I. INTRODUCTION
Students vary in how they react to the learning activities their teachers provide. Some students work harder, with greater joy, and more strategically. According to Reeve and Tseng (2011), these behavioral, emotional, and cognitive differences are important in predicting students’ learning and achievement. These are the three components that comprise the conventional model of engagement (Abdelhalim & Almaneea, 2025). Cognitive engagement refers to students’ intellectual effort, whereas emotional engagement refers to their affective states when engaging with language learning tasks. Behavioral engagement means observable evidence of cognitive and emotional engagement (Mercer, 2019). It refers to students’ behaviors in learning situations and is often indicative of their cognitive and emotional engagement (Reeve, 2012). The three constructs are interrelated and jointly contribute to student engagement, which has been regarded as a crucial construct in learning activities (Abdelhalim & Almaneea, 2025). It has been well-understood and heavily researched in the educational context (Fredricks et al., 2004).
However, the three-component model of engagement, which primarily focuses on emotional, cognitive, and behavioral aspects, often neglects the crucial role of students actively shaping their own learning experiences. In other words, the conventional tripartite model fails to consider students’ agency in the learning process, as it does not sufficiently reflect how students contribute actively to their classroom (Reeve & Tseng, 2011). According to Bandura (2006), an agent refers to an individual who impacts their own life and functioning consciously. In the realm of education, agency refers to both motivation—the desire to influence one’s learning—and action—the behaviors reflecting their desire. Agentic engagement is characterized by its proactive nature. Students take the lead in interacting with and influencing the learning process, making their learning environment more personally relevant and challenging. Drawing on the self-determination theory framework, agentic engagement has been integrated into the traditional model alongside behavioral, cognitive, and emotional dimensions (Dincer et al., 2019).
As a key feature of the language learning process, engagement is frequently considered one of the strongest influences on successful educational outcomes (Dörnyei, 2001). Particularly, engagement in language learning is increasingly related to educational technology, personalizing students’ learning experiences in innovative ways (Henrie et al., 2015). Among such technologies, the emergence of generative AI tools such as ChatGPT has transformed English language teaching and learning. Providing authentic language experience, ChatGPT can help differentiate students’ proficiency levels (Kohnke et al., 2023). It also expands learning opportunities by making instruction more adaptive, personalized, and student-centered. These affordances can provide timely scaffolding in line with Vygotsky's Zone of Proximal Development to enhance meaningful language learning (Woo & Choi, 2021).
Furthermore, in educational contexts, ChatGPT provides personalized feedback that scaffolds language practice and supports interactive, responsive dialogue similar to conversations with peers or teachers. Generative AI offers real-time feedback matched to context and helps provide low-stress, motivation increasing conditions for students (Kim & Cha, 2025). According to Radford et al. (2019), a conversational platform such as ChatGPT, has great potential for foreign language education. The growing integration of generative AI in language education provides new opportunities for enhancing student engagement (Abdelhalim & Almaneea, 2025).
Beyond addressing gaps in prior research, examining AI-integrated PBL is pedagogically meaningful in EFL education. In exam-oriented EFL contexts, sustaining learners’ cognitive and agentic engagement remains a persistent challenge, particularly for lower-proficiency students who often experience anxiety and limited participation. Integrating generative AI into project-based learning offers an instructional response by providing timely linguistic support, reducing affective barriers, and promoting strategic engagement, while PBL encourages authentic language use, collaboration, and learner autonomy. Investigating how these approaches influence multidimensional engagement is therefore essential for designing more inclusive and engaging EFL learning environments.
In spite of its rapid emergence, the affordances of ChatGPT in language learning contexts have been underexplored. Its use in education is gaining momentum, but relevant research is still rare in English language classrooms (Kim et al., 2024). Empirical evidence of its effectiveness remains limited. Previous studies have been descriptive rather than experimental, which contributed to uncertainty about improving actual learning processes (Kim & Cha, 2025). Only a few studies have included a small number of participants and most have focused largely on the perceived benefits of employing generative AI (Kim et al., 2024; Woo & Choi, 2021). Taken together, the need to establish the validity of using generative AI as a tool for second or foreign language learning is substantial.
Moreover, it is required to study the impact of AI-integrated project-based learning (PBL) on student engagement in English as a foreign language (EFL) education. Abdelhalim and Almaneea (2025) pointed out that although the growing integration of generative AI in language learning has revealed new opportunities for enhancing student engagement, empirical evidence on its effectiveness remains limited, particularly in PBL contexts. Previous research has not examined how AI assists learning processes related to student engagement, specifically across cognitive, affective, behavioral, and agentic dimensions.
This study examines the effects of generative AI-supported PBL learning on EFL students’ engagement, with particular attention to gender and proficiency differences. Although previous studies have suggested that learners’ gender and proficiency influence learning outcomes, empirical findings regarding these factors remain inconsistent (Hsieh et al., 2017; Liu, 2011), and limited attention has been given to their relationship with engagement in EFL contexts (Kim, 2019). To address this gap, the present study focuses on four dimensions of engagement—behavioral, emotional, cognitive, and agentic—based on the framework proposed by Reeve and Tseng (2011). A mixed-methods pre-/post-test design was employed, combining quantitative and qualitative data to examine how AI-supported PBL influences student engagement across different proficiency levels. The research questions are as follows:
1. How does participation in an AI-integrated book-making project influence EFL learners’ behavioral, emotional, cognitive, and agentic engagement?
2. To what extent do changes in engagement differ by learners’ English proficiency level following participation in the AI-integrated book-making project?
3. To what extent do changes in engagement differ by gender following participation in the AI-integrated bookmaking project?
4. How do learners perceive the benefits and challenges of AI-integrated book-making projects, and how are these perceptions reflected across different engagement dimensions?
II. Theoretical Background
1. Project-Based Learning in EFL Settings
Project-based learning has attracted attention as a highly effective teaching method that motivates EFL students by fostering their use of English in real-world scenarios that mirror authentic communicative situations (Imbaquingo & Cárdenas, 2023). PBL emphasizes student-centered, authentic, and meaningful tasks. As noted by Kim (2017), one type of PBL is a book-making project. The book-making project has been viewed as an effective means of cultivating cultural literacy. According to Ellis and Brewster (2002), this project encourages students to actively engage in cultural learning, recognize cultural differences without bias or judgement, and develop critical thinking skills to understand and reflect on their own culture. Through book-making, students may express a desire to meet people from the cultures represented in the books or interact with peers from other countries, indicating successful achievement of cultural educational objectives in the classroom.
A book-making project allows students to participate in the process and freely express their thoughts and feelings. By integrating the four language skills and critical thinking to support comprehensive language development, students create a unique, personalized book through various creative techniques. There have been some research articles and case studies that explore book-making activities within PBL in language and literature education. Literature reviews consistently report benefits of PBL including improved language skills, motivation, autonomy, critical thinking, and collaboration, across EFL contexts both domestically and internationally.
Previous empirical research conducted by Park and Lee (2009) found that English reading skills improved through the English book-making activity. The study also demonstrated that students in the English book-making activity significantly improved vocabulary acquisition. In particular, female students showed greater improvement than male students and mid-level students improved more than both high- or low-level students. In addition, meaningful changes were also observed in affective domains. Positive shifts related to English learning, especially in confidence, interest, and participation, were reported. These findings suggest that as students engage in the English writing process through book-making, they naturally reread their own writing repeatedly. In other words, rereading their own work during the book-making activity may lead to improved English proficiency. By naturally fostering language skill enhancement, students can also develop positive attitudes towards English learning.
