Authors: Vishal Singh, Hemant Singh, Ajay Rawat, Shivam Kumar Jha
Abstract: Nutrition-analysis applications traditionally provide qualitative, binary guidance such as “healthy,” “unhealthy,” or “avoid.” However, recent advances in generative artificial intelligence (AI) enable hyper-personalized, quantitative moderation advice that recommends specific serving sizes, risk thresholds, and actionable alternatives. This paper investigates whether quantitative, personalized recommendations enhance user comprehension, confidence, and dietary intent compared to generic, qualitative warnings. We conduct a randomized controlled A/B user study with 100 participants and compare a qualitative control interface against a quantitative, generative-AI- powered interface offering explicit serving guidance and alternatives. Results show that quantitative moderation advice significantly improves comprehension accuracy, user confidence, trust, and positive dietary intent. These findings provide strong HCI evidence supporting the integration of precise, personalized guidance in digital nutrition applications.
