clicked. Internal cross-references have to be followed by hand. This is normal and not a sign the file was mangled in transit.
- Handle accessibility as a format request, early. An image-only PDF is effectively empty to a screen reader. If you or a colleague needs this document read aloud, OCR on your end is a workaround of last resort; ask the sender for a tagged, text-based version instead. Raise it as soon as you know, because it usually takes the producer a few minutes and takes you a lot longer.
Judging the redactions themselves
Separately from usability, you will want to know whether the missing material matters. A flattened file gives you less to work with than you might expect, and two habits help.
Note the shape and position of what is covered rather than guessing at its content. A box the width of a line in a signature block is a name or an address; a box covering a whole table column is a category of data. That tells you what kind of information was withheld, which is often enough to decide whether to ask about it. Guessing the actual characters from the box width is a different matter and a weak inference, because box dimensions are a product of how someone dragged a pointer, not of what the text measured.
Then ask for a basis rather than for the content. "Which pages were redacted, and under what exemption or on what ground?" is a question a producer can answer without disclosing anything, and the answer is far more useful than a dispute about one rectangle. Many producers keep a redaction log for precisely this purpose and will send it if asked.
Common mistakes and misconceptions
"I can't search it, so they're hiding something." Flattening to images is the standard way to make a redaction irreversible. It is what careful producers do. The file you can search is the one more likely to be leaking.
"I searched for the name and got nothing, so it isn't in there." In an image-only PDF every search returns nothing. A clean search result is a statement about the file format, not about the contents. This cuts both ways and is the single most common error in reading these documents.
Trying to brighten, invert or "enhance" a black box. Where the page is a true raster the original pixels were overwritten before the image was encoded. There is no hidden data in a solid black rectangle to pull out, no matter what is done to the levels. Time spent here is wasted even before considering whether it would be appropriate.
Treating the OCR text file as the document. It is a derived index with an error rate. Keep the PDF as the authority, and keep the OCR output clearly labelled so it does not get circulated as a transcript.
Annotating it and sending it back as a shared record. You can mark up an image-only PDF in most viewers, but if the file is ever run through a flattening tool again, every comment is either burned into the picture or discarded. For anything that needs to survive, put your comments in a separate document that cites page numbers.
Reading file size or page count as evidence. A flattened file is often much larger than the original because text has become photographs, and page counts can shift if pages were withheld entirely rather than covered. Neither number tells you anything reliable about how carefully the redaction was done.
Assuming the sender knows your workflow broke. Most producers never open their own output in anger and have no idea that you cannot search it. A short, specific message asking for a searchable copy is usually met with a shrug and a file, not an argument.
Related guides
Explore more ways to redact PDFs privately, or use the redaction tool above:
Frequently asked questions
Why can't I search or copy any text in the redacted PDF I was sent?
Most likely because the file is no longer a text document. The common way to make a redaction permanent is to flatten each page into an image and rebuild the PDF from those images, which destroys the text objects along with everything that was hidden under the black. The tool on this page works exactly that way: every page is re-rendered to a canvas and embedded as a JPEG, so the export contains no selectable text anywhere, not only where boxes were drawn. Nothing is being concealed from you by that choice, and it is not a sign of a sloppy job. It is closer to the opposite: a file you can still search is a file where the original text objects may well have survived underneath the black rectangles. Before you assume flattening, confirm it. Press Ctrl+F or Cmd+F and search for a word you can plainly see on the page. If a visible word returns no hit, the page is an image.
Can I run OCR on it to make it searchable, and how accurate will that be?
You can, and it is often worth doing, but treat the result as a rough index rather than a transcript of record. Resolution is the limiting factor. The Tesseract documentation states that it works best on images which have a DPI of at least 300 dpi. A PDF page is measured in points at 72 to the inch, so a tool that rasterises at twice page scale produces roughly 144 pixels per inch: a US Letter page at 612 by 792 points becomes about 1224 by 1584 pixels. That sits well under the 300 dpi figure, and the pages are usually JPEG compressed on top of it, which softens exactly the character edges that OCR depends on. Body text at a normal size often comes through usable. Small print, dense tables, footnotes, handwriting and unusual typefaces are where errors cluster, and OCR mistakes in numbers are both common and easy to miss. Use OCR to find candidate pages, then read the figure off the image with your own eyes before you rely on it or quote it.
The black boxes look like they are drawn on top. Can I see what is underneath?
If the page is genuinely an image, no. The redaction was applied to the rendering before the image was encoded, so the pixels that used to show the text were overwritten with black pixels and there is no separate layer to move, delete or brighten. Adjusting levels, contrast or exposure on a solid black rectangle recovers nothing, because nothing is left in those pixels to recover. The case that behaves differently is the one where the text layer survived and a black rectangle was merely drawn over it. There you may be able to select, copy or search the hidden text with no special effort at all, and that is a defect in the document you were given rather than a tool you should quietly reach for. If you find it, the appropriate step is to tell the sender that their redaction did not take effect, and to be careful about what you do with what you saw. Checking a file someone sent you for exactly this is a separate job with its own method.
What should I ask the sender for if I need a version I can actually work with?
Be specific, because a vague complaint usually produces another copy of the same file. Three requests are reasonable and easy for a sender to act on. First, ask for a text-searchable version: the producer can OCR the redacted file from their own high-quality original and hand you a searchable copy without undoing any redaction. Second, ask for a page index or a redaction log, a short list stating which pages were touched and on what basis, which lets you work out whether a gap matters to you. Third, if you need the document for accessibility reasons, say so plainly and ask for a tagged, text-based version, because a flattened image carries nothing a screen reader can read aloud. Keep that request separate from any challenge to the redactions themselves. Asking for a usable format is a format question, and framing it that way tends to get a faster answer than bundling it with a dispute about what was withheld.