Studies conducted abroad have likewise reported positive effects of English book-making projects. For example, Lin and Chang (2009) found that both students and teachers expressed favorable perceptions of an English bookmaking project in a Taiwanese EFL writing class. Students reported positive attitudes, increased motivation, and heightened interest in English learning, while teachers highlighted benefits such as active participation, improved language learning, and enhanced peer collaboration and interaction. Coba and Auquilla (2024) conducted a study in Ecuador to examine the impact of creating digital books (e.g., digital storytelling or book creation) on the productive skills of high school EFL students. A mixed method approach was applied including surveys and journals to document classroom interactions while working on digital books. The findings emphasize the potential benefits of digital book creation as a strategy to enhance speaking and writing skills. This research on creating digital books in EFL revealed that students’ productive skills (speaking, writing, and creativity) improved significantly, along with learning motivation and engagement.
The present study aims to investigate how a project-based English book-making activity affects Korean EFL students’ language learning. While international EFL studies have actively explored the educational benefits of bookmaking projects such as Taiwan (Lin & Chang, 2009) and Ecuador (Coba & Auquilla, 2024), empirical studies conducted in Korean EFL contexts is scarce. Although Park and Lee (2009) examined the impact of book-making activities on English learning, their study focused on elementary school classroom settings, and Kim (2017), despite examining college English classrooms, did not conduct an experimental study. This imbalance reveals a significant research gap in domestic EFL contexts, particularly in higher education settings.
Furthermore, previous PBL research has predominantly focused on the impact of book-making activities on students’ productive language skills, such as speaking and writing (Coba & Auquilla, 2024; Lin & Chang, 2009). Regarding reading, studies have explored PBL’s role in promoting critical reading, suggesting PBL enhances comprehension and develops reflective and contextual reading strategies (Hmelo-Silver, 2004). However, limited attention has been paid to the potential benefits of book-making projects within the context of EFL reading instruction. According to Harputra and Ramadhani (2025), there is a lack of empirical research developing and testing instructional designs combining PBL and AI for reading comprehension. From this point of view, the present study seeks to investigate how a book-making project can support EFL students in a reading classroom. The current study aims to address this gap by providing updated and contextualized insights into the application of book-making projects in Korean EFL classrooms.
2. Student Engagement
Engagement refers to an important educational outcome. As a marker of learners’ positive functioning, it predicts highly valued learning outcomes such as academic progress and achievement (Ladd & Dinella, 2009). A general consensus has emerged to characterize student engagement as a three-component construct: behavioral, emotional, and cognitive (Reeve & Tseng, 2011). Cognitive engagement is defined as learners’ intellectual effort such as focused attention and active self-regulation. It also includes critical thinking or problem-solving, as well as the use of deep learning strategies and sophisticated learning approaches (Abdelhalim & Almaneea, 2025). Emotional engagement indicates the affective states that learners experience when engaging in learning tasks such as enjoyment, enthusiasm, interest, and motivation. These are positive emotions that inform learners’ relationships with their tasks. In contrast, there can be negative emotions such as frustration, anger, anxiety, or boredom, expressing learners’ disengagement and inhibiting their learning experience and performance. Behavioral engagement refers to observable learning behaviors in learning situations, such as classroom engagement and persistence (Reeve, 2012). These behaviors often reflect learners’ cognitive and emotional engagement. As observable evidence of emotional and cognitive engagement, it also refers to on-task attention, effort, and lack of conduct problems (Fredricks et al., 2004).
In classroom settings, it is often difficult for learners to employ sophisticated learning strategies during the ongoing flow of instruction. Learners also face challenges in constructively modifying what is learned or how learning activities are experienced and carried out. From a social-cognitive perspective, Bandura (2006) emphasized that when learners act agentically, they expand their range of available options, enhance their autonomy and self-efficacy, and increase their opportunities for meaningful learning experiences, such as internalization and conceptual understanding.
Reeve and Tseng (2011) defined agentic engagement as “students’ constructive contribution into the flow of the instruction they receive” (p. 2). This concept captures the process in which learners intentionally and somewhat proactively try to personalize and enrich what is to be learned as well as the environment where it is to be learned. During instruction, students can demonstrate agentic engagement by expressing their needs and preferences, seeking support and clarification, contributing ideas, and actively influencing learning goals and activities.
Agentic engagement has been integrated into engagement models alongside behavioral, cognitive, and emotional dimensions (Abdelhalim & Almaneea, 2025). It can enhance teachers’ understanding of how students learn and benefit from their learning experiences (Reeve & Tseng, 2011). Unlike the other three engagement dimensions— emotional, cognitive, and behavioral—agentic engagement stands out due to its proactive nature. Students can take the lead in interacting and influencing their learning process, making their learning environment more personally relevant and challenging. They can express learner autonomy, choice, and active involvement during the project. The students can also take a purposive role in their own learning, making an intentional and constructive contribution into the flow of the instruction they receive (Reeve & Tseng, 2011).
Engagement in language learning has become increasingly associated with educational technology, which personalizes learners’ experiences in innovative ways (Henrie et al., 2015). In this context, agentic engagement plays a critical role by enabling learners to take an active part in shaping their own learning experiences, particularly in technology-enhanced and project-based learning environments. Previous studies have shown that agentic engagement is positively related to academic outcomes and is especially important in learning contexts where students actively contribute to the learning process (Dincer et al., 2019). Recent research has suggested that integrating generative AI into PBL learning can enhance multiple dimensions of student engagement, including behavioral, cognitive, emotional, and agentic engagement (Abdelhalim & Almaneea, 2025).
With advancement of AI technology, AI-driven PBL has supported tailored language learning experiences with contextually relevant language practice and prompt feedback, which can enhance language proficiency and student engagement. It has also been found to increase students’ motivation (Baskara, 2023). Language students have considered AI technology as accurate, effective, useful, and easy to use. They have reported increases in positive perceptions on AI-enhanced language learning while reporting reductions in negative ones. However, several limitations warrant further investigation regarding language learning with AI technologies. In spite of the promising potential of AI-assisted PBL in language learning, there are challenging issues such as cultural biases, inaccurate content generation, inconsistent integration, and reduced teacher-student interaction.
Despite these promising findings, empirical research examining how AI-driven PBL influences student engagement in EFL contexts is still limited. In particular, few studies have systematically examined engagement changes across learner groups and dimensions, or qualitatively explored how ChatGPT is integrated into PBL in EFL settings. To address this gap, the present study examines the effects of integrating generative AI into PBL on student engagement in an EFL classroom.
III. METHOD
1. Participants
The participants of the study were 38 students enrolled in a first-year general English class at a university in Korea. All students belonged to one of four fields of study: Business and Media, Humanities Convergence, Theology, and Global Convergence. All participants were freshmen and were required to take this mandatory English class during the fall semester. It was a 3-credit-hour class conducted over a 15-week semester. During the semester, students completed a book-making project. Before and after the completion of the project, the participants carried out a perception questionnaire that included items on their English learning backgrounds. Out of 38 students, 22 were male, and 16 were female. The participants ranged in age from 18 to 23 years old, with a mean age of 19.20 (SD = 0.93).
After the midterm examination, students’ scores were used to divide them into two proficiency groups: high and low. Based on the midterm results, students whose scores were above the class average were classified as the highproficiency group (n = 18), while those whose scores fell below the average were categorized as the low-proficiency group (n = 20). This grouping was adopted to explore differences in how students at varying proficiency levels perceived the AI-enhanced PBL experience.
