A new study finds that people rate fictional stories more favorably when they believe they were written by AI, even though the same stories receive lower ratings when attributed to human authors.
The research, which involved 1,682 U.S. participants in its first phase, suggests that people's stated preference for human-written creative work does not reflect their actual judgments when authorship is hidden or misrepresented.
The findings contradict common assumptions about "algorithm aversion" and reveal that AI writing's fluency and positive tone influence reader perception more than preconceived bias against machines.
What happened
Researchers conducted experiments where participants read short fictional stories and rated their quality. In Study 1, 1,682 participants were either told the truth about a story's authorship or given false information; participants rated stories more favorably when told they were AI-generated, regardless of actual origin. In Studies 2 and 3, participants who read unmarked story pairs were asked to identify which was AI-written.
Why it matters
The findings contradict the widespread belief that people prefer human-written creative work. Although previous research showed people claim to dislike AI-generated content when they know its source ("algorithm aversion"), this study reveals that when authorship is unknown or mislabeled, people actually judge AI stories more favorably—suggesting that fluency, positive emotional tone, and ease of parsing in AI writing override skepticism about machine creativity.
What to watch
The research extends prior findings from poetry and expository writing into fictional storytelling, a creative domain where human experience was thought to matter most. The study also examines whether expertise in AI or literature affects people's ability to distinguish the two, with results to come from Studies 2 and 3.
The research investigates a fundamental question raised by the 2023 Writers Guild of America strike: can people actually tell the difference between AI-authored and human-authored creative work, and how do their beliefs about AI shape their judgments? The study focuses on fictional stories specifically because creative writing is perceived as a domain where human experience and insight are particularly important.
Study 1 recruited 1,682 adult U.S. participants through Prolific, paying them $3 for a 15-minute study. The sample included 843 female participants, 809 male, 26 nonbinary or third gender, 2 other, and 2 who preferred not to answer, with a mean age of 45.9 years and a range from 18 to 81 years. Each participant read a single short fictional story and answered questions about its quality and their engagement with it. Crucially, half were told the truth about the story's origin (whether human or AI-written) and half were deliberately given false information—told it was AI-generated when it was human-written, or vice versa. This experimental design allowed the researchers to measure how much authorship labels influenced ratings independent of the actual text.
The article notes that this approach differs from prior work in an important way. Earlier studies found that when participants learn a poem or artwork is AI-generated, they rate it lower on aesthetic and emotional dimensions—a bias so persistent that attempts to counteract it by highlighting AI's impressive capabilities have failed. However, the same research also showed that when authorship is not disclosed, people struggle to distinguish AI from human work and sometimes prefer AI output. The current study tests whether this paradox extends to fictional stories.
Studies 2 and 3 take a different approach: all participants receive two unmarked stories—one human-written and one AI-generated—without being told which is which. They are then asked to identify which story was written by AI and to answer questions about their own prior experience with and expertise in both AI software and fictional literature. This design allows the researchers to test whether expertise in either domain—knowledge of AI systems or deep familiarity with creative writing—affects people's ability to discriminate between the two sources. The article indicates that the full results from Studies 2 and 3 are forthcoming but does not report them.
The study addresses a paradox in how people perceive AI-generated creative work. While previous research documented "algorithm aversion"—where people rate content more poorly when told it is AI-generated—those same studies found that when participants cannot identify the source, they often cannot distinguish AI from human work and sometimes rate AI output as superior. The current research builds on findings from poetry and expository writing to test whether this pattern holds for fictional storytelling, a domain where character complexity and narrative depth have traditionally been seen as uniquely human strengths.
The body notes that AI-generated content benefits from structural advantages: it tends to be more fluent and easier to parse than human writing, and it carries a more positive emotional tone. These features trigger what the study identifies as a "preference for material" with such characteristics, a tendency documented in earlier research on processing fluency. The study's experimental design—where half of participants receive truthful attribution and half receive false attribution for the same stories—isolates the effect of belief from the effect of actual quality, making it possible to measure how much of people's judgment comes from preconception versus genuine textual features.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
The AI news that matters, in one minute each morning.
Sign up free