Clinical Nutrition Open Science (Aug 2024)

Personalized dietary recommendations for obese individuals – A comparison of ChatGPT and the Food4Me algorithm

  • Isabell Agne,
  • Kurt Gedrich

Journal volume & issue
Vol. 56
pp. 192 – 201

Abstract

Read online

Summary: Background: The globally rising prevalence of obesity highlights the importance of implementing effective methods for prevention and treatment. In this context, personalized nutrition (PN) is being discussed as a promising approach. The Food4Me study, a clinical trial investigating the effectiveness of PN interventions, demonstrated that tailoring dietary advice to individual needs is more likely to initiate behavior changes than standard dietary guidelines. However, the accessibility of PN remains a challenge due to financial and structural barriers. ChatGPT, a freely available, natural language processing model published in 2022, might provide a solution to increase the availability of personalized nutrition to a broader population. Methods: To investigate ChatGPT's potential to provide accurate and reliable personalized dietary recommendations, we conducted a comparative analysis, comparing its recommendations with those of the Food4Me algorithm. We selected 20 obese subjects of the German Food4Me sub-cohort and entered their baseline data into ChatGPT (version 3.5). For each subject, a separate chat space was set up, maintaining consistent wording and prompt order throughout all chats to ensure comparability. The prompts were written from a first-person perspective, simulating an authentic scenario of average users asking for dietary recommendations. ChatGPT's responses were compared to the Food4Me feedback reports that had been provided to the selected subjects within the study. Results: ChatGPT may provide suitable personalized dietary advice and holds some noteworthy advantages over the Food4Me algorithm, but still tends to produce recommendations with inconsistent and unpredictable errors, for example, regarding intakes of macro- or micronutrients. Conclusion: Currently, it is not advisable for individuals without nutritional expertise to rely on ChatGPT for personalized dietary recommendations. Nevertheless, ChatGPT remains a promising approach and may develop into a reliable tool for PN with further updates and fine-tuning in the future.

Keywords