In this study, students’ English proficiency levels were operationalized based on their midterm examination scores. The midterm examination was a reading-focused assessment designed to evaluate core comprehension skills, including identifying main ideas, understanding references, making inferences, and answering detailed content-based questions. These components reflect essential reading and language-processing skills that are widely recognized as key indicators of academic English proficiency in EFL contexts (Grabe, 2009). The examination consisted of 50 reading comprehension items.
Importantly, the midterm examination was administered as a common assessment across all general English courses at the university, rather than being limited to a single class. As a result, the test was developed under institutional guidelines that emphasize objectivity, fairness, and consistency across sections, with careful alignment to course learning objectives to minimize instructor-specific bias. Because the exam was standardized and focused on core reading skills, it provided a valid and objective indicator of students’ relative English proficiency within the course.
2. Project Procedures
The study was conducted at a university in Korea during the fall semester of 2025. The general English course was mandatory for first-year students and aimed to improve reading comprehension skills. Typically, students completed four reading passages before the midterm examination and four additional passages afterward. Beginning in the spring semester of 2025, however, one of the post-midterm readings was replaced with a book-making project, a form of PBL. The objectives of the project were also aligned with the broader goals of language education, which emphasize the integration of the four core language skills and the development of critical thinking for comprehensive language learning.
This study is a follow-up study of Kim and Cha (2025); however, it extends the earlier work through a more detailed, methodologically complex design with a broader research focus. While Kim and Cha (2025) examined students’ perceptions of an AI-enhanced book-making project, the present study conceptualizes student engagement as a multidimensional construct, encompassing behavioral, emotional, cognitive, and agentic engagement. Although both studies were conducted in general English courses and involved a reading-based project, the current study employs a mixed-methods pre-post design and examines differences by English proficiency level and gender. These distinctions indicate that the present study extends the earlier work rather than replicating it.
In this study, students created a travel guidebook using AI technology. The project aimed to examine whether authentic and engaging tasks, such as designing a travel guidebook, could promote students’ ability to independently search for and analyze English-language information and support the production of diverse English texts. As the regular reading curriculum was structured across three 75-minute sessions (pre-reading, while-reading, and postreading), the book-making project was designed to follow the same instructional framework. Prior to the project, the instructor thoroughly explained all procedures to ensure that students could concentrate on task completion. The project was implemented over three class periods across two weeks and required active participation during class time; therefore, class attendance was mandatory.
The first session began with group work. Students were randomly assigned to groups of five through the school’s Learning Management System (LMS) and received the task instructions in advance. Each group selected a main theme or goal for its travel guidebook (e.g., “Australia: Where you can take a rest,” or “Paris: Art and Beauty Experience”) and then developed a five-day itinerary, each group member responsible for one day. Students identified key destinations and activities, justified their selections, and incorporated relevant images. During this stage, groups engaged in brainstorming to generate ideas. The completed group work was submitted via the LMS.
The second session focused on individual work. Students created a five-slide PowerPoint presentation based on the itinerary day assigned to them. Slide 1 included basic identifying information and a brief overview. Slides 2-4 each described one destination or activity, with a minimum of six sentences addressing the site description, rationale for selection, and recommended activities, and each slide included an accompanying image. Slide 5 listed all references, including websites, blogs, AI-generated content, and social media posts. All presentation slides were uploaded individually to the LMS.
The third session also involved individual work. For the video production task, students prepared a final twominute video presentation based on their slides. They recorded a video presentation of their travel guidebook content. Students were allotted 10 minutes for final revisions, 35 minutes for presentation rehearsal, and 30 minutes for video recording. The final video, revised slides, and rehearsal recordings were submitted during class.
Most importantly, the current study examined the impact of AI-driven collaborative projects on student engagement in EFL education. To investigate the effects of AI integration, the book-making project required students to use ChatGPT throughout the entire process, including idea generation, content drafting, reformulation of English expressions, summarizing background information, and creating frequently asked questions (FAQs) for foreign tourists. In addition, students used Grammarly alongside ChatGPT for final text refinement. Throughout the implementation of the project, ethical use of AI was strongly emphasized. All AI-generated text and images were required to be properly cited, and students were expected to critically evaluate and revise AI-generated content to ensure accuracy, originality, and alignment with the project theme.
The book-making project process consisted of theme selection, goal setting, research, content development, AI integration, and final production. Students autonomously selected AI tools and managed project outcomes across multiple stages. For example, they used ChatGPT to generate ideas and draft scripts, particularly for presentation dialogues and narrative components. Canva was used to support the creation of visual content, while Grammarly assisted with language refinement and linguistic editing. Text-to-speech tools including Naturalreaders, were utilized to support students’ pronunciation and intonation.
This student-driven project fostered students’ creativity and agency, resulting in outputs such as AI-enhanced slides and video presentations. Students produced presentations featuring AI-narrated visuals and AI-generated illustrations of cultural sites. For example, they used ChatGPT to script historical narratives while Canva to created visual content. These practices demonstrate the effective use of AI tools in promoting cultural awareness and multimodal expression (Abdelhalim & Almaneea, 2025).
Overall, the study seeks to provide updated and contextually grounded insights into the implementation of AIassisted book-making projects in Korean EFL classrooms. By designing a culturally inclusive, tourism-oriented guidebook project supported by generative AI tools, the study contributes to a deeper pedagogical understanding of PBL and AI integration in language education.
3. Data Collection
The data collection and research procedures were carried out over a six-week period during the 2025 academic semester. In Week 7, all participants in the study completed a pre-survey consisting of close-ended questionnaire items designed to measure baseline levels of student engagement (behavioral, cognitive, emotional, and agentic engagement) in English learning. The pre-survey was administered before the implementation of the AI-integrated book-making project.
In Week 8, students took the mid-term examination, which served as an objective measure of English proficiency. Based on their mid-term exam scores, participants were divided into high- and low-proficiency groups for subsequent analyses. This grouping enabled the examination of proficiency-based differences in students’ engagement before and after the intervention.
The AI-integrated book-making project was conducted during Weeks 9 and 10. Students worked collaboratively to design and produce English travel guide books using generative AI tools, including ChatGPT, for idea generation, content development, language support, and revision. The project was structured to encourage active participation, collaboration, and continuous revision, while allowing students autonomy in topic selection and creative decisionmaking. Instructional guidance and scaffolding were provided as needed to support students’ engagement throughout the project.
In Weeks 11 and 12, participants completed the post-survey that included both close-ended and open-ended questionnaire items. The close-ended items assessed changes in student engagement across the four dimensions, while the open-ended items elicited students’ reflections on their learning experiences, perceived benefits, and challenges related to the AI-integrated book-making project. Collecting post-survey data over a two-week period ensured sufficient time for student reflection and increased response completeness.
Overall, this sequential procedure allowed for systematic examination of changes in student engagement over time and facilitated comparisons across proficiency and gender groups, as well as integration of quantitative and qualitative findings.
4. Instrument: Questionnaire
All participants completed pre- and post-questionnaires intended to assess changes in student engagement before and after the intervention. The questionnaire consisted of two sections. The first section gathered demographic information including age, perceived technology proficiency, and frequency of AI tool use for both general and project-based English learning. The second section assessed student engagement across four dimensions—behavioral, emotional, cognitive, and agentic—using items adapted from two validated scales: Language Learning Engagement Scale (Bandura, 2006; Mercer, 2019) and Agentic Engagement Scale (Mameli & Passini, 2019).
