When hybrid (ChatGPT and teacher) feedback does not outperform teacher-only feedback: evidence from A2 EFL writing in Ecuador
Abstract
This study investigated whether hybrid (ChatGPT-generated and teacher-provided) feedback could yield greater gains in grammatical accuracy than teacher-only feedback among on-site A2-level English as a foreign language (EFL) students at a technical higher education institute in Quito, Ecuador. A six-week quasi-experimental study involved 41 students from two intact classes: teacher-only ( n = 18) and hybrid feedback ( n = 23). Both groups completed six handwritten writing tasks. Initial feedback was provided in English by the teacher or ChatGPT. Students in both groups reviewed this feedback using the textbook, class notes, and teacher clarification in English and Spanish as needed before rewriting their sentences. All students completed the same 18-item Grammar Accuracy Writing Test before and after the intervention. The test assessed verb tense usage, word order, subject-verb agreement, articles, prepositions, and conjunctions. Both groups improved from pretest to posttest. Teacher-only scores increased from M = 7.17 to M = 9.61 and hybrid scores from M = 9.17 to M = 10.78. The unadjusted gain difference (hybrid minus teacher-only) was −0.84 points [95% confidence interval (−2.68, 1.00), p = 0.364]. Analysis of covariance controlling for pretest scores also detected no statistically significant difference [adjusted mean difference = −0.40 points, 95% confidence interval (−2.25, 1.45), p = 0.664]. None of the six grammatical-category comparisons reached statistical significance after Holm correction. Interpretation is limited by the small sample, intact classes with different schedules, repeated use of the same test, and scoring by one non-blinded rater. The confidence intervals include effects favoring either condition, so the nonsignificant results should not be interpreted as evidence of equivalence. The study did not detect a statistically significant advantage for hybrid feedback in this A2 EFL context. Because both conditions included teacher mediation, the findings do not isolate the independent contribution of ChatGPT. Future research should examine longer interventions, larger samples, and how learners interpret and apply AI-generated feedback.