<![CDATA[Purpose: This study examined the effectiveness of ChatGPT as a conversational partner for Indonesian EFL learners by investigating changes in speaking performance, learners’ perceptions, and the association between perceived usefulness and speaking gains. Subjects and Methods: The study employed a quantitative one-group pretest–posttest design involving 40 intermediate-level EFL learners in Makassar, South Sulawesi, Indonesia. Participants engaged in structured ChatGPT conversations for four weeks, comprising three 30-minute sessions per week. Speaking performance was assessed before and after the intervention across fluency, vocabulary, and grammatical accuracy by two EFL instructors. Learners’ perceptions were measured using a five-point Likert-scale questionnaire. Data were analyzed using descriptive statistics, Shapiro–Wilk testing, paired-samples t-tests, Cohen’s d, and Pearson correlation. Results: Speaking performance increased across all measured dimensions. Fluency increased from 5.60 to 7.05, vocabulary from 5.45 to 6.80, and grammatical accuracy from 5.10 to 6.20, while the total speaking score increased from 16.15 to 20.05. All differences were statistically significant (p < .001), with large effect sizes for individual dimensions and a very large effect for the total score (d = 1.77). Learners reported favorable perceptions, with mean scores ranging from 4.45 to 4.70. Perceived usefulness was positively associated with speaking gains (r = .67, p < .001), indicating that more favorable evaluations tended to coincide with greater improvement. However, the findings represent within-participant changes rather than definitive causal effects because no control group was included.]]>
