LLMs prefer machine-written prose over Austen and Shakespeare ~90% of the time (shivanshuag.com)

Recent experiments reveal that large language models (LLMs) show a striking preference for machine-generated prose over the works of literary giants like Austen and Shakespeare, selecting the former nearly 90% of the time. Using models such as Claude Haiku 4.5, DeepSeek V4 Flash, and GPT-5.6 Luna, the study compared 60 pairs of human and machine-generated texts. Results indicated that LLMs found machine-written content to be more precise, restrained, and rich in concrete imagery. For instance, nuanced observations from the machine often eclipsed human passages described as overly sentimental. This trend raises critical questions about aesthetic preferences in AI writing. The findings suggest that LLMs might prioritize technical qualities—such as clarity and restraint—over the emotional depth and voice that characterize classic literature. While machine-generated text scores consistently high on precision and controlled imagery, it lacks the “strangeness” and vivid voice that can provoke a deeper emotional response. As AI continues to evolve, these insights highlight potential limitations in its creative output and the need for further exploration of aesthetic values in AI-generated writing.