The integration of AI in education, particularly in language learning, has gained substantial momentum over the past years. Generative tools such as ChatGPT, Grammarly, and QuillBot offer real-time language support and personalized feedback, presenting new possibilities for enhancing students’ writing competence. In the context of EFL, such tools can provide linguistic scaffolding, increase learner engagement, and foster greater writing fluency (Xu & Wang, 2023; Sun, 2022). However, the effectiveness of these technologies in educational settings depends not only on their functional capabilities but also on learners’ willingness to adopt and use them.

The TAM has been extensively applied to examine user adoption of emerging technologies. It identifies perceived usefulness (PU) and perceived ease of use (PEOU) as primary determinants of attitudes toward technology and subsequent behavioral intention (BI) to use it. Extensions of TAM have incorporated additional constructs, including social influence (SI), the perceived expectations of significant others such as peers, instructors, or societal norms. Other factors include (BI) an individual’s conscious plan to engage in a technology-related behavior; self-efficacy, the belief in one’s ability to use the technology effectively; and facilitating conditions, the perceived availability of organizational, technical, or environmental resources that enable usage. (Venkatesh & Bala, 2008).

In the context of EFL academic writing at Korean universities, these variables offer a valuable analytical lens for understanding generative AI adoption. Recent research applying the extended TAM to ChatGPT adoption among Korean EFL learners found that PU significantly predicted BI, indicating that students who perceive AI tools as capable of improving writing quality and linguistic accuracy are more likely to use them. While PEOU did not directly predict BI in this study, it was influenced by factors such as output quality and playfulness, suggesting that ease of integration remains a relevant enabler (Hwang et al., 2025). SI is particularly salient in Korea’s collectivist and hierarchical academic culture, where instructor and peer approval can strongly affect adoption decisions; this effect has also been observed in EFL contexts using other technologies, such as smartphone translators (Park & Kim, 2024). BI captures students’ readiness to incorporate AI into their writing practices, while self-efficacy influences their confidence in applying AI tools for language learning, a relationship supported by prior research on mobile-assisted language learning in Korea (Lee et al., 2016). Finally, facilitating conditions, such as institutional access to AI platforms, training programs, and supportive policy environments, align with the enabling factors identified in extended TAM models, including TAM2 and the Unified Theory of Acceptance and Use of Technology (UTAUT) (Venkatesh et al., 2003). Together, these constructs provide a robust framework for analyzing how Korean EFL students evaluate and integrate generative AI in academic writing contexts. In educational research, the TAM has been applied to investigate the adoption of learning management systems (e.g., Moodle), mobile learning apps, and digital writing tools (Teo, 2011; Alshammari et al., 2016). However, relatively little attention has been given to its application in the context of generative AI tools for academic writing.

Recent studies indicate that learners’ acceptance of this varies depending on factors such as language proficiency, prior experience with technology, and instructional context. For instance, Lee and Lin (2023) found that EFL students who perceived AI writing tools as useful for improving grammar and structure showed higher adoption rates. Similarly, research by Al-Khateeb (2023) emphasized the importance of ease of use, noting that complex interfaces and unfamiliar functions could deter student engagement. These findings underscore the relevance of TAM in examining user attitudes toward generative AI in EFL settings.

Moreover, ethical concerns and pedagogical implications are emerging as critical dimensions in discussions about AI in writing instruction. While some educators view generative AI as a supplementary tool to enhance language learning, others worry about issues related to academic integrity, over-reliance, and the potential erosion of learners’ critical thinking and writing skills (Bai & Li, 2023; Yoon, 2024). This complexity highlights the need for empirical research that not only assesses the functional acceptance of such tools but also considers learners’ perceptions, attitudes, and concerns in authentic classroom environments.

Although the TAM framework has been extensively validated in technology-related educational research, its application to generative AI usage in EFL academic writing classes remains underexplored. This gap presents an opportunity to deepen our understanding of this area, what influences their acceptance, and how these technologies can be effectively and ethically integrated into academic writing instruction.

