Effect of AI-driven Conversational Agents on EFL Learners’ Writing Anxiety, Agentic Engagement and Language Investment

Abstract

The purpose of the study was to investigate the impact of AI-powered conversational agents (CAs) on Iranian EFL learners' writing anxiety, active participation, and commitment to language learning. To this end, 73 Iranian male and female EFL learners were randomly chosen from two language institutes in Iran. Following this, the OPT was administered, and a sample of 60 learners was selected and divided equally into two groups: an experimental group and a control group. The data collection process began with administering pretests, which included the second language writing anxiety inventory (SLWAI), the agentic engagement scale, and the language learning investment questionnaire. Participants were given 30 minutes to complete these pretests.  Subsequently, the treatment phase commenced for each group. The study utilized materials from *American English File 1*, which consists of four sections per unit, along with two pages dedicated to practical English and writing, as well as a two-page review and check section. The experimental group was taught using AI-driven CAs. Participants were informed that their performance and class activities would be recorded for research purposes, but the specific aim of the study was not disclosed to avoid the Hawthorne effect. They engaged in sessions facilitated by AI-driven CAs and received feedback on their language errors. After each feedback session, learners were given the opportunity to reflect on their mistakes in preparation for upcoming tasks.  In contrast, the control group did not receive any specialized intervention during the study period. They only received feedback directly from their instructor, as all tasks were corrected manually by the instructor. Upon completing the course, both groups were given posttests, which included the same SLWAI, agentic engagement scale, and language learning investment questionnaire. To analyze the results, descriptive statistics such as the mean and standard deviation for each group were calculated and reported. Additionally, skewness and kurtosis indices were examined to confirm the normality of data distribution. Once the assumptions for conducting parametric tests were met, a one-way ANCOVA was used to compare the two groups. The results indicated that AI-driven conversational agents significantly reduced writing anxiety and enhanced agentic engagement among the learners. Finally, some theortical and pedagogical implications are provided.