2026 Theses Doctoral

The Effect of Multimodal Conversational AI on Job Interview Anxiety and Performance among ESL Students

Alkhayat, Amany

Speaking a second language is often considered one of the most anxiety-inducing skills to acquire. For many ESL learners, limited vocabulary, fear of judgement, and lack of sufficient practice in a crowded classroom can impede their speaking performance. Although the literature has predominantly focused on interventions in general English classrooms, English for Specific Purposes courses, where English focuses on professional communication, remain underdeveloped, with most research focused on writing and reading. This gap is evident in the context of job interviews, which require not only linguistic knowledge but also familiarity with domain-specific vocabulary and discourse competence.

This study employed a quasi-experimental, mixed-method design, where experimental groups practiced job interview questions with a multimodal embodied conversational AI (MECAI) in mixed reality (MR) simulation (Group1), or ChatGPT voice application (Group 2). The control group did not receive any treatment and relied on traditional classroom activities instead. Students (N = 55), ages 18-64 years from a community college participated in the study. All students completed a pretest and were randomly assigned to three conditions. The experimental conditions were assigned to a six-week intervention. After the intervention, all students completed a posttest questionnaire.

Finally, 36 students were interviewed to explore their perceptions of the study intervention. The pre-and posttest measures comprised eight mock job interview questions, where students’ responses were evaluated through a rubric that included seven criteria: fluency, pronunciation, intonation and stress, grammar and sentence structure, vocabulary use, content relevance, and the use of work-based examples. The pretest also included 33 Foreign Language Anxiety Scale (FLCAS) items rated on a 5-point Likert scale and heart rate measures during the pre-and post-mock job interviews. The interaction with AI conversation logs negotiation of meaning (NoM) patterns within the Task-based Language Teaching method (TBLT) were examined using a Computer Mediated Discourse Analysis (CMDA) coding scheme. The reflection interviews consisted of 11 questions examining students’ perceptions of their conditions, the teacher’s role, experimental students’ interaction with AI and the traditional condition experience, and whether each condition contributed to anxiety reduction and improved speaking performance.

Key findings showed that although there was no significant difference between all groups in the pre-intervention mock job interview, FLCAS, or heart rate, both AI groups (MECAI in MR and ChatGPT voice app) showed significant post-intervention improvement, with the MR group demonstrating the most improvement in fluency, vocabulary, and total language performance. However, despite the fact that students in the AI conditions expressed anxiety reduction during reflection interviews, posttests of FLCAS and heart rate demonstrated no statistically significant reduction in physiological or self-reported anxiety levels. The experimental groups conversational logs showed more NoM pattern production than the ChatGPT voice group. Both AI groups helped address gaps in teachers and curriculum knowledge in ESP-oriented pedagogy, where speaking remains underexplored.