Artificial intelligence (AI)-assisted translation is increasingly used in tourism and hospitality education to help students produce English texts for international audiences. Previous research has mainly examined grammatical accuracy, fluency, lexical choice, and user perceptions, while the treatment of culture-specific meaning in tourism discourse has received less attention. This study examines the linguistic features and cultural representation found in Gemini-assisted English translations of Balinese tourism texts. It does not seek to establish Gemini’s effectiveness or improvement in students’ independent writing. Using a qualitative descriptive design, the study analyzed seven Indonesian-English text pairs purposively selected from a classroom dataset of twenty student texts, together with student reflections. The seven cases were chosen as information-rich examples representing ritual, agrarian heritage, religious practice, culinary culture, and local philosophy. Analysis was informed by Nababan et al.’s dimensions of accuracy, acceptability, and readability and Molina and Hurtado Albir’s translation techniques. Four main strategies were identified among seven focal culture-specific items: explanation (three items), retention (two items), semantic adaptation (one item), and generalization (one item). The English outputs generally displayed readable sentence organization and tourism-oriented lexical choices. However, because no independent student translation or control condition was available, these features cannot be interpreted as evidence of writing improvement. Textual comparison also showed that some local labels were retained, while others were replaced with broader descriptions. Such changes indicate differences in lexical specificity, but their effects on readers’ cultural understanding require further empirical validation. The findings support using Gemini as a classroom comparison and revision resource rather than as evidence of autonomous translation quality.