Readers rated short stories written by ChatGPT as more absorbing and higher quality than human-written fiction, according to a new study published in the journal Judgment and Decision Making. But researchers found a complication: people still judged stories more generously when they believed a human wrote them, even when an AI actually did.
The study involved 1,682 adults, each of whom read one of six short stories. Three were written by people and three by ChatGPT, with each AI-generated piece created around a theme similar to one of the human stories. Participants were told whether the story had supposedly been written by a person or by AI - but the label was not always accurate.
Two standards of judgment
The AI stories generally received stronger ratings for quality and absorption, yet works labelled as human-written were judged more generously regardless of who actually produced them. That split suggests readers respond to two different qualities: the immediate experience of reading a story, and the sense of authenticity attached to the person they believe created it.
Senior researcher Deena Skolnick Weisberg of Villanova University told The Guardian that AI systems could already produce short fiction regarded as comparable to, or even better than, human writing. She stressed, however, that this did not make human authors obsolete.
Writing is not valued solely for the finished product, she said. People also write to express themselves, explore their experiences, and test their creative abilities - purposes that remain meaningful even when a machine can generate convincing prose.
Readers' existing views of AI also shaped their responses. Those who felt more positively towards the technology tended to rate stories identified as AI-generated more highly. This is worth paying attention to for anyone working in content creation, and training resources like AI for Writers Courses can help clarify where machine writing ends and human craft begins.
Can readers tell the difference?
In two additional experiments involving 905 adults, participants were shown pairs of stories with similar themes - one written by a person, the other by AI - and asked to identify which was which. Only about 40% guessed correctly in one experiment, while 52% did so in the other. There was little evidence that people could reliably distinguish machine-generated fiction from human work.
Greater familiarity with AI appeared to help. Weisberg suggested that AI writing may have a recognisable style that readers can learn to identify through experience. Its appeal, she added, could lie in its relative simplicity, making stories easier to follow and digest.
Similar findings have emerged in research on poetry, where non-expert readers rated AI-generated poems more highly and often found them more straightforward than works by established poets. Researchers cautioned, however, that accessibility should not automatically be mistaken for depth or artistic achievement.
The experience of AI drafting a story alongside a human revising it is a workflow many writers are exploring, and AI for Creatives offers practical guidance on those tools.
Limits of the study
The latest study is based on only six short stories, meaning it cannot establish that AI writes better fiction generally, much less that it can sustain the complexity, originality and emotional development required of a novel. The growing role of AI in the arts also faces strong opposition: Luke Kennard, a professor of film and creative writing at the University of Birmingham who was not involved in the research, told The Guardian that machine-generated work could not be separated from concerns about copyrighted training material, environmental costs and the livelihoods of human creators.
Why this matters for writers
Writers who treat AI output as a serious competitor risk misreading the situation. This study suggests that readers judge machine-written stories on two separate axes: how much they enjoy the experience, and how much they value the person behind it. Both matter. If you are professionally invested in the craft of writing, the evidence here points one practical direction: focus on what machines cannot replace - original insight, emotional depth, and your own voice - and build a track record readers recognise. The study also suggests that readers who can identify AI styles may eventually prize human authorship more, not less. The advantage is not in avoiding the tools entirely but in knowing where they genuinely add value to your workflow.
Your membership also unlocks:
