Preserve intent first. Facts, commitments, and regulatory language remain locked; rewrites adapt rhythm and word choice without drifting meaning.
Transparency beats magic. Editors can diff AI input vs. Truuly output, inspect detector drops, and export reports for reviewers.
Stay on-voice. Style presets clone human cadence instead of defaulting to generic “AI tone,” so releases, LMS submissions, and support macros sound authored by your team—not your model.