This study examines the cultural and discursive gaps that emerge when machine translation systems render Spanish gourmet tourism products into Chinese, focusing particularly on texts involving Iberian ham. While neural machine translation models such as DeepL, ChatGPT, Kimi, DeepSeek and Copilot have made significant progress in lexical accuracy and contextual fluency, they still struggle to convey symbolic, sensory and culturally embedded meanings characteristic of gastronomic discourse. Through a comparative analysis of automatic outputs and translations produced by university students, this paper identifies recurrent patterns of cultural neutralisation, loss of aesthetic tone and weakened promotional discourse in machine-generated versions. By contrast, human translators employ interpretive strategies, such as amplification, domestication and idiomatic recreation, that restore cultural depth and communicative intent. The findings highlight the indispensable role of cultural mediation in translation training and argue for a critical, pedagogical integration of AI to foster intercultural competence and translational creativity in future professionals.