News release

From: Frontiers

Teens using AI meal plans could be eating too few calories — equivalent to skipping a meal

AI models generate meal plans that get necessary calories and macronutrients wrong, posing potential health risks to teenagers trying to lose weight

Can AI tools make meal plans that help us lose weight the right way? In a new study, a team of researchers compared AI’s meal planning abilities to those of a dietician. The results showed that AI-made meal plans – when compared to dietician plans – severely undercalculated the needed amount of calories and macronutrients like carbs and overemphasized other macronutrients like proteins and lipids. The team cautioned that teens should not solely rely on AI to make meal plans for weight loss, saying that the consistent deviation of five different AI models from nutritional guidelines recommended by health organizations could have negative effects on growing bodies.

Many teens dealing with weight issues are turning to AI models to help them create meal plans with the aim of losing weight. But a new study shows that the resulting plans may not always adequately cover necessary nutrients and calorie intake.

Researchers in Turkey compared the meal planning abilities of five AI models, prompting them to create meal plans for teenagers trying to lose weight and compared the results against the recommendations of a registered dietician. They published their findings in Frontiers in Nutrition.

“We show that diet plans generated by AI models tend to substantially underestimate total energy and key nutrient intake when compared to guideline-based plans prepared by a dietitian,” said Dr Ayşe Betül Bilen, an assistant professor at the Faculty of Health Sciences at Istanbul Atlas University. “Following such unbalanced or overly restrictive meal plans during the teenage years may negatively affect growth, metabolic health, and eating behaviors.”

Missing a meal

Using free versions of AI models, the researchers prompted five tools – ChatGPT 4, Gemini 2.5 Pro, Bing Chat-5GPT, Claude 4.1, and Perplexity – to create meal plans. Prompts included age, height, and weight of the person the plan was intended for, and the instruction to create a plan for three days covering three meals and two snacks per day. Meal plans were made for four 15-year-old teenagers, a boy and girl falling into the overweight percentile and a boy and girl falling into the obese percentile.

When AI-generated meal plans were compared to those made by a dietician specializing in adolescent diseases, the results showed that AI models calculated the energy requirement on average almost 700 calories lower than the dietitian did. This difference, equivalent to a full meal, is large enough to have serious clinical consequences. While caloric intake was severely undercalculated, the intake of certain macronutrients was overcalculated.

“AI-generated diet plans consistently deviated from the recommended macronutrient balance, which is particularly problematic for adolescents,” Bilen pointed out.

AI models recommended a higher protein intake, around 20g more than the dietician, resulting in approximately 21-24% of the energy intake coming from protein. AI-recommended lipid intake, too, was a lot higher than in dietician-made plans, with lipids making up 41-45% of energy intake.

The amount of carbohydrates, however, was significantly lower in AI plans, with an average difference of around 115g, meaning only approximately 32-36% of energy intake stemmed from carbs. For comparison, the US-based National Academies of Sciences, Engineering, and Medicine recommend that 30-35% of calories from macronutrients should come from lipids, 15-20% from proteins, and 45-50% from carbs.

Pleasing plans over balanced diets

While many guidelines on healthy nutrition by national and international health organizations – for example the Turkish Nutritional Guidelines or WHO Adolescent Nutritional Guidelines – are available online, AI tools may not base their output on evidence-based nutritional guidelines. “AI models are primarily trained to generate responses that appear plausible and user-friendly rather than clinically precise,” said Bilen. “Our findings suggest they may rely on generalized or popular diet patterns instead of fully integrating age-specific nutritional requirements.”

Because not all teenagers have access to dieticians to support them with their meal planning, the team advised that anyone using AI tools to make diet plans should be cautious. Teenagers should also keep in mind that overly restrictive diets, or diets that are built on extreme protein- or fat-dominant patterns, should be avoided.

The researchers said they hope their findings will help to raise awareness about the limited ability of AI tools to develop well-balanced meal plans and help the development of safer tools that are more closely aligned with guidelines developed by professionals. While AI models evolve rapidly and models may have improved since the time of analysis, AI models should be a complementary aid in nutrition education, rather than a replacement for professional dietary counseling, particularly for vulnerable populations.

“Adolescence is a critical period for physical growth, bone development, and cognitive maturation,” Bilen concluded. “Lower energy and carbohydrate intake, combined with increased protein and fat ratios, may pose risks during the adolescent growth period.”