I tested whether AI could maintain a distinct voice across multiple drafts. The results were mixed. Tone drifted when instructions were loose, but improved when I defined structural rules and examples.
It became clear that voice is engineered through constraints. Specific vocabulary, sentence rhythm, and formatting rules reduced randomness. Consistency required intentional design.