The concept of applying the TAM to technology used in English writing language instruction has been visited in a number of previous studies. One done in Indonesia (Ardiningtyas et al., 2024) examined how 30 students in an essay writing course viewed generative AI through the lens of the TAM. This project concluded that over 70% of the respondents found the technology both easy to use and effective in improving writing ability. There were also strong indications they would continue to use it in the future; however, concerns were also raised regarding potential negative impact on critical thinking and creativity. Another study by Aksakallı & Daşer (2025) explored the use of ChatGPT in EFL writing courses in accordance with the TAM. Eight hundred and seventy-four undergraduate students at a university in Turkey took part in a quantitative survey with results showing while perceptions of this technology were positive, it was rarely used. Furthermore, gender-based analysis showed differences in frequency of usage, but not in overall perceptions. In Taiwan, Tan (2024) employed the TAM to explore the perceptions of 256 undergraduate students enrolled in EFL writing courses. This study concluded that once again while adoption rates, PU, and PEOU were high, there were also concerns regarding ethics, creativity, and ownership of the learning process meaning educations programs for both teachers and students how to use generative could be useful.

Specific to the Korean university context, with writing courses often a required component of general English programs or major-specific coursework, many students struggle with formal writing conventions, argument structure, and coherence. For these learners, generative AI can offer timely support, particularly in areas such as grammar correction, paraphrasing, and vocabulary enhancement. Given the high-stakes nature of university assessments in Korea, students may view these tools not only as learning aids but also as performance enhancers, which could significantly affect their motivation to adopt and use them. Lee, Davis, and Lee (2024) found that Korean university students recognized both the strengths and drawbacks of AI-based writing tools, noting improvements in writing accuracy but cautioning against overreliance. Hallemans (2023) demonstrated the usefulness of TAM in understanding ChatGPT acceptance in terms of general English learning among Korean university students, highlighting external factors, PU, PEOU, attitudes, behavioral intention, and actual use. More specific to writing, Copeland & Franzese (2021) used data from 97 undergraduate students enrolled at Korean university. Using the TAM framework, they evaluated students’ perceptions of using virtual learning environments to aid with EFL writing. Their results showed higher levels of PU and PEOU generally resulted in greater willingness to continue to use this technology. However, this study was done prior to the development of generative AI in its current form and further research into this area could be beneficial.

Beyond the cognitive and technical dimensions captured by the TAM, broader societal, emotional, and practical factors increasingly shape how students engage with generative AI in English writing instruction. Emotional responses such as anxiety, perceived threat, or overreliance can influence attitudes toward adoption, particularly when students feel uncertain about the ethical implications or long-term impact on their creativity and autonomy (Alessandro et al., 2024). Societal norms and trust in institutional governance also play a mediating role, as students may be more inclined to adopt AI tools when they perceive clear guidelines and responsible integration into curricula (University of Melbourne and KPMG, 2025; Shrivastava, 2025). Practical considerations such as time constraints, academic pressure, and the need for immediate feedback further reinforce the appeal of generative AI, especially in high-stakes environments like Korean universities. Studies have shown that while perceived usefulness and ease of use remain central, factors like self-efficacy, perceived enjoyment, and compatibility with existing learning habits significantly enhance behavioral intention to use these tools (Ursavaş et al., 2025). Thus, a more holistic understanding of technology acceptance in writing instruction must account for these intersecting layers of influence.

