Guideline Compliance of Artificial Intelligence-Generated Diet Plans After Bariatric Surgery: A Cross-Sectional...

Background/Objectives: Artificial intelligence (AI)-based tools are increasingly being used in tailored nutrition management, and evaluating their compliance with guidelines is significant in clinically sensitive areas, including bariatric surgery. This study aimed to investigate the extent to which diet plans recommended by AI models in the early period following sleeve gastrectomy (SG) align with current clinical nutrition guidelines (ASMBS, AACE/TOS). Methods: A total of 360 menu plans were generated using three AI platforms-ChatGPT-4o, DeepSeek V3, and Grok-3-for 40 simulated patients (20 females, 20 males; BMI 32-45 kg/m²) across three postoperative stages: liquid (day 5), puree (day 16), and solid (day 35). The energy and nutrient contents of the menus were analyzed using BeBiS 8.1; an experienced dietitian assessed compliance with the guidelines using a structured checklist. Nutrient intakes and guideline compliance scores were examined using within-patient Friedman tests followed by Bonferroni-adjusted pairwise comparisons. Results: ChatGPT-4o demonstrated the highest overall compliance scores, particularly in the liquid and puréed phases, while DeepSeek produced higher values for several micronutrients. All models showed substantial gaps in essential postoperative recommendations, most notably thiamine and multivitamin supplementation. Conclusions: Although LLMs can generate partially guideline-concordant postoperative diet plans, they consistently omit several critical elements of bariatric nutrition care. These findings indicate that LLM-generated menus may serve as supportive educational tools, and diet planning must be performed under the guidance of a specialist dietitian. This simulation does not assess clinical safety, efficacy, or patient outcomes and should not be used as a substitute for dietitian-led postoperative nutrition care.