Evaluating the limits of AI-based nutritional advice for pregnant women from ChatGPT and Gemini
- 17.12.2025
- English Articles
- Jasmin Mössinger
- Lisa Kopp
- Maren C. Podszun
Peer-reviewed / Manuscript (original) submitted: 19 February 2025; revision accepted: 5 June 2025
Introduction
The launch of ChatGPT in November 2022 marked the beginning of a societal transformation driven by artificial intelligence (AI), which is increasingly influencing the field of nutritional sciences. ChatGPT is a generative AI model based on a language model (Large Language Model, LLM), specifically OpenAI’s GPT (Generative Pre-trained Transformer). The model was trained on a large dataset comprising a wide range of sources, including books, scientific articles, websites, and other publicly accessible content [1]. Gemini is an LLM developed by Google DeepMind and was released in December 2023. While an LLM refers to the technology that processes and generates natural language, a chatbot is the application built upon it, allowing users to engage in interactive dialogue with the model. ChatGPT and Gemini are, therefore, chatbots based on LLMs and made accessible through user-friendly interfaces.
In addition to pure text processing, these models can also handle audio, video, and image files. Chatbots do not possess knowledge; instead, they generate text by predicting the most likely next word. However, it is always necessary to verify the accuracy of the information, as texts generated in this way may contain so-called hallucinations. These statements sound plausible, but do not reflect reality. Despite these potential risks, there is growing interest in chatbots both among the public and in scientific research, particularly for health- and nutrition-related questions. A recent preprint presenting a representative study from Germany shows that 30% of respondents have already used a chatbot to obtain nutrition information [2]. The most frequent questions concerned weight loss and disease-specific dietary advice, but there were also inquiries about nutrition during pregnancy. ...
Abstract
Large Language Model (LLM)-based chatbots such as ChatGPT and Gemi- ni, present new opportunities for delivering dietary recommendations. However, particularly for vulnerable groups like pregnant women, their outputs require careful assessment.
In this study, 20 daily meal plans were generated for a fictional pregnant woman using ChatGPT and Gemini, and recommendations for critical foods and dietary supplements were requested. The results were compared with the DGE reference values and the practical guidelines issued by the German network ‘Healthy Start – Young Family’ (Netzwerk Gesund ins Leben).
Both chatbots failed to meet the reference values for carbohydrates, vitamin D, fluoride, and iodine. In addition, Gemini showed deficiencies in energy, total water intake, folate, iron, and calcium. While both tools consistently recommended avoiding raw animal products and alcohol, they fell short of covering all relevant practical guidelines. Although chatbots can assist in generating nutritional information, they show weaknesses in meeting critical nutrient requirements and providing evidence-based recommendations. Consequently, at this point in time, they do not represent a reliable source of information for pregnant women.