Although the TAM framework has been extensively validated in technology-related educational research, its application to generative AI usage in EFL academic writing classes in Korea remains underexplored. This gap presents an opportunity clarify how EFL learners interact with generative AI, what influences their acceptance, and how these technologies can be effectively and ethically integrated into academic writing instruction, particularly within specific educational contexts such as Korean universities. Thus, this study will attempt to answer the following two research questions:

This study was conducted at a private 4-year university located in the greater Seoul area. The participants were all students enrolled as undergraduates in the software convergence department. As part of their graduation requirements, all of them need to pass an academic writing course. The participants were divided into 4 separate 2 credit classes all of which were taught by one member of this research team. Sections 1 and 2 were taught in-person and sections 3 and 4 were taught online. Each offline courses had 45 students enrolled, and each online course had 70 students enrolled. Classes were held once per week for two hours for the 15-week semester. The in-person courses were taught on campus and consisted of teacher-led lectures, individual practice activities, pair and group, and class-based discussions. The online courses followed an asynchronous approach with pre-recorded videos and prepared materials being uploaded to the university learning management system (LMS) every Monday morning for students to access. They then had 1 week to review the lesson and submit the required work. All students, regardless of teaching approach, used the same materials, submitted the same homework assignments, and took the same mid-term and final examinations.

From a total pool of 230 students, one hundred and eleven voluntarily responded to a request to provide data for this project. Although the response rate was low at 48%, the researchers concluded that 111 respondents was sufficient enough to give a true representation of the total population. Five initial questions focusing on the participants’ demographic information showed that 80 (72.1%) were male and 31 (27.9%) were female. Slightly over half of them (56, 50.5%) said they were 23 or older, 14 (12.6%) said they were 22, 34 (30.6%) were 21 and of the remaining 7, 5 (4.5%) were 20, and 2 (1.8%) were 19. This academic writing course was not offered to freshman students; thus, the majority were sophomores (89, 80.2%), followed by juniors (13, 11.7%) and seniors (9, 8.1%). Of the 111, 70 (63.1%) took the course in-person, and the remaining 41 (36.9%) took it online. Prior to enrollment, no form of level testing was used and all four courses should be considered as being of mixed ability.

The primary form of data collection, done between June 1st and June 20^th, 2025, employed a quantitative approach using questionnaire developed by the research team. Following on from five initial demographic questions, participants were asked to respond to a series of 20 statements rated on a 5-point Likert scale. They explored participants’ experiences using generative AI tools for classroom and homework assignments, as well as their perceptions of how such tools may influence the future of EFL writing instruction. Specific areas of inquiry included perceived benefits and drawbacks, the role of generative AI in completing homework, and broader pedagogical implications. Consistent with the recommendations of Evans and Mathur (2005), the use of a questionnaire was deemed advantageous for several reasons. Primarily, it allowed for flexibility in administration, as responses could be collected asynchronously without restrictions related to time or location. Moreover, this approach facilitated the efficient collection of a relatively large dataset, making it an appropriate method for the objectives of this study.

A QR code and a hyperlink were generated and used as a method of sharing the questionnaire. For the two in-person classes, they were shown on a PowerPoint slide towards the end of the lesson in the 14th week of the semester. For the online courses, they were included in the week 14 video and also shared via the LMS. Students were invited to participate voluntarily and were informed that their responses would remain anonymous and have no impact on their course grades, GPA, or any other aspect of their academic standing.

A Cronbach’s Alpha test was conducted separately for both the PU and PEOU constructs. For the 7 items relating to PU, a score of 0.91 indicated excellent internal consistency. The 4 items concerned with PEOU produced and score of 0.87 signaling very good internal consistency. To promote clarity and comprehension, all survey questions were presented in both English and Korean. The original items were written in English and then translated into Korean using a reverse translation method. To ensure linguistic consistency and accuracy between the two versions, a bilingual colleague from the research team verified the translations. The final question asked participants for consent to include their responses in this research to which all responded positively.

