Inline AI highlighter

A verification report tells you what happened. An editor tells you what to do next. Highlighting inside the writing surface removes the step where you read a list of findings, scroll back through your draft, find the sentence and remember which term was flagged. The flag sits where the sentence sits, so the edit is one gesture.

This matters most for the tells that hide at sentence level: a hedge stacked on a qualifier, an announcement connective that carries no meaning, an absolute claim nobody can support. Those are not vocabulary errors you can find with a find-and-replace pass. They are shapes, and a shape is easiest to see while you are rewriting it.

Every mark comes from one of four pattern families, and the colour tells you which family caught it. Turquoise marks stock vocabulary: the twenty or so words a model reaches for when it has nothing specific to say. Amber marks hedges, the qualifiers that soften a sentence until it asserts nothing.

Purple marks announcements — furthermore, moreover, ultimately — the connectives that describe structure instead of doing work. Red marks absolutes: always, never, guaranteed, proven. Each colour is a different instruction. Turquoise usually wants a concrete replacement, amber wants a decision, purple is often deleted outright, and red wants a source or a quieter claim.

Marks find shape, not truth. A sentence can be free of every term in the list and still be wrong, and a sentence can carry three flags while being exactly right. Use the inline pass to fix the prose the reader will judge you on, then take the substance somewhere else: the verification console checks structure, claims and formatting, and a named reviewer covers whatever has consequences.