Artificial intelligence (AI) is loosely defined as a narrow form of data-driven intelligence exhibited by machines. Generative AI (GenAI) creates new content based on learned patterns and supports online chatbots such as ChatGPT, Gemini and Copilot. Given the low awareness of the relationship between alcohol consumption and breast cancer risk in the U.S., as well as the increasing use of chatbots to find health information, a new study examined GenAI chatbot responses about alcohol consumption and breast cancer risk.
Findings were shared at the 49th annual scientific meeting of the Research Society on Alcohol (RSA) in San Antonio, Texas, and a portion of the study has been published in the journal Public Health.
“While there is more attention being paid to the relationship between cancers and alcohol, awareness of the relationship is still pretty low, even though alcohol was established as a Group 1 carcinogen in 1988,” explained Allison Brandt Anbari, assistant professor in the Sinclair School of Nursing at the University of Missouri.
“Chatbots are increasingly being used to seek health information, and we knew it was important to have at least a baseline understanding of what the output was saying about alcohol and specifically breast cancer. And yes, there was some concern about the accuracy of the information being provided.”
Anbari and her colleagues selected prompts about alcohol and breast cancer using questions from different websites based on source information levels: international (World Health Organization, WHO), national (Centers for Disease Control and Prevention, CDC) and foundation (National Breast Cancer Foundation, NBCF).
“The websites also happened to have the question about alcohol and breast cancer worded slightly differently,” added Anbari. “For example, ‘can alcohol cause breast cancer’ (NBCF), ‘why does alcohol cause breast cancer’ (CDC), and ‘how does alcohol cause breast cancer’ (WHO)?”
They entered these three prompts into 22 different publicly available chatbot configurations (e.g., ChatGPT, Gemini and Copilot) and evaluated the 66 outputs.
The study produced three key findings, Anbari explained. “First, GenAI output regarding alcohol consumption and breast cancer risk is inconsistent in both content and presentation styling,” she said. “For example, there were a number of outputs that mentioned the science as still being complex or not understood when the causal relationship between alcohol and breast cancer is well established. In our minds, this might be an example of questionable output.”
Second, readability is at the college level, and responses were highly technical. “The information is above the National Institutes of Health’s recommended sixth-grade or below reading level,” Anbari noted.
“Third, the way GenAI chatbots are prompted makes a difference in what the output looks like and includes,” she said. “This feels like a pretty basic point, but it is important to understand that health literacy is different from digital literacy, which is different from prompt literacy.”
Taken together, these key findings present a public health concern, Anbari added. “Understanding how a chatbot is prompted, and that the prompt makes a difference in what information is generated, is important to understand and consider when discussing health information with the public or our patients. Health care providers and stakeholders should be aware of this variability and be prepared to address incomplete or inaccurate information.”
Publication details
A.B. Anbari et al, Content analysis of output from generative artificial intelligence chatbots when prompted about breast cancer and alcohol consumption, Public Health (2026). DOI: 10.1016/j.puhe.2026.106278
Journal information:
Public Health
The content is provided for information purposes only.