There were a total of 25 items: five items for behavioral engagement, four for emotional engagement, eight for cognitive engagement, and eight for agentic engagement. Each item was rated on a 6-point Likert scale from 1 (not all true) to 6 (very true). According to the previous research (Abdelhalim & Almaneea, 2025), minor wording adjustments were made to tailor the items to the specific context of English language learning and AI-supported project work (See Appendix). The questionnaire showed strong Cronbach’s alpha values of .78 (behavioral), .80 (emotional), .80 (cognitive), and .80 (agentic).
5. Analysis
This study employed a pre-/post-test design to examine the impact of AI-driven collaborative projects on student engagement in EFL education. A mixed-methods approach was used, combining quantitative data from structured questionnaires with qualitative insights from student interviews. This research design enabled a comprehensive analysis of both measurable changes and participants’ experiences, strengthening the depth and validity of the findings.
Quantitative data collected through the pre- and post-questionnaires were analyzed using IBM SPSS Statistics (Version 24). To assess within-group differences from pre- to post-treatment, paired-samples t-tests were performed. Between-group comparisons on the post-test were analyzed using independent-samples t-tests. Descriptive statistics, including means and standard deviations, were also reported. Prior to performing the t-tests, assumptions of both normality and homogeneity of variance were verified. Specifically, Shapiro-Wilk test was used to confirm that the mean differences were normally distributed. Q-Q plots also supported the assumption of normality. Levene’s test was also performed to check the assumption of homogeneity of variance. The results indicated that equal variances could be assumed. Cohen’s d was used to determine effect size, following the guidelines proposed by Cohen (1988). 0.2, 0.5, and 0.8 represent small, medium, and large effects, respectively.
For Analysis of covariance (ANCOVA) results, partial eta squared was reported to convey information about the magnitude of the effect. Commonly interpreted as small (0.01), medium (0.06), and large (0.14). ANCOVA was used to compare means across two groups while statistically controlling for the effect of continuous variables, known as covariates. If the groups in this quasi-experimental study were not randomized and differed on a baseline measure, there might have been pre-existing differences. For pre-existing differences, the study controlled for pre-test scores in a post-test comparison. ANCOVA assumptions were verified. The Shapiro-Wilk test revealed normal distribution of residuals. Levene’s test satisfied the assumption of homogeneity of variances. The interaction between the independent variable and the covariate was not significant, confirming the homogeneity of regression slopes.
To examine the construct validity of the 25-item for the pre-test, an Exploratory Factor Analysis (EFA) was carried out using Principal Component Analysis with Varimax rotation. The data met all requirements for factorability, evidenced by a Kaiser-Meyer-Olkin (KMO) value of .761 and a significant Bartlett’s Test of Sphericity (p < .001). Four distinct engagement factors were extracted, explaining a robust 79.86% of the total variance. Likewise, the posttest factor analysis successfully identified four distinct components with a robust 79.86% of the total variance 75.45%. KMO value was .600. and a significant Bartlett’s Test of Sphericity was p < .001. These results indicate that the scale has strong construct validity.
Qualitative data from open-ended questions were analyzed using a content analysis approach to deepen interpretive insight. Based on previous study (Reeve & Tseng, 2011), a coding framework was developed. The four-dimensional model of engagement—emotional, behavioral, cognitive, and agentic—was employed. Additional themes related to the benefits, challenges, and implications of AI integration were also incorporated. The resulting themes were organized and summarized. Inter-rater reliability between two independent authors was established at 0.985, indicating substantial agreement.
IV. RESULTS AND DISCUSSION
1. Whole-Class Results
To assess the impact of AI-integrated book-making projects on student engagement, both within- and betweengroup comparisons were conducted in the current study. First, a paired-samples t-test was performed on pre- and post-tests for the whole class to examine changes over time. As shown in Table 1, there were statistically significant improvements across all engagement dimensions. Among them, the most substantial gain was observed in cognitive engagement (MD = 0.35), followed by agentic engagement (MD = 0.32), behavioral engagement (MD = 0.27), and emotional engagement (MD = 0.27). These results demonstrate the positive impact of AI-integrated projects on student engagement and learner agency. The effect sizes, as indicated by Cohen’s d, ranged from 0.40 to 0.50, reflecting moderate effects of the AI-integrated PBL intervention across all engagement dimensions.
To be specific, for cognitive engagement, the mean score of the pre-test was 4.14 (SD = .91) while that of the posttest was 4.49 (SD = .89). There was a statistically significant difference between the pre- and post-tests for the cognitive engagement (t = -3.057, p = .004). Evidence of cognitive engagement includes critical thinking, deep learning, and problem-solving (Abdelhalim & Almaneea, 2025). Participants in the study increased their intellectual effort like focusing attention and active self-regulation to complete the projects. The findings of the study indicate that EFL students can benefit from AI-integrated projects in terms of cognitive engagement with the use of sophisticated learning strategies.
Regarding agentic engagement, the mean score increased from the pre-test (M = 3.22, SD = 1.11) to the post-test (M = 3.54, SD = .95). The t-test results showed a statistically significant difference (t = -2.488, p = .017), indicating significant improvement over time. Findings indicate that students increased their constructive contribution to the learning process after the treatment. During the flow of instruction, they expressed their likes and dislikes, added personal relevance to the lesson, or requested assistance such as background knowledge, feedback, or a concrete example of an abstract concept more often than before. In other words, students took a more interactive and influential role in class, making their learning environment more personally relevant and challenging (Reeve & Tseng, 2011). This finding confirms the positive impact of the AI-integrated projects on agentic engagement.
For behavioral engagement, the mean score also increased from the pre-test (M = 4.37, SD = .87) to the post-test (M = 4.64, SD = .71). A statistically significant difference was also found between the pre- and post-tests (t = -3.040, p = .004). These results reveal that there was significant improvement in behavioral engagement after the intervention. That means their observable behaviors in learning situations such as on-task attention and persistence of effort increased after getting involved in AI-integrated projects. Overall, the results demonstrate that the projects led to increased classroom engagement, lesson involvement, consistent and sustained participation, and collaborative teamwork among students.
Students also showed significant improvement in emotional engagement (t = -2.905, p = .006), indicating statistically significant difference between the pre- and post-tests. The mean score of the post-test (M = 4.34, SD = .91) was higher than that of the pre-test (M = 4.06, SD = 1.03) reflecting an increase from pre- to post-test. This means that significant improvement was found in students’ emotional engagement after conducting AI-integrated projects. Findings of this study described heightened emotional engagement, indicating the positive impact of the AI-integrated projects on learner engagement. In sum, the project positively affected students’ experiences when engaging in learning tasks.
2. Proficiency Differences in Engagement
1) Within-Group Results
To evaluate the impact of AI-integrated PBL on student engagement by proficiency, paired-samples t-tests were conducted. As shown in Table 2, there were statistically significant improvements only among low-level participants. In all dimensions, t-test results showed significant differences between the mean scores of the pre- and post-tests for the low-level group.
To examine changes over time by proficiency, paired samples t-tests were performed. Among the engagement dimensions, the most substantial gain was found in cognitive engagement (MD = 0.54), followed by agentic engagement (MD = 0.47), emotional engagement (MD = 0.38), and behavioral engagement (MD = 0.36). These findings highlight the positive effects of AI-integrated book-making projects on student engagement, particularly for students with lower English proficiency. Effect size estimates (Cohen’s d) revealed small effects for high-proficiency students (d = 0.19 - 0.36) and moderate to large effects for low-proficiency students (d = 0.59 - 0.72), indicating that the AI-integrated PBL intervention had a substantially stronger practical impact on learners with lower English proficiency.
