Study: AI Short Stories Outscore Human Writing
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Study published in the journal Judgment and Decision Making had 1,682 adults read one of six short stories—three written by humans and three by ChatGPT, with each AI story matched in theme to a human-authored work
- Participants who read AI-generated stories rated them as more absorbing and of higher quality than those who read human-written stories, while participants simultaneously gave higher ratings to stories they were told were human-authored
- Follow-up experiments with 905 adults asked them to identify which of two stories was human-written: only 40% guessed correctly in one trial (worse than chance) and 52% in the other, with researchers concluding there is no general evidence readers can distinguish the two
- Dr Deena Skolnick Weisberg, senior author from Villanova University, said AI can "already generate short stories that are seen as being at least as good as – if not better than – human-written stories" while arguing human creativity still holds independent value as self-expression
- Greater self-reported expertise with AI systems was linked to more accurate detection of AI-written stories, whereas fiction expertise showed no such connection, suggesting AI style can be learned but literary expertise doesn't help
- Luke Kennard, professor of film and creative writing at the University of Birmingham, called AI an "existential threat" and "predictive code based on a massive stolen database of actual writing" with "catastrophic repercussions" tied to energy-hungry datacenters
- Participants with more positive attitudes toward AI gave higher ratings to stories they were told were ChatGPT-generated than those with a less favorable outlook, revealing that prior beliefs color reading experience
Why it matters: For publishers and writers, the finding that 1,682 readers rated AI stories as more absorbing—and couldn't reliably tell them apart from human ones—suggests a near-term market shift where machine-written content may satisfy readers. The labeling bias (higher ratings for stories labeled "human") preserves a measurable premium for authenticity that authors and editors may need to market explicitly rather than assume.
