GenAI in nutritional sciences (GAINS): A systematic review and reporting framework for future research

Abstract

The growing integration of large language model (LLM)-powered chatbots in healthcare has raised interest in their potential to provide nutritional advice. This pre-registered systematic review (CRD42025619448) aimed to synthesize current evidence on the quality of nutritional advice provided by LLM-powered chatbots. We focused on advice related to metabolic diseases, food allergies or intolerances, nutrient intake, and nutrition during pregnancy or lactation. Out of 2469 records found through an extensive search, 13 studies satisfied the inclusion criteria. We conducted standardized data extraction and quality evaluations. While the chatbots in the included studies demonstrate potential as tools for nutrition advice, caution is necessary as they tend to be less effective with complex cases and may occasionally produce incorrect responses. Synthesis of the included studies revealed substantial methodological heterogeneity, particularly in evaluation criteria and study design, which precluded meaningful cross-study comparisons. Key limitations included the frequent use of subjective or poorly defined assessment measures as well as a lack of reproducibility testing. Despite these issues, all but 1 study agreed that while LLM-powered chatbots show potential for supporting nutritional advice, they are not yet ready for unsupervised use. In response to the methodological gaps identified, we propose a reporting guideline for the use of Generative AI in Nutritional Sciences (GAINS) to promote greater rigor, transparency, and comparability in future studies evaluating chatbot-generated nutritional advice.