<![CDATA[This qualitative exploratory study examines communication strategies, phonological errors, and pragmatic patterns that English as a Foreign Language (EFL) learners report experiencing when interacting with AI conversational systems such as ChatGPT, Gemini, and Elsa Speak. Data were collected through an open-ended questionnaire completed by 15 students from the English Language Education Study Program at Universitas Negeri Manado, spanning Semester 2 to Semester 12. Responses were analyzed using Error Analysis for phonological aspects, the Cooperative Principle and politeness-based speech act frameworks for pragmatic aspects, and a communication-strategy taxonomy for strategic aspects. Findings reveal three patterns. First, students reported difficulty with interdental fricatives /θ/ and /ð/, unstressed vowel reduction (e.g., comfortable, vegetable), and vowel-length contrasts (e.g., ship-sheep), reflecting first-language transfer. Second, achievement-oriented strategies paraphrase, circumlocution, and appeal for assistance were reported more frequently than avoidance strategies, suggesting AI availability encourages compensatory behavior. Third, students described AI feedback prompting awareness of pragmatic infelicity, particularly excessive directness, leading to increased use of mitigated, hedged forms (e.g., could you, would you mind). Students also reported gains in speaking confidence, reduced anxiety, and concerns about generic feedback and over-reliance. The study suggests AI functions as a rehearsal space complementing classroom instruction.]]>
