Feature-based: Analyzes statistical properties of text, like perplexity (randomness in word choice) and burstiness (variability in sentence structure). AI models tend to have lower perplexity and consistent burstiness over ~15 words, while human writing shows more variability.