Text Diff Checker

A text diff aligns two versions and reports the smallest set of insertions and deletions that turns one into the other. It does not understand meaning, so a paragraph moved from the top of a document to the bottom shows as a deletion in one place and an addition in another rather than as a move. Knowing this prevents a common misreading, where a reorganised document looks like it was rewritten.

Granularity matters too. A line-level diff marks an entire line as changed even if one character differs, which is why a reflowed paragraph can appear entirely rewritten when only a comma moved. A word-level or character-level diff shows the real change inside the line. When reviewing prose, prefer the finer granularity; when reviewing code, line-level is usually what you want because line boundaries are meaningful.

A large share of confusing diffs come from characters nobody can see. Windows ends lines with carriage return plus line feed while macOS and Linux use line feed alone, so a file edited on both platforms can show every single line as changed while the visible text is identical. Editors that convert tabs to spaces, or that strip trailing whitespace on save, produce the same effect.

Three more invisible culprits are worth knowing. A byte order mark at the start of a UTF-8 file makes the first line differ. Non-breaking spaces pasted from a web page or a Word document look exactly like ordinary spaces and are a different character. And smart quotes, em dashes, and ellipses inserted by autocorrect differ from their typed ASCII equivalents. If two texts look identical but the diff disagrees, one of these is almost always the reason.

The most useful thing a diff does is narrow attention. In a 4,000-word contract, the diff turns a full re-read into a review of six changed clauses, which is both faster and more reliable, because attention holds for six passages and does not hold for 4,000 words.

A few practical habits. Compare like with like: diffing a formatted export against an unformatted original produces noise that buries the real change, so normalise both first. Read deletions as carefully as additions, since removed text is where obligations quietly disappear. Check the surrounding lines of every change, not just the highlighted span, because a change of "shall" to "may" is one word and a complete reversal of meaning. And for anything contractual or regulatory, keep the diff itself alongside the signed version as a record of what was negotiated.