How it works

Why PDF text is hard to edit

A PDF page does not store paragraphs or words. It stores drawing instructions: use this font at eleven points, go to this position, draw these characters. There is not even a guarantee the text appears in reading order — a page is free to draw its last line first.

So when a tool offers to edit a PDF, it has to guess the layout before it can change anything. It reads the characters, guesses where the words and lines are, and builds a document from its guess. Then it writes a new PDF from that document. What comes back is a rebuilt page — usually recognisable, frequently not identical, and for anything with a table or a distinctive typeface, obviously not the original.

Why covering the old text causes problems

The other common approach is to leave the page alone and cover it: paint a white rectangle over the old words and draw the new ones on top. This is quick and it looks right in a thumbnail. It also means the document now contains both versions. Select the text and you get the old words. Search it and you find the old words. Open it in a reader that draws the layers differently, or copy it into another document, and the rectangle can move or vanish.

What this does instead

It edits the drawing instructions themselves. The instruction that draws your word is found, the characters inside it are replaced, and the file's own embedded font draws the new ones. Nothing about the page is guessed, because nothing needs to be: the change is made where the page actually stores its text.

Keeping the line in place

Replacing May with June makes that piece of text wider, and everything after it on the line would move right. A line of PDF text already carries small spacing adjustments between groups of characters; the width your edit gains or loses is balanced back out inside the same line, so the character after your edit stays at exactly the position it had.

How each edit is checked

After each edit the new page and the original are both read back, and every character outside the edited region is compared: same character, same position, within a twentieth of a point. The pages are also rendered and compared as images, with only the edited region left out of the comparison.

If anything moved, the file is discarded before it can become something you download. This is why the tool refuses as often as it does: when it says no, that is the check doing its job.

Why it says no

Almost every refusal is one of these:

• The font does not carry the character. Fonts in a PDF usually include only the characters that document already uses — a heading font may carry a few dozen. There is nothing to draw your letter with, and borrowing one from elsewhere would put a second typeface on your page.
• There is no text there. The page is a scan, so it holds an image of words.
• The text cannot be located exactly. Some pages draw a word across several instructions in an order that cannot be matched with certainty. Guessing would risk editing the wrong thing.
• The same drawing is reused on several pages. Editing it once would change all of them, which is not what you asked for.

Against a sample of 150 real-world documents, about four in five could be edited. The remainder were almost all the first case — heading fonts carrying between nine and forty-three characters.