People Prefer AI-Written Stories When They Believe They Are Human-Authored, Study Finds
Research indicates readers cannot reliably distinguish between human and AI-generated text and often favor AI narratives when the author's identity is concealed.
Readers frequently rate stories generated by artificial intelligence more highly than those written by humans, particularly when they are unaware of the AI's involvement, according to a new study. The research, which involved over 2,500 adults, found that participants could not reliably distinguish between human-written and AI-generated content.
In one experiment detailed in the study, participants could only identify AI-generated stories 39.9% of the time, a rate worse than random guessing. In another, their success rate was 52%, which is statistically no better than chance. This suggests that public perceptions of artificial intelligence capabilities are falling behind its actual advancements.
Dr. Deena Weisberg, a senior author of the study from Villanova University's Department of Psychological & Brain Sciences, explained that when participants were told an AI-written story was authored by a human, they consistently gave it the highest ratings. "I'd argue that this reveals a bias towards narratives written by real people," Weisberg stated. "We assume creative writing requires uniquely human qualities, such as emotional understanding and lived experience, which leads people to underestimate AI's capabilities."
AI Writing Preferences and Characteristics
The study analyzed participants' preferences for AI-generated narratives, which often scored higher for quality and reader engagement than human-written stories when the AI's authorship was masked. The researchers suggest that AI-generated writing tends to be clearer and more direct, making it easier to process. In contrast, human-written narratives are often more subtle and complex, requiring readers to infer meaning rather than having themes explicitly stated.
This preference for predictability may lead readers to rate AI stories higher, as processing subtle material demands more cognitive effort. The findings align with previous research indicating a preference for AI-generated content in various forms, including short stories, poetry, and essays.
Identifying AI-Generated Text
While readers struggle to identify AI authorship, researchers have noted certain patterns characteristic of AI-generated text. These include repetitive sentence structures, overuse of em dashes, and formulaic phrases such as "it's not just X, it's Y." AI writing may also feature perfect grammar, generic descriptions, balanced arguments, frequent buzzwords, and a consistent tone throughout.
Inconsistencies, such as abrupt shifts in tone or repeated phrases, can also be indicators. AI may sometimes reference specific details without adequate context or produce text that feels basic and formulaic. An excessive use of buzzwords and jargon can signal AI filling knowledge gaps with generic vocabulary. Additionally, very quick response times could suggest AI generation, as human thought processes often involve more deliberation.
The study also found that familiarity with AI systems can help individuals recognize patterns typical of AI-generated writing. This suggests that improving AI literacy could equip people to better navigate an increasingly AI-influenced world. Developers have also created AI detection tools to assist in identifying AI-generated or enhanced text, particularly in academic and professional contexts.