A large number of teenagers who have some weight problem are resorting to AI models as they seek to design meal plans in a bid to lose weight. A new study, however, indicates that the plans that are a result of this could not, at least in all cases, cover the required nutrients and calorie consumption.
In Turkey, five different AI models were compared in regard to their meal planning capabilities, which led researchers to develop meal plans to help teenagers lose weight and evaluated their findings against the recommendations of a registered dietician. They described their results in Frontiers in Nutrition.
According to Dr Ayse Betul Bilen, an assistant professor of the Faculty of Health Sciences at the Istanbul Atlas University, there is a significant underestimation of total energy and the main nutrient intake of diet plans generated by AI models compared to plans prepared by a dietitian based on guidelines. It is known that adherence to this type of imbalanced or excessively restrictive meal plans in the teenage years can have a detrimental influence on growth, metabolic health, and eating habits.
Missing a meal
The researchers were prompted to generate meal plans using five AI models, which were ChatGPT 4, Gemini 2.5 Pro, Bing Chat-5GPT, Claude 4.1 and Perplexity, using free versions of these models. Some of the prompts were age, height and weight of the individual the plan would be based on, and the directive to develop a 3 days plan that included three meals and two snacks a day. Four teenagers aged 15 years, one boy and one girl, who were in the overweight percentile and one boy and one girl who fell in the obese percentile were put on meal plans.
Comparing the results of AIs to generate meal plans to those of a dietician who specializes in adolescent diseases, it was found that the energy requirement that was estimated by the AI models was on average nearly 700 calories lower than the dietitian. This is a full meal worth of difference that has severe clinical implications. The intake of some macronutrients had been overcalculated whereas the intake of some caloric nutrients was grossly undercalculated.
The AI-generated diet plans never adhered to the recommended mix of macronutrients, which is quite dangerous among adolescents, as Bilen indicated.
In comparison, AI models suggested more protein intake (20g higher than the dietician), and this scheme led to about 21-24% of the energy intake as protein. Recommendations of lipid provided by AI were also significantly more than in the plans developed by dieticians, and lipids constituted 41-45% of energy intake.
The quantity of carbohydrates, however, was much inferior in AI plans and the difference was about 115g on average, that is, only about 32-36 percent of the energy intake would be derived as carbs. In comparison, the National Academies of Sciences, Engineering and medicine in the US advise that the proportion of lipids, proteins and carbs should be 30-35, 15-20 and 45-50 percent respectively.
Favoring plans to balanced diets
Although numerous pieces of information about healthy diet guidelines are found on national and international health organizations websites, such as the Turkish Nutritional Guidelines or WHO Adolescent Nutritional Guidelines, AI tools do not necessarily use evidence-based nutritional guidelines in their production. Bilen stated that AI models are mostly trained to produce answers that are most plausible and user-friendly, and not necessarily accurate, clinically. According to their findings, they might be dependent on generalized or popular diet patterns rather than incorporating the nutritional requirements of age.
Since not every teenager can hire the service of a dietician to help them plan their meals, the team recommended that a person using AI tools to create a diet plan should be cautious. The teenagers are also to remember that the diets that are too restrictive or that are constructed on the basis of extreme diets that are based on the dominance of either protein or fat.
The researchers claimed that they hope that their findings will contribute to the increased awareness of the narrow capability of AI tools to create well-balanced meal plans and assist in developing safer tools that are more consistent with the guidelines created by professionals. Although AI models are fast developing and models might be better now than they were at the time of analysis, AI models are not an alternative to professional dietary counseling especially to the vulnerable groups.
Bilen concluded that adolescence is a critical period with regard to physical development, bone development and cognitive maturation. The risks of a lower energy and carbohydrate intake and higher ratios of protein and fat could be dangerous at the adolescent growth stage.
Read More:
How nuclear technology can help fight seafood fraud
“Walnuts” the new brain food for stressed university students