Generative AI (GenAI) tools, such as Gemini, offer new possibilities for English language learning; however, empirical evidence remains limited, particularly in resource-constrained contexts. This quantitative survey examined how 116 English Education undergraduates at Pattimura University in Eastern Indonesia perceived the usefulness of Gemini and the challenges they encountered in listening, speaking, reading, and writing skills. Students rated Gemini highly for all four skills, with reading receiving the strongest endorsement, followed by writing, speaking, and listening. Text-based features proved more reliable and accessible than audio-based functions. The most valued features included explaining complex vocabulary, improving sentence structure, generating presentation scripts, and creating transcripts from audio materials. However, significant challenges emerged. Academic integrity concerns dominated, with students uncertain about proper citation practices and worried that using AI suggestions might constitute plagiarism or reduce originality. Technical barriers, particularly internet connectivity and file format compatibility, hindered consistent access for many students. Information overload from lengthy responses also created difficulty selecting helpful feedback. The results underscore the importance of establishing clear institutional policies on AI collaboration, providing targeted training in the critical evaluation of AI output, and investing in infrastructure to ensure equitable access to AI. While students recognized Gemini's potential benefits, ethical uncertainties and technical constraints limited its optimal use, highlighting the importance of coordinated support across pedagogical, policy, and infrastructural dimensions.