Humans can detect AI text with 87.6% accuracy, study finds

A new study challenges previous findings that humans struggle to distinguish AI-generated text from human writing. Researchers found that annotators could identify AI text with 87.6% accuracy across nine languages and nine domains. Key differences were noted in concreteness, cultural nuances, and diversity, though explicit prompting could bridge some gaps. Interestingly, humans did not consistently prefer human-written text when the source was unclear. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Challenges assumptions about AI text detection, suggesting potential for improved content authenticity verification.

RANK_REASON Academic paper on human detection of AI-generated text.

Read on arXiv cs.CL →

• arXiv
• Rui Xing
• LLMs

• paper
• safety