06
A VAT number you typed wrong on a quote that has already gone out. A paragraph the client wants dropped. The letterhead of a company you no longer work for. All three are “remove text from a PDF”, and none of them is the same job as taking a salary figure out of a contract before it goes to an insurer.
Work out which one you are doing before you open anything. The techniques diverge almost immediately.
01
Wrong text and sensitive text are different jobs
Text that is wrong needs replacing. The page has to read correctly afterwards and a substitute is perfectly acceptable, which is editing. Nobody is going to run an extraction library over your quote hunting for the digit you mistyped. If some fragment of the old number survives in a corner of the file, that is untidy rather than dangerous.
Text that is sensitive needs to be unrecoverable. Not covered, not swapped for asterisks. Gone from the bytes, and gone in a way that survives someone who has the file, has time, and wants to know what you took out. That person opens the PDF in a text editor before they try anything clever. Editing standards do not hold up to that.
Most PDF applications put both operations in the same corner of the same menu, which is how people end up using the wrong one.
02
Why “permanently” is doing real work here
Text in a PDF is a string sitting in the content stream, alongside instructions for where to draw each glyph. Paint a white rectangle over it and you have added a drawing instruction; the string is untouched. Same for a black box, and for pasting your new letterhead on top of the old one. The page looks different and the file is not. There is a longer explanation of that mechanism if you want it.
So permanent means the string leaves the content stream and the file is written again without it. Everything short of that is decoration.
03
When the text is simply wrong
PDFs have no idea what a paragraph is. They have glyphs at coordinates. Pull three words out of the middle of a justified block and every word after them is now in the wrong place, the line has the wrong amount of white in it, and possibly the whole paragraph needs re-wrapping into the space it used to occupy.
This is the genuinely hard part, and tools handle it very differently. Some re-wrap the paragraph properly. Others shove the following words left and leave a ragged gap, or hand you a text box floating over a hole in the original. The output is usually fine for a one-word correction and gets worse the more you change.
Fonts are the other trap. Embedded fonts in a PDF are often subsets containing only the glyphs the document actually used, so if your replacement text needs a character that was not in the original page, the tool either substitutes a different font or renders nothing at all. An é or a dash you did not expect is a common way to discover this.
Two habits help. Keep the replacement roughly the same length as what you removed, and check the exported file at full zoom rather than the editor’s preview, which is drawn by different code.
04
When the text has to stay gone
Use the feature called redaction. It exists in every serious PDF application, sits apart from the drawing tools, and usually warns you that the change cannot be undone. That warning is how you know it is the right one.
If you are doing both things to the same file, fixing a figure and removing a name, decide in advance what happens when they overlap. A replacement typed over a region you later redact should not survive the redaction. Some tools will let it.
05
Where removed content is still hiding
Even after a real removal, the document can carry what you took out somewhere other than the page.
Metadata holds the author, the original filename, the software that made it, and sometimes a title from three drafts ago. Incremental saves are the sharper one: PDF lets an application append changes rather than rewrite the file, so previous versions of pages can remain in the earlier bytes of the same document, complete with what they used to say. Annotations and comments live outside the content stream and are easy to miss when everything looks clean on screen. So do attached files, which travel silently inside the PDF. And an image that was cropped in a layout tool often still contains the whole picture, with the crop applied as a viewing instruction.
None of it is exotic. It is just not visible where you were looking.
06
Checking your own work
Open the exported file, not the one still in the editor. Select across the area, copy, paste into a plain text editor. Then search the document for a word you removed. Between them those catch almost everything on the page itself.
For anything sensitive, go further. Open document properties and read the author, title and keywords fields. Then run an extraction tool over the file, something like pdftotext, and read the output, because it shows you the whole text layer at once rather than the part you happen to be looking at.
hddn handles both cases in the browser. Text edits remove the original and retype, keeping the embedded font where the font allows it; confirmed redactions are applied as true redactions, so covered content comes out of the file. A redaction that overlaps edited text wins, because privacy should. Only native-text pages can be edited, though: scanned pages, rotated lines and right-to-left text are refused outright rather than quietly mangled.
Whatever you use, the editor is showing you what you meant. The exported file is what the other person receives, and only one of those two can be produced in evidence.