This qualitative study examined how EFL students at UIN Walisongo utilize DeepL, an AI translation tool, to manage speaking anxiety and improve their English communication skills. Through semi-structured interviews with 10 out of 172 students from the English Language Education Study Programme, purposively selected based on their Perceptions using DeepL for at least two months, the research identified that speaking anxiety primarily stems from limited vocabulary, grammatical concerns, and pronunciation fears, with heightened anxiety when speaking to instructors versus peers. The findings revealed that students strategically use DeepL as a preparatory tool rather than real-time assistance, employing it for vocabulary kalo enhancement, sentence structure verification, and pre-speaking practice. This strategic usage effectively reduces anxiety levels, increases speaking confidence, and improves linguistic competence by providing scaffolding that bridges the gap between students' current abilities and communication goals. The study demonstrates that AI-powered translation tools can serve as effective anxiety management resources in technology-enhanced language learning when integrated thoughtfully into pedagogical practices, offering valuable insights for EFL educators seeking to address both cognitive and affective barriers in language acquisition.