Specifically, for cognitive engagement, the mean score for low-level participants increased from 3.67 (SD = 0.69) at the pre-test to 4.21 (SD = 0.94) on the post-test. This difference was statistically significant (t = -3.232, p = .004). The results suggest that low-level participants exerted greater intellectual effort during the projects, including sustained attention and active self-regulation. These findings indicate that AI-integrated projects can support lowlevel learners’ cognitive engagement by encouraging the use of more sophisticated learning strategies. One plausible explanation for this finding is that AI tools provided linguistic and cognitive scaffolding—such as idea generation, language support, and feedback—which may have reduced cognitive load and enabled low-level learners to engage more deeply with the task content.
Regarding agentic engagement, the mean score increased from the pre-test (M = 2.93, SD = .71) to the post-test (M = 3.40, SD = .87). The paired-samples t-test results showed a statistically significant difference (t = -2.663, p = .015), indicating significant improvement over time. This result suggests that the low-level students became more proactive contributors to their learning processes following the intervention. During instruction, they more frequently expressed preferences, connected content to personal experiences, and requested support such as background information, feedback, or concrete examples. In line with Reeve and Tseng (2011), these behaviors reflect a shift toward a more interactive and influential learner role. This finding confirms the positive impact of AI-integrated projects on lowlevel students’ agentic engagement. From a pedagogical perspective, AI-integrated projects may have lowered affective and linguistic barriers, thereby empowering low-level learners—who are often more hesitant in traditional EFL classrooms—to take a more active and influential role in shaping their learning (Reeve & Tseng, 2011).
Students with low-level proficiency also improved emotional engagement significantly (t = -2.703, p = .014), indicating a statistically significant difference between the pre- and post-tests. The mean score of the post-test (M = 3.99, SD = .71) was higher than that of the pre-test (M = 3.61, SD = .86). These results suggest that participation in AI-integrated projects enhanced low-level students’ emotional engagement. In particular, students experienced more positive emotional states during the projects, reflecting a more supportive and motivating learning environment. The availability of AI assistance may have alleviated frustration associated with limited language proficiency, thereby enhancing positive emotional experiences during task completion.
For behavioral engagement, the mean score increased from the pre-test (M = 4.02, SD = .73) to the post-test (M = 4.38, SD = .62) and this difference was statistically significant (t = -2.651, p = .016). These results indicate a significant improvement in behavioral engagement following the intervention. Specifically, students demonstrated increased observable learning behaviors, such as on-task attention and persistence of effort, after participating in AIintegrated projects. Overall, the findings reveal that the intervention fostered greater classroom engagement, consistent participation, and cooperative learning among low-level EFL students. This improvement may be attributed to the structured yet flexible nature of AI-supported project-based learning, which allows learners to participate meaningfully regardless of their linguistic limitations.
In contrast, although high-level students showed increases in mean scores across all engagement dimensions, these changes did not reach statistical significance (see Table 2). One possible explanation is that AI-integrated support may have been less transformative for high-level students, as they already possessed sufficient linguistic resources and learning strategies to engage effectively in conventional instructional settings. In sum, these findings suggest that AI-integrated book-making projects served as an equalizing instructional tool, disproportionately benefiting students with lower proficiency by providing scaffolding that enhanced their cognitive, emotional, behavioral, and agentic engagement.
2) Between-Group Results
The current study explored AI-enhanced PBL within the Korean EFL setting to understand how AI integration can effectively support English learning. In particular, this study focused on whether there were any significant differences between language proficiency levels in terms of students’ perspectives on English book-making project. To assess the impact of AI-integrated book-making projects on student engagement, between-group comparisons were conducted. Table 3 demonstrates the independent t-test results of the close-ended questionnaire, comparing high-level students’ perspectives on EFL learning after participating in the book-making project using ChatGPT with those of low-level students.
To examine differences by proficiency level, independent-samples t-tests were conducted on the pre-test scores. As can be seen in Table 3, there were significant differences between high- and low-levels regarding three dimensions, except for the agentic dimension. The results did not confirm initial equivalence but instead indicated initial differences between high- and low-level students. As shown in Table 3, statistically significant differences were found in the two proficiency groups for behavioral, emotional, and cognitive engagement at the pre-test. Specifically, high-level students reported significantly higher levels of behavioral engagement (t = 2.853, p = .007), emotional engagement (t = 3.134, p = .003), and cognitive engagement (t = 3.984, p < .001) than low-level students. In contrast, there was no statistically significant difference was found for agentic engagement at the pre-test (t = 1.699, p = .102), indicating initial equivalence in this dimension.
Considering the lack of initial equivalence for behavioral, emotional, and cognitive engagement, ANCOVAs were subsequently employed on the post-test scores for these three dimensions, using the corresponding pre-test scores as covariates. This was adopted to control for initial group differences when examining post-test outcomes. As presented in Table 4, the ANCOVA results revealed no statistically significant differences between high- and low-level students on the post-test for behavioral engagement (F = 0.459, p = .503), emotional engagement (F = 0.087, p = .770), or cognitive engagement (F = 0.301, p = .587), after controlling for pre-test differences. The effect sizes, as indicated by partial eta squared, were very small (behavioral: ηp² = .013; emotional: ηp² = .003; cognitive: ηp² = .009), suggesting minimal practical differences between proficiency groups.
For agentic engagement, for which initial equivalence was established, an independent-samples t-test was conducted on the post-test scores. As shown in Table 5, no statistically significant difference was found between high- and low-level students at the post-test (t = .957, p = .345). Although the high-level group (M = 3.69, SD = 1.02) scored higher than the low-level group (M = 3.40, SD = 0.87), post-test comparisons by proficiency using independent samples t-tests did not show any significant difference for the agentic engagement. Overall, although high-level students demonstrated higher engagement levels, no significant differences by proficiency level were observed at the post-test across all engagement dimensions.
The present study examined the effects of AI-integrated PBL on EFL students’ engagement across different proficiency levels in a Korean university context. The findings suggest that AI-enhanced book-making projects functioned as an equalizing pedagogical approach, particularly benefiting low-level students. Although high- and low-level students initially differed in behavioral, emotional, and cognitive engagement, these differences were no longer evident at the post-test. This pattern indicates that AI-integrated PBL was especially effective in supporting low-level students, enabling them to reach engagement levels comparable to those of their high-level peers. Instead of widening existing gaps, AI-enhanced book-making projects appear to have served as a compensatory mechanism that fostered engagement growth among students who initially struggled.
With respect to agentic engagement, post-test results showed no differences between proficiency groups; however, this equivalence should not be interpreted as an absence of change. Instead, low-level students demonstrated meaningful growth in agentic engagement over time, effectively closing the initial gap with their high-level peers. Given this, the finding is especially noteworthy because agentic engagement is widely recognized as the most difficult dimension of engagement to develop, particularly for students with limited linguistic resources (Reeve & Tseng, 2011). The results indicate that AI-integrated PBL not only supported participation but also repositioned low-level learners as more agentic contributors, enabling them to express preferences, seek support, and influence the learning process. To sum up, these findings highlight the potential of AI-integrated PBL to foster learner agency and reduce proficiency-based disparities in EFL classrooms.
