CLOSE MENU
Nutrition articles can offer useful health information, but their language and emotional tone may affect how readers receive the message. A new study by Assistant Professor Ann Gaba and other CUNY SPH colleagues explored whether ChatGPT could help researchers assess those features.
Published in JAMIA Open, the study examined 475 nutrition articles from 11 legacy media publications collected in 2020 including Time, Scientific American, Vogue, and Men’s Journal. Researchers used ChatGPT to evaluate emotional tone and calculate readability using the Simple Measure of Gobbledygook, or SMOG, which estimates the education level needed to understand a text.
The articles averaged an 11th-grade reading level, with differences across publications and topics. Articles about weight loss had the highest average reading level, while those about agriculture and farming had the lowest. Overall, emotional tone was close to neutral. Food insecurity coverage had the most negative average sentiment, while articles about diet, stress, and mental health were the most positive.
“Effective nutrition communication requires attention to both the language we use and the tone of the message,” says Gaba. “AI may help us assess these features more efficiently, giving public health professionals another tool to develop information that is accessible and relevant to the people they want to reach.”
To check the AI-generated results, a human researcher independently assessed a random sample of 47 articles. Average readability and sentiment scores did not differ significantly between the human and AI assessments, although this does not establish that their ratings were interchangeable for individual articles.
The authors caution that the convenience sample and focus on English-language articles limit how broadly the findings apply. Still, the study suggests that AI could help researchers and public health professionals evaluate how nutrition information is presented, informing efforts to better reach their intended audiences.
