PP172 - NUTRITIONAL ADEQUACY OF AI-GENERATED OBESITY MEAL PLANS: A CROSS-CULTURAL REPEATED-MEASURES ANALYSIS

PP172

NUTRITIONAL ADEQUACY OF AI-GENERATED OBESITY MEAL PLANS: A CROSS-CULTURAL REPEATED-MEASURES ANALYSIS

D. N. Cirak^1,*, E. Ozkul Erdogan¹

¹Nutrition and Dietetics, Bahcesehir University, Istanbul, Türkiye

Rationale: Rationale: AI tools are increasingly used for dietary planning, yet their reliability across cultural contexts remains unverified. We examined whether AI-generated weight-management diets differ by platform and country in energy/macronutrient concordance and micronutrient adequacy.

Methods: Methods: 96 meal plans were generated by entering 8 obesity profiles into ChatGPT, Gemini, and Copilot across four countries (Japan, Türkiye, Italy, USA). Content was assessed via ASA24. Concordance was percentage deviation from AI targets; adequacy was the proportion of micronutrient targets met. Repeated-measures ANOVA tested platform, country, and interaction effects.

Results: Results: AI platform significantly affected energy and macronutrient deviation (p<0.001). As shown in Table 1, Copilot had the best energy concordance, while ChatGPT had the highest micronutrient adequacy despite significant energy overdelivery. Platform×country interactions were significant for carbohydrate (p=0.004) and protein (p<0.001). Italy had the highest country-level adequacy (71.6%); Türkiye the lowest (59.7%). Sodium targets were met in 1.0% of plans; saturated fat in 19.8%. Vitamin K and iron were the most frequent deficiencies.

Image:

Conclusion: Conclusions: AI-generated obesity diets vary by platform and culture. Current tools frequently fail safety standards for sodium and micronutrients. Clinical oversight is essential to ensure nutritional adequacy before use in practice.

Disclosure of Interest: None declared