Redacting a PDF Someone Else Already Redacted
You are not the redactor and you are not the reviewer. You are the second link in a chain — inheriting another person's marks, their scope decisions and their mistakes, and adding your own on top.
Almost every guide on this site addresses one of two people. The first is the person holding an original document who has to decide what to cover and then cover it properly. The second is the person on the receiving end, who has to work out whether what arrived is actually safe — that is the job described in how to check a redacted PDF someone sent you. Both of those are single-party problems. One document, one set of decisions, one person accountable for them.
This page is about the awkward third position, and it comes up constantly once you notice it. An agency sends you a partially redacted response and you need to redact further before passing it to a client. Opposing counsel produces documents already marked up and you must cover your own client's identifiers before they go into an exhibit bundle. A journalist receives a leaked file with names already blacked out and needs to remove a few more before publishing. A compliance team gets a vendor report with pricing hidden and has to hide employee names too. In every one of those, you are redacting a document that has already been redacted by someone whose work you cannot inspect, whose reasoning you were never told, and whose marks will shortly become indistinguishable from yours.
That changes the shape of the task in three specific ways. You cannot audit scope, because you have no original to compare against and no record of what the first party removed. You cannot verify their method, only guess at it from the artefacts. And you are about to create a document in which two sets of redactions, made by two parties for two different reasons, look exactly alike. None of those problems is solved by being careful with the mouse.
Start by placing yourself in one of three situations
The right move depends entirely on which of these you are in, and it is worth deciding explicitly rather than drifting.
- Their redaction is sound and you are adding to it. You have tested the file and the covered content is genuinely unrecoverable. Your pass is additive: you cover your own items and the result is a two-generation document with one honest set of pixels.
- Their redaction is broken and you are also fixing it. The covered text selects, copies or searches. You are now doing two jobs at once — repairing their failure and performing your own redaction — and you should treat the exposure as an incident in its own right, not a detail of your workflow.
- You cannot tell, and the answer matters. This is the most common case and the one people handle worst, usually by assuming the best. If the document is going anywhere consequential, the correct response is to treat it as unverified and flatten it, which makes everything on the page permanent whether their marks were real or not.
Only the second case obliges you to go back to the sender. All three oblige you to write down which one you decided you were in.
What a second pass actually does to the file
The mechanics matter here more than usual, because a second pass is where people form wrong expectations about what is being preserved.
Every page is rebuilt as an image. The export renders each page of the document — including pages you never touched — to an offscreen canvas at a fixed scale of two, which is roughly 144 dots per inch, burns your rectangles in as solid black, encodes each page as a JPEG, and assembles a brand new PDF from those images. Nothing from the incoming document's internal structure is carried across. This is why the second pass is a useful repair for a bad first pass, and it is explained from first principles in permanent redaction versus visual masking.
Nothing but the pictures survives. The output document is created empty and receives only the page images, so it has no annotations at all: no comments, no sticky notes, no form fields, no bookmarks and no clickable links. If the first party left review comments or an annotation-based markup layer in the file, the visible appearance of those marks is painted into the pixels, but the objects themselves do not come through. That cuts both ways — stray reviewer comments disappear, and so does any annotation trail that documented their redaction decisions.
The document's identity is reset. The export sets the title to Redacted Document and the producer and creator both to HidePDF, and stamps a fresh creation and modification date at the moment you export. The incoming file's author, dates and producer string are not preserved. If you were relying on metadata to show that the document came from a particular agency on a particular date, capture that before your pass, because the file will no longer say it.
The filename accumulates. The download is named after the file you opened with -redacted appended. Feed in response-redacted.pdf and you get response-redacted-redacted.pdf, which is a genuinely useful signal that two passes happened — and a genuinely confusing one once there are four files in a folder. Rename deliberately rather than letting the suffixes stack.
Cropped-away content is dropped, not hidden. If the first party narrowed the page's crop box to push content out of view, only the cropped region is rendered and the new page is built at the cropped dimensions. The off-page content is genuinely gone after your pass, which is good — but the page size changes, so compare dimensions before and after if the document has to match a filed or printed original.
Where an inherited redaction can quietly come undone
A covering mark only becomes permanent if it is actually drawn when the page is rendered. There are two ways that fails, and both are worth knowing because the symptom is a page that looks fine to the first party and wrong to you.
The first is optional content — PDF layers. If the covering marks live in a layer, the document carries a default configuration saying which layers are on. This tool does not override that configuration on either the preview or the export, so the document's own default decides. A layer that is on by default is painted and becomes permanent. A layer that is off by default is painted in neither the preview nor the export, so a redaction layer shipped switched off never reaches your delivered file, and the content it was meant to cover is simply visible.
The second is annotation visibility flags. A mark drawn as an annotation carries flags that control whether it is shown on screen, in print, or both. An annotation flagged hidden or no-view is excluded from on-screen rendering, which is exactly the path the export uses — so a print-only black stamp covers the content on paper and covers nothing in your output.
The reassuring half of this is precise and worth leaning on: because the preview and the export use the same rendering path with the same defaults, the preview is an exact predictor of what the export will contain. Whatever you can read on screen in the editor is what the recipient will be able to read. That makes a slow, page-by-page look at the preview the single highest-value step in a second pass, and it is a check no amount of trust in the first party can replace.
A workflow for the second pass
- Test the incoming file before you draw anything. Try to select and copy the covered regions, run a document-wide search for a term you expect to be hidden, and note whether any text selects at all. This takes two minutes and determines which of the three situations you are in.