3. Gender Differences in Engagement
1) Within-Group Results
To assess the impact of AI-integrated book-making projects on student engagement by gender, paired-samples ttests were conducted. As shown in Table 6, statistically significant improvements were found only among female participants. For the female group, significant differences between pre- and post-test mean scores were revealed across all engagement dimensions except agentic engagement. Among the significant dimensions, the largest gain was observed in cognitive engagement (MD = 0.59), followed by behavioral engagement (MD = 0.49) and emotional engagement (MD = 0.47). Overall, these findings suggest that AI-integrated book-making projects positively influenced female students’ engagement across multiple dimensions. Effect size estimates (Cohen’s d) indicated small effects for male students (d = 0.25 - 0.39), none of which reached statistical significance, whereas moderate to large effects were observed for female students (d = 0.69 - 0.82) in behavioral, emotional, and cognitive engagement. These results suggest that the AI-integrated PBL intervention had a substantially stronger practical impact on female students’ engagement.
Regarding cognitive engagement, female participants’ mean score increased from 3.91 (SD = .83) on the pre-test to 4.50 (SD = .76) on the post-test. This increase was statistically significant (t = -2.824, p = .013). The result indicates that female students demonstrated enhanced levels of intellectual investment, such as deeper processing, sustained attention, and active problem-solving during AI-integrated PBL. One possible explanation is that the open-ended, creative, and reflective nature of book-making tasks—combined with AI-assisted idea generation and language support—may have encouraged sustained cognitive involvement, especially for students who tend to engage more deeply with narrative-driven and meaning-focused activities. The findings are aligned with studies showing that female students are more likely to demonstrate deeper cognitive involvement in tasks that emphasize narrative construction, reflection, and sustained meaning-making (Oxford, 2017). In other words, AI-assisted book-making projects in the current study may have amplified this tendency by providing continuous linguistic and conceptual scaffolding, thereby allowing female students to invest greater intellectual or cognitive effort.
With regard to behavioral engagement, there was a statistically significant improvement over time for female students (t = -3.288, p = .005). The mean score increased from 4.04 (SD = .70) on the pre-test to 4.53 (SD = .61) on the post-test. The significant increase suggests that female students were more consistently involved in classroom activities and demonstrated greater effort and sustained participation throughout the AI-integrated book-making projects. Prior research has suggested that collaborative and PBL environments can foster higher behavioral engagement, particularly when tasks emphasize cooperation, organization, and sustained task completion (Gillies, 2016). The significant increase in behavioral engagement among female students in the current study echoes these findings that cooperative PBL environments promote observable participation among female students. The structured yet flexible nature of AI-integrated book-making projects may have provided conditions that facilitated female students’ active classroom involvement, especially when tasks involved structured collaboration and sustained task completion rather than competitive interaction. The integration of AI tools may have further reduced performance pressure, thereby encouraging more consistent participation.
Similarly, emotional engagement among female participants improved significantly following the intervention (t = -2.748, p = .015). The mean score rose from 3.77 (SD = 0.95) on the pre-test to 4.23 (SD = 0.83) on the post-test. This increase indicates that female students experienced more positive affective responses—such as interest, enjoyment, and enthusiasm—while engaging in the AI-assisted book-making tasks. The emotionally supportive nature of creative, AI-assisted projects may have contributed to a more enjoyable and motivating learning experience for the female students. The emotionally supportive affordances of AI, including immediate feedback and reduced anxiety associated with language production, may have contributed to a learning environment in which female students felt more comfortable and motivated to engage with the tasks.
In contrast, a statistically significant difference was not found for the female students’ agentic engagement. Although there was an increase in mean score from 2.91 (SD = 1.07) on the pre-test to 3.34 (SD = 0.89) on the posttest, this change did not reach statistical significance (p > .05). This result suggests that although female participants may have become more proactive in their learning processes, the intervention was not sufficient to produce a statistically reliable increase in agentic engagement. While students participated more actively and positively, they were not consistently encouraged to influence instructional decisions or take control over their learning. Agentic engagement often requires explicit pedagogical support, such as opportunities for choice, negotiation, and coconstruction of learning goals, as well as sustained exposure to learner-centered practices. This finding aligns with prior research emphasizing that agency does not automatically emerge from participation or positive affect alone (Reeve & Tseng, 2011).
For male participants, descriptive statistics showed increases in mean scores from pre- to post-test across all engagement dimensions (see Table 6); however, none of these changes were statistically significant. These findings suggest that although male students may have benefited from the AI-integrated projects to some extent, the magnitude of change was insufficient to demonstrate a reliable effect within the duration of the intervention. One possible explanation is that male students exhibited relatively higher initial engagement levels, resulting in a ceiling effect that limited further gains. As a whole, the gender-based findings indicate that AI-integrated book-making projects may be particularly effective in enhancing cognitive, behavioral, and emotional engagement among female EFL learners, while their impact on agentic engagement and male students’ engagement requires further investigation.
2) Between-Group Results
The current study also investigated whether there were any significant gender differences in terms of students’ perspectives on the English book-making project. Independent-samples t-tests were first conducted on the pre-test scores to assess initial differences in engagement between male and female students. Table 7 demonstrates the results of the independent-samples t-test for the close-ended questionnaire, comparing male students’ engagement in EFL learning after participating in the book-making project using ChatGPT.
The results confirmed initial equivalence between the male and female groups for three engagement dimensions: emotional, cognitive, and agentic. Specifically, no statistically significant differences were identified for emotional (p = .136), cognitive (p = .198), or agentic engagement (p = .144), indicating initial equivalence for these three engagement dimensions. In contrast, there was a statistically significant gender difference was found for behavioral engagement at the pre-test, with male students reporting higher levels than female students (t = 2.097, p = .043). Given the significant pre-test group difference in behavioral engagement, ANCOVA was administered, using the pretest score as a covariate. The ANCOVA results (see Table 8) indicated no statistically significant difference between male and female students in behavioral engagement at the post-test (F = 1.223, p = .276, ηp² = .034).
For emotional, cognitive, and agentic engagement, for which initial equivalence was established, independentsamples t-tests were conducted on the post-test scores. However, post-test comparisons by gender using independent samples t-tests did not show any significant differences across all the three engagement dimensions (see Table 9). As shown in Table 9, no statistically significant gender differences were found for emotional engagement (t = 0.577, p = .568), cognitive engagement (t = -0.077, p = .939), or agentic engagement (t = 1.107, p = .276).
In sum, these results indicate that although some gender differences were present at the pre-test, no statistically significant differences were noted between male and female students at the post-test across these engagement dimensions. When paired-samples t-test results are taken into account together (see Table 6), a more nuanced picture of gender-related engagement emerges. While between-group analyses revealed no statistically significant post-test differences between male and female students, within-group analyses showed that statistically significant engagement gains occurred only among female students. This finding indicates that AI-integrated PBL influenced engagement development differently across genders, differential developmental effects across genders, even when their final engagement levels were similar.
At the pre-test, a statistically significant gender difference was found only for behavioral engagement, with male students reporting higher engagement levels than female students. This finding is consistent with prior EFL research suggesting that male students may initially exhibit higher observable participation, such as classroom involvement and task persistence (Fredricks et al., 2004). However, after controlling for pre-test behavioral engagement through ANCOVA, no statistically significant gender differences were identified at the post-test. The disappearance of this gender difference does not imply a lack of engagement development; rather, it indicates that female students’ behavioral engagement improved to a level comparable to that of male students, resulting in AI-enhanced PBL. In this respect, AI-integrated book-making projects appear to be particularly effective in fostering behavioral engagement among female EFL students.
