Letting AI do all the writing makes it worse, a study finds
AI-generated summary A study published in PNAS Nexus in 2026 examined how Kartik Hosanagar and Daehwan Ahn tested 285 students writing 1,000-word stories with ChatGPT 3.5. The study found that stories produced when AI handled all the thinking were rated worse than human-only writing, while other designs scored better.