- Record the incoming file's identity first. Note its filename, size, page count, dates and producer string, and keep a hash if your process uses them. Your export preserves none of it.
- Keep the file you received, unmodified, as its own generation. Not as a working copy. You will need it if anyone later asks what you were given as opposed to what you produced.
- Page through the whole preview before marking. You are looking for content that should obviously have been covered and is not — the signature of a layer or annotation that did not render for you.
- Mark only your own items, and list them as you go. Write down page number and what each rectangle is for while you are drawing it. Once exported, your boxes are anonymous black pixels like everybody else's.
- Do not cover their marks with yours. Redrawing over an existing black area adds nothing and destroys the only visual evidence of where the first party's redactions were.
- Download, then verify from the downloaded file. Check the export itself, not the editor: no text selects, your items are covered, nothing new has appeared, and the page count and dimensions are what you expect.
- Log two generations explicitly. One block for what you received and in what state, one for what you removed and why. A single undifferentiated list is worse than useless when somebody later asks who hid a particular name.
Common mistakes and misconceptions
Assuming their black boxes were real redactions. They often are not, and the file looks identical either way. This is the error that puts organisations in the news, and inheriting the document does not make it someone else's problem once you have passed it on.
Assuming their black boxes were fake and re-covering everything. The opposite over-correction. Drawing your own rectangles over theirs does not make their redaction more permanent — the flattening does that, whether you draw over them or not — and it erases the boundary between two parties' decisions.
Treating a second pass as a way to audit scope. It cannot be. Flattening makes the current appearance permanent; it tells you nothing about whether the first party covered everything they should have. Scope is a question about a document you do not have.
Expecting the output to show who did what. It will not. Both sets of marks are black pixels in one raster, with no annotations, no layers and no edit history in the file. The distinction has to be written down outside the document or it does not exist.
Letting the filename suffixes stack unchecked. Three generations of -redacted-redacted-redacted in a shared folder is how the wrong version gets sent. Rename each generation to something that says what it is.
Forgetting the quality floor compounds for non-flat sources. Two passes of this tool land on nearly the same pixel grid, so the cost is a second JPEG generation rather than a resize. A high-resolution scan put through even one pass is a different story: it is rendered toward 144 dots per inch and does not come back. Test a page with the smallest type in it.
Reporting an exposure quietly, or not at all. If the covered text was readable, that is an incident for the party that produced it, and they may have sent the same file to others. Your repair protects your copy only. Tell them, in writing, separately from your own delivery.
Assuming anything left your device during any of this. It did not, in either pass you perform here. The page's code reads your file inside the browser tab, no upload is required at any point, and the only file written anywhere is the one you choose to download.
Related guides
See also, or use the redaction tool above:
Frequently asked questions
Will my second pass make the first party's redactions permanent as well as my own?
In the ordinary case, yes, and that is the single most useful thing a second pass does. The export rasterizes every page — not only the pages you marked — and rebuilds a fresh document from those images, so no selectable text survives anywhere in the file you get back. If the first party drew a black rectangle over live text and never removed the text, your pass flattens rectangle and text together into pixels, and the text is gone for good. The qualification is that this only covers content that your copy actually renders. Anything the first party hid in a way that your viewer and this tool both skip — a covering mark that is not drawn, or content that is drawn where a mark is missing — is handled according to what appears on screen, not according to what they intended. The preview is the authority. Read the preview carefully before you download, because what it shows is what the export will contain.
How can I tell whether the file I received was already rasterized by a tool like this one?
Try to select text on a page you know has words on it. If nothing highlights and a document-wide search finds no matches, the page is an image and someone has already flattened it — either by rasterizing, or by scanning paper. A second clue is the document's own metadata: an export from this tool sets the title to Redacted Document and both the producer and the creator to HidePDF, and carries no document identifier array. Those strings are easy to read with any properties dialog. Be careful about reading too much into their absence, though. Metadata is trivially editable and plenty of tools flatten pages without advertising it, so a file with ordinary-looking metadata may still have been through a flattening pass. Treat the text-selection test as the reliable one and the metadata as a hint.
Does running an already-redacted PDF through a second pass hurt the image quality?
Less than you would expect, for a specific reason. The export always renders at a fixed scale of two, which works out to roughly 144 dots per inch, and embeds each page as a JPEG at quality 0.92. If the file you received came out of this same tool, its pages are already images at that same scale, so re-rendering lands on essentially the same pixel grid and there is no meaningful resize — the cost is a second generation of JPEG compression rather than a resampling blur. If the file came from somewhere else at a different resolution, the picture changes: a 300 dots-per-inch scan gets rendered down toward 144, and that loss is real and not recoverable. For anything that will be printed, or where small type matters, test one page end to end before you commit to the whole document.
Can the recipient tell which redactions were mine and which were the first party's?
Not from the file. Both sets of marks end up as black pixels in the same raster, and the export copies nothing from the original object tree — it builds a new document with only the page images, so there are no annotations, no comments, no form fields and no outline in the result. There is no layer, no author tag and no edit history distinguishing your rectangles from theirs. The new document also gets a creation and modification date stamped at the moment you export, so the file's own dates describe your pass and say nothing about the first one. If the distinction matters, and in a production or disclosure context it usually does, it has to live outside the file: a written log that records who redacted what, on which generation of the document, and under which exemption or rule. Build that log as you work rather than reconstructing it afterwards.