Study: AI Stories Outrated Human Writing by 1,682 Readers
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Judgment and Decision Making published a study in which 1,682 adults rated one of six short stories—three human-written, three generated by ChatGPT on matching themes.
- Readers rated the AI-generated stories as more absorbing and of higher quality than the human-written ones, the researchers found.
- Participants simultaneously gave higher ratings to stories they were told were human-authored, suggesting a bias toward human authorship labels over actual content.
- Follow-up experiments with 905 adults found readers could not reliably distinguish human from AI stories—40% guessed correctly in one experiment (worse than chance) and 52% in another.
- AI expertise, not expertise with fiction, was linked to more accurate identification of AI authorship, according to the researchers.
- Dr. Deena Skolnick Weisberg of Villanova University, a senior author, said the AI preference likely reflects easier readability, while bias toward human labels reflects a desire for authenticity.
- Luke Kennard, a professor of creative writing at the University of Birmingham not involved in the study, dismissed AI writing as "predictive code based on a massive stolen database" and called it an "existential threat."
Why it matters: Over 1,600 readers rated ChatGPT's stories as more engaging than human-written ones, yet couldn't reliably tell them apart—40% accuracy was worse than chance. That gap—AI output is both preferred and undetectable—pressures publishers and educators to confront disclosure and what readers are actually valuing in fiction, while critics flag the ethics of training data.
