Why Word-Level Diff Is More Useful Than Character-Level for Prose
A character-level diff treats every individual letter as its own unit, so changing a single letter within a word (fixing a typo, or changing verb tense) can make the entire word appear partially highlighted in a way that's visually noisy and hard to read at a glance; a word-level diff treats each whole word as the unit of comparison, so a single-letter change inside a word simply shows that one whole word as changed, producing a dramatically cleaner, more readable result for ordinary prose and paragraphs.
This is a deliberate design choice most diff tools built for prose make, as opposed to diff tools built for source code, which often favor a finer level of detail.
Where character-level diff creates visual noise
Comparing "jumps" against "leaps" character by character would highlight a confusing, partial mix of matching and non-matching letters within what a reader intuitively experiences as one whole word being replaced by a different one — word-level comparison instead cleanly marks the entire word as removed and the entire replacement word as added.
Where a finer level of detail is still useful
Source code diff tools often do benefit from a finer level of granularity, since a single-character change (a typo in a variable name, or an operator swap) can be functionally significant in code in a way it usually isn't in ordinary prose — which is part of why code-focused diff tools and prose-focused diff tools often make different default granularity choices.