Similarly, independent-samples t-tests revealed no significant post-test differences in emotional, cognitive, or agentic engagement. Although there were differential within-group gains over time (see Table 9), these results suggest that AI-integrated PBL functioned as a gender-neutral learning environment. It supports comparable levels of deep learning engagement outcomes for both male and female EFL students. As for the lack of gender differences in cognitive engagement reveals that intellectual investment was driven by task characteristics rather than by gender. The book-making project required sustained attention, meaning construction, and the integration of multimodal resources, while AI tools reduced linguistic and cognitive barriers, fostering equitable intellectual involvement. Similarly, the absence of gender differences at the post-test indicates that both male and female students experienced similar levels of interest, enjoyment, and anxiety during the AI-integrated project. Generative AI tools may have contributed to this convergence by providing nonjudgmental, private, and immediate feedback.
To further explain the absence of gender differences in agentic engagement, it is necessary to consider the instructional conditions of the present study. Agentic engagement refers to students’ proactive contributions to shaping instruction (Reeve & Tseng, 2011) and requires explicit opportunities for choice, negotiation, and influence. In this study, AI tools enabled students to seek feedback and personalize content; however, instructional decisionmaking remained largely teacher-directed, which constrained the extent to which students could meaningfully practice learner agency regardless of gender. This finding aligns with prior research emphasizing that agentic engagement does not emerge automatically from technology use or increased participation but instead requires intentional pedagogical design (Lai, 2019).
In summary, the absence of gender differences in cognitive, emotional, and agentic engagement indicates that AIintegrated PBL can foster engagement equity in EFL classrooms. Instead of favoring one gender, the intervention created conditions under which engagement was shaped primarily by task design, scaffolding, and learner-tool interaction. However, the findings also point to the need for more explicit agency-oriented instructional strategies, such as student-driven goal setting, choice of AI tools, and opportunities to influence task parameters. Without such supports, agency may remain constrained regardless of gender.
4. Students’ Open-Ended Questionnaire Responses
1) Perceived Benefits
To examine changes in student engagement with English learning after the AI-integrated project, participants were responded to open-ended questions asking what they found most valuable about participating in the project. Their responses were coded according to four engagement dimensions based on established engagement frameworks (Fredricks et al., 2004). Each response was categorized according to its dominant engagement dimension. The number of responses exceeded the number of participants, as some students provided multiple responses. Table 10 illustrates benefits perceived by students after engaging in the AI-integrated project in the EFL classroom.
Table 10 presents analysis of students’ responses regarding the perceived benefits of AI-integrated PBL, which varied across four engagement dimensions: behavioral, cognitive, emotional, and agentic. Cognitive engagement emerged as the most salient dimension, followed by agentic and behavioral engagement. As can be seen from Table 10, students most frequently reported gains in cognitive engagement, describing sustained effort, deeper processing, and strategic approaches to problem-solving. Many students noted that AI support enabled them to persist when encountering difficult vocabulary or complex sentence structures by facilitating comprehension through repeated checking, revision, and self-monitoring. These experiences suggest that AI functioned as a cognitive scaffold that supported deeper engagement with learning materials. Students also described instances of deep learning in which they were able to comprehend previously unfamiliar sentence structures or effectively apply learning strategies, often with the assistance of AI tools.
In terms of agentic engagement, many students described proactively using AI as a language learning tool. They reported asking targeted questions, verifying their interpretations, and making independent decisions about how to revise or proceed with tasks. This proactive stance was evident in their strategic use of AI tools and help-seeking behaviors. Students reported that they deliberately turned to AI when facing comprehension breakdowns and selectively employed learning strategies (e.g., using AI for interpretation or pronunciation support). A small number of students explicitly referred to taking initiative, exercising choice, or actively shaping their learning process. Such responses indicate an increased perceptions of learner control and learner autonomy, suggesting that AI-integrated environments may create conditions that support agentic engagement.
In line with Reeve (2013), the current study also activates behavioral engagement through self-initiated action. Regarding behavioral engagement, students described increased persistence, repeated practice, active participation, and task completion. Although references to behavioral engagement appeared less frequently than other dimensions, they were mainly associated with active participation in project tasks and collaboration with peers. For example, several students reported practicing pronunciation multiple times until they were satisfied with their performance or rerecording presentation videos to improve clarity and accuracy. These behaviors were described as self-initiated efforts motivated by improvement and personal satisfaction, rather than externally imposed requirements. Overall, these responses revealed sustained effort and involvement during the class projects and team-based activities. The findings suggest that behavioral engagement emerged not merely as task compliance, but as effortful participation driven by perceived learning value, particularly when the students experienced progress or mastery.
The engagement patterns observed in this study align well with the previous engagement theory. For example, Fredricks et al. (2004) pointed out that behavioral engagement includes persistence and effort when positively framed. The presence of positively framed behavioral engagement in this study indicates that AI-integrated PBL supported sustained effort and practice, particularly when students perceived progress or improvement. By transforming behavioral engagement from compliance-based activity into self-directed participation, AI-supported learning reduces barriers, enabling repeated practice without anxiety, which supports behavioral engagement.
Alongside these behavioral outcomes, emotional benefits were also evident. Students frequently described increased interest, enjoyment, motivation, and confidence in learning English. Many of them reported feeling motivated especially when they successfully comprehended English texts, engaged with content connected to their personal interests, or experienced academic achievement through meaningful learning outcomes. Particularly the students reported feeling encouraged when comprehension improved and noted reduced anxiety when AI tools lowered the psychological barrier associated with speaking and writing in English.
2) Perceived Challenges
In this study, students were also asked to describe the challenges or difficulties they encountered during the project through open-ended questions. Their responses were analyzed using a four-dimensional engagement framework adapted from established engagement models (Fredricks et al., 2004). Each response was classified according to its predominant engagement dimension.
Table 11 shows challenges or difficulties perceived by students during the AI-integrated project in the EFL classroom. Perceived challenges were most prominent in emotional and cognitive engagement. Emotional engagement emerged as the most salient category. Students frequently reported frustration when comprehension was hindered and boredom when instructional content lacked personal relevance. They described pressure associated with exam-oriented language learning environments. These negative affective responses often preceded or accompanied disengagement, particularly when students perceived a mismatch between task demands and their interests or abilities. Notably, several students referred to long-term emotional disengagement rooted in prior schooling experiences, suggesting that affective responses to English learning were shaped by accumulated instructional histories. As Fredricks et al. (2004) pointed out emotional engagement strongly influences other engagement dimensions, including cognitive engagement. The prominence of emotional engagement challenges highlights the pivotal role of affect in shaping students’ engagement trajectories and underscores the importance of emotionally supportive and interest-driven instructional design.
Cognitive engagement constituted one of the most prominent sources of perceived challenge. Students frequently reported difficulty comprehending complex sentence structures, unfamiliar grammatical forms, and abstract or specialized vocabulary. Students often expressed feeling burdened when confronted with exam-oriented content. Several students described losing their interest when instructional content was perceived as monotonous or excessively test-focused. They expressed boredom at repetitive memorization tasks. Cognitive overload was also reported when task difficulty exceeded students’ perceived proficiency levels. These cognitive demands often led to frustration, particularly when students felt unable to bridge gaps in understanding despite sustained effort. However, many students also described employing adaptive strategies—such as consulting AI tools, repeatedly analyzing sentences, or linking new information to prior knowledge—indicating that cognitive struggle often co-occurred with active problem-solving efforts.
