AI stories rated higher than human ones—until readers learn the truth

The study reveals a fundamental mismatch between what readers experience and what they believe about AI creativity. ChatGPT's stories outscored human-written texts on both quality and immersion—a finding that stands even before the labeling manipulation. Yet the moment readers learned an AI wrote the story, their ratings dropped regardless of which text they were actually judging. This suggests readers hold a conviction that AI cannot produce work as good as humans, even when their own experience contradicts that belief.

The researchers propose that AI-generated text succeeds partly because it tends to be smoother, easier to read, and emotionally upbeat—traits that make material easier to process. Literary fiction, by contrast, is often intentionally difficult, asking readers to work for meaning. Perceived ease does not equal literary quality, but it does influence how people rate what they read. The short-story format—just 1,000 words—may also favor AI, since maintaining narrative coherence is a different challenge in that length than across hundreds of pages.

A key factor is the reader's own attitude toward AI. Those with positive attitudes gave higher ratings across the board; when also told the story was AI-generated, their scores rose even further. Skeptical readers showed the opposite effect. This pattern mirrors an earlier finding in AI-generated poems: human judgment of creative work is filtered through prior belief and disposition, not just aesthetic response. Interestingly, self-reported experience with AI systems correlated with the ability to correctly identify a story's origin, while experience with fiction did not—suggesting that familiarity with how AI works matters more than literary expertise when trying to detect its hand.