09/07/2026
A scientific reference generated by AI can look perfectly credible and still be completely fabricated.
Researchers asked an early version of ChatGPT to explain a medical mechanism and provide supporting PubMed references. The response sounded convincing. The citations looked legitimate.
But when the references were checked, all five paper titles were fabricated, and the PubMed IDs pointed to unrelated articles.
The researchers observed a similar problem when testing another rare-disease topic: ChatGPT confidently described an association that had not been established in the published literature.
The takeaway for scientific writing remains relevant: fluency is not evidence.
Claims and references generated by AI require independent verification.
Read the full article here: https://hubs.la/Q04wRgJT0
This study evaluated an early version of ChatGPT following its release in 2022.