Behavioral engagement also emerged as a salient source of perceived challenge. Students commonly cited observable learning behaviors such as test preparation and practice routines. Many described feelings of discouragement when engagement took the form of compulsory or repetitive activities, such as memorizing vocabulary or texts for examinations. Several students reported that despite repeated effort, they struggled to retain information, leading to diminished motivation and increased fatigue. Unlike behavioral engagement framed as a benefit, these instances were characterized by effort without perceived payoff, suggesting that sustained participation alone was insufficient to support positive learning experiences when students lacked a sense of progress or autonomy. The findings indicate that behavioral engagement may function as a source of challenge when sustained effort is experienced as obligatory rather than meaningful, reinforcing prior research distinguishing productive persistence from compliance-based participation. Although such behaviors sometimes contributed to learning gains, they were often accompanied by emotional burden, suggesting that behavioral engagement alone did not guarantee positive learning experiences. According to Fredricks et al. (2004), behavioral engagement can be maladaptive when decoupled from emotional or cognitive investment.
Although less frequent than other engagement challenges, difficulties related to agentic engagement were evident in students’ reports of constrained autonomy and uncertainty about their role in learning activities. Some participants expressed hesitation in decision-making or difficulty asserting their ideas during group tasks, particularly when project goals or expectations were unclear. A smaller but notable set of responses also reflected concerns about overreliance on AI, with some students indicating that excessive dependence reduced intrinsic motivation or led to superficial task completion rather than deep engagement. These findings suggest that the mere availability of AI tools does not guarantee enhanced learner agency, especially when instructional structures limit opportunities for meaningful choice or self-direction. As Reeve (2013) pointed out, agentic engagement requires explicit affordances for choice and influence. Without intentional autonomy-supportive scaffolding, learner agency may remain underdeveloped even in AI-supported environments. Overall, these findings suggest that while AI-integrated learning environments can enhance engagement, they may also introduce new challenges, underscoring the importance of instructional design that encourages strategic and reflective use of AI tools.
V. CONCLUSION
The current study examined the effects of generative AI-integrated PBL on Korean EFL students’ engagement across proficiency levels and gender. By adopting a mixed-methods design, the study provided a comprehensive account of how AI-supported book-making projects influenced student engagement across four dimensions: behavioral, emotional, cognitive, and agentic.
The findings demonstrated that although high-level students initially reported greater behavioral, cognitive and emotional engagement, these group differences were no longer evident after the AI-integrated book-making project. Paired-samples analyses further revealed that statistically significant improvements occurred only among low-level students across all four engagement dimensions. These results suggest that AI-integrated PBL functioned as an equalizing pedagogical approach, benefiting students who typically face linguistic and affective barriers in EFL classrooms. The AI tools appeared to provide timely linguistic scaffolding, reduce cognitive load, and lower anxiety, thereby disproportionately enabling low-level students to participate more actively, think more deeply, and exert greater agency in their learning processes.
Gender-based analyses revealed a similar pattern of convergence in engagement outcome. Although male students reported higher behavioral engagement at the pre-test, no significant gender differences were observed at the posttest across any engagement dimension. Moreover, within-group analyses indicated that female students experienced statistically significant gains in behavioral, cognitive, and emotional engagement, whereas changes among male students were not significant. These findings suggest that AI-integrated book-making projects may be particularly effective in fostering engagement among female students, possibly due to the collaborative, reflective, and meaningoriented nature of the tasks combined with the emotionally supportive affordances of AI. At the same time, the absence of post-test gender differences indicates that the intervention promoted equitable engagement outcomes across genders.
Despite these encouraging results, agentic engagement did not show any significant between-group differences by proficiency or gender at either the pre- or post-test. Although low-level students demonstrated significant withingroup gains in agentic engagement, these gains were not sufficient to produce statistically significant group-level differences. This finding indicates that while AI-integrated PBL can support participation, motivation, and cognitive investment, the development of learner agency may require more explicit instructional support. Agentic engagement, which involves learners’ proactive contribution to the learning process (Reeve & Tseng, 2011), is unlikely to emerge automatically from technology integration alone and may necessitate deliberate pedagogical design, such as structured opportunities for choice, negotiation, and learner voice.
The findings of this study offer several pedagogical implications for EFL instruction. First, integrating generative AI into PBL can be a powerful strategy for supporting low-level students by reducing linguistic limitations and enabling meaningful participation in complex, open-ended tasks. In reading-focused EFL classrooms, generative AI can be strategically used to support text comprehension processes, such as vocabulary clarification, sentence-level paraphrasing, and inferential understanding during pre- and post-reading activities. For low-proficiency EFL learners, generative AI can function as an instructional scaffold when its use is pedagogically guided. By embedding AI support within reading tasks, providing structured prompts, guided AI use, and opportunities for reflective revision can support learners’ participation and reduce anxiety, while still maintaining learner responsibility and critical engagement with reading materials in complex PBL tasks.
Second, the results suggest that AI-integrated PBL can contribute to more equitable learning environments by narrowing engagement gaps related to proficiency and gender. Designing projects that emphasize creativity, collaboration, and real-world relevance—such as book-making—may enhance emotional and cognitive engagement for a wide range of learners. In reading-centered EFL classrooms, such projects can be used to extend text-based instruction by requiring students to interpret, summarize, and reorganize reading materials for communicative purposes. For example, tasks that require students to collaboratively design a travel guidebook allow learners to contribute at different levels, such as selecting visuals, organizing content, or drafting simplified descriptions based on assigned reading texts, thereby reducing pressure on linguistic accuracy alone. In particular, such tasks may be effective in fostering sustained engagement among learners who are less confident in traditional, exam-oriented reading-focused EFL settings.
Third, to promote agentic engagement more effectively, instructors should incorporate explicit instructional strategies that encourage learner agency. Within reading instruction, these may include offering meaningful choices in task topics or formats, inviting students to set personal learning goals, encouraging critical questioning of AIgenerated output, and creating opportunities for students to negotiate task procedures or evaluation criteria. For instance, teachers can allow students to choose between summarizing a reading passage as a written guidebook or transforming it into a video-based presentation, select which AI tools to use for text analysis, paraphrasing or revision, or decide how their final products will be evaluated through negotiated rubrics. Without such intentional design, gains in agency may remain limited, even in technology-rich reading classrooms.
Regarding limitations of the study, first, the study was conducted with a relatively small sample from a single university. This limits the generalizability of the findings. Therefore, future studies should include larger and more diverse participant groups across institutional and cultural contexts. Second, the duration of the intervention was relatively short (six weeks) which may have constrained the development of agentic engagement and limited the detection of long-term effects. Although there have been some previous AI-based PBL studies (e.g., Mingyan et al., 2025; Rostam et al., 2024) showing positive results within short period of time under six to ten weeks, longitudinal research is needed to examine whether sustained exposure to AI-integrated PBL leads to more robust and enduring changes in engagement. Third, student engagement was measured primarily through self-reported questionnaire data, which may be subject to response bias. Future research could triangulate self-reports with classroom observations, learning analytics, or discourse-based measures of engagement to obtain a more comprehensive picture. Finally, this study focused on engagement outcomes rather than language achievement; future research should explore how changes in engagement relate to gains in language proficiency and performance.
PDF Links
PubReader
ePub Link
Full text via DOI
Download Citation
Print
