A study published in BMJ Open has analyzed five widely used chatbots and their responses to medical questions.
According to the study, around half of the answers were assessed as problematic, while a smaller proportion constituted serious errors.
Only a few questions were rejected by the systems.
The results also showed that performance varied depending on the topic.
Also read: Analysis of 18 studies finds increased risk of stroke among people with chronic pain
Areas with a strong research base, such as cancer and vaccines, produced fewer errors, while nutrition and exercise more often led to imprecise answers, according to The Conversation.
Issues with sources
A central challenge lies in the use of references. The study documents that citations were often incomplete or misleading.
The average completeness of references was around 40 percent, and no chatbot consistently provided accurate source lists.
According to the researchers, this can be particularly problematic, as structured references may create an impression of credibility even when the content is not verified.
Also read: Tired of potatoes? Try this forgotten root vegetable
At the same time, open-ended questions were identified as a significant risk factor, as they more often led to misleading answers.
The role of humans
Other studies support this picture. Research from Nature Medicine shows that chatbots can often generate correct answers, but users reach correct conclusions in less than a third of cases.
Similarly, analyses from JAMA Network Open and Communications Medicine show that models either overlook diagnoses or propagate incorrect concepts.
Overall, the findings suggest that the challenge lies not only in the technology but in the interaction between system and user.
Also read: Tick-borne encephalitis: Understanding TBE and how to stay protected
Chatbots can serve as a supplement to information-seeking, but should not be used as independent medical authorities.
Sources: The Conversation, BMJ Open, Nature Medicine, JAMA Network Open and Communications Medicine.