The Polish-Authenticity Paradox

AI-generated content often scores higher on surface quality metrics, informativeness, consistency, clarity, while simultaneously being more detectable as artificial.

In persona generation studies, AI outputs were rated significantly more informative, consistent, and clear than human-crafted equivalents. Yet participants could still distinguish them from human work. The very polish that makes AI output look “better” can be what gives it away.

This creates a counterintuitive situation: optimizing for quality metrics may actually reduce perceived authenticity. The absence of rough edges becomes its own tell.

Related: 05-atom—flaws-as-authenticity-signals, 05-atom—uniform-confidence-problem, 05-atom—llm-stereotype-defaults