Domains such as movie or music, typically addressed in recommendation studies, focus on suggesting items that do not need thorough analysis to estimate their social impact, with the main goal often being to increase sales. Conversely, recommending items with higher social involvement, such as insurances and food, requires the introduction of specific constraints, including socioeconomic, sustainability, and healthiness considerations. In this paper, we address the intricate domain of food recommendation, focusing on analyzing the nutritional intake of food recipes suggested by automatic systems. We base our investigation on a public user-recipe knowledge graph, which embeds information about ingredients, nutrients, and healthiness scores of food recipes. Our experiments highlight the importance of assessing factors beyond accuracy in food recommendation and opens discussions about the complexity in interpreting nutritional intake based solely on raw values.