Contra Labs introduced the Human Creativity Benchmark to untangle two signals that surface when designers judge AI output: convergence on shared best practices and divergence reflecting personal taste. Traditional benchmarks collapse these into a single score, erasing the nuance professionals need. By keeping the signals separate, the framework reveals where models must be correct versus where they must be steerable.

The study tapped a network of over 1.5 million independent creatives who have collectively earned $250M. Evaluators across five domains—landing pages, desktop apps, ad images, brand assets, and product videos—rated AI‑generated pieces at ideation, mockup, and refinement stages. Their effort produced roughly 15,000 judgments, exposing which dimensions (prompt adherence, usability, visual appeal) generate agreement and which remain subjective.

Findings show that prompt adherence and usability yield high consensus, while visual appeal stays highly divergent, especially in ad video and brand assets. Models that default to safe, averaged aesthetics fail to support designers who rely on AI for inspiration and rapid iteration. The benchmark therefore offers a practical tool for measuring both correctness and creative flexibility.