Study finds nearly two-thirds of AI-generated citations are fabricated or contain errors.

Study reveals AI-generated citations often fabricated or erroneous, posing risks for scientific research.

Why it matters

  • AI tools like GPT-4o are increasingly used in research but may produce unreliable citations.
  • Accurate citations are crucial for scholarly communication and research integrity.

By the numbers

  • Nearly two-thirds of AI-generated citations were either fabricated or contained errors.
  • 35 out of 176 citations were entirely fabricated.
  • Fabrication rate was 6% for major depressive disorder, 28% for binge eating disorder, and 29% for body dysmorphic disorder.

The big picture

  • AI models struggle more with less familiar or specialized topics.
  • The reliability of AI-generated content varies significantly based on the topic's prominence in training data.

What they're saying

  • Users report similar issues with AI-generated references in their own research.
  • Some suggest that while AI-generated citations are often unreliable, AI can still be useful for initial research when paired with online search tools.

Caveats

  • The study focused on one AI model (GPT-4o) and may not represent other models.
  • Findings are based on a specific set of prompts and topics.

What’s next

  • Further research could explore a wider range of topics and AI models to confirm these patterns.
  • Academic institutions may need to develop new standards for verifying AI-generated references.