In addition to the quantitative data, a secondary qualitative element was also used to identify any possible trends and provide further insight. At the beginning of the course, students were informed they would be allowed to use generative AI in their lessons and to assist with homework assignments. One caveat of this was for them to explain how they decided to use it, and if and how they felt it was beneficial prior to submission of two homework assignment during the semester the students were asked to write short responses to the following questions:

At the conclusion of the semester, each student was required to hand in an academic style essay using the skills and techniques covered during the semester. Prior to submission, they were required to answer the following two questions:

The analysis was conducted in two distinct phases. Initially, inferential statistics analysis alongside a descriptive statistical approach was employed to summarize and examine the responses for the quantitative data. Measures such as means (m), standard deviations (SD), and frequency distributions were calculated to identify general trends, central tendencies, and the overall distribution of responses. In addition, an analysis of variance (ANOVA) test was employed to determine whether statistically significant differences exist between the means of different groups to assess whether at least one differs significantly from the others. In the present study’s context, ANOVA is appropriate for examining whether variables such as perceived usefulness, perceived ease of use, or behavioral intention toward generative AI in EFL academic writing vary significantly across different demographic or academic groups. This helped to highlight recurring patterns and key themes, their perceived benefits and drawbacks, and their anticipated impact on EFL writing instruction.

Following this, the second phase involved a qualitative thematic review of the data. The researchers independently examined the aggregated results, noting salient themes that emerged across items. Through this process, recurring response patterns were further categorized and interpreted in relation to the study’s research questions. To ensure minimize individual bias, the research team held a series of collaborative meetings to compare interpretations, discuss discrepancies, and reach consensus on thematic classifications. This two-phase process allowed for a nuanced understanding of the data, integrating both quantitative trends and qualitative insights to address the study’s aims.

Whilst employing several of the more traditional constructs associated with the TAM (PU; PEOU; BI; SI), this study also attempted to integrate a number of other elements to gain a deeper insight into the research questions. To understand the prevalence of the use of generative AI in this context, actual usage (AU) was examined. Furthermore, the emotions associated with the use of this technology such as pleasure or guilt were explored under an affective component (AC) construct. The two other areas that this study focused on were how much trust the students had in using this technology and their external beliefs (EB) on how it might affect EFL academic writing courses of the future.

In terms of the AU construct, both the quantitative data and qualitative data indicate overall high levels of usage. A mean of 4.44 and SD of 0.81 for question 10 indicates most students have actively used generative AI in some form despite a few outliers. The same trend was found when looking at the responses to the qualitative data, with over 90% saying they used generative AI to help with their given assignments. A deeper dive in the quantitative data reveals no statistical differences in the AU construct based on gender and/or the format of the class. A two-way ANOVA was conducted to examine whether actual usage differed by gender, class format (online vs. in-person), or their interaction. Results showed no statistically significant differences for gender (p = .27), class format (p = .16), or the interaction between the two (p = .54). While mean usage scores varied slightly among groups, these differences were not meaningful, suggesting that students, regardless of gender or class modality engaged with generative AI at similar levels.

The PU was evaluated across seven items (Q1:Q5, Q19, Q20) measuring performance enhancement, productivity, effectiveness, helpfulness, and tool efficiency. Table 1 shows the mean and SD for each individual item.

With an overall mean of 4.31 and SD of 0.92, these results suggest participants generally perceived generative AI as useful. However, the item regarding belief that generative AI is the only tool needed for improvement (Q20) received the lowest (M = 3.85, SD = 1.06), indicating more variability in perception and a higher degree of caution among students regarding sole reliance on AI. This was also reflected in the responses to the questions students answered prior to submitting their final essays. A significant number felt that AI is only helpful when used in conjunction with weekly lessons. The general nuance of these comments showed little variation depending on whether the course was taken online or in-person. A selection of them is shown in Table 2.

While not statistically significant, subtle differences were detected when examining the PU construct in relation to the class format. The online group reported a slightly higher PU score (M = 4.34, SD = 0.63) compared to the in-person group (M = 4.29, SD = 0.68). While this suggests online learners may find generative AI marginally more useful, the difference was minimal. Likewise, no significant variances were found when comparing gender.

Questions 6 to 9 were designed to understand how easy the students felt generative was to use. Data of the responses to these questions can be seen in Table 3.

These results suggest that students broadly perceived generative AI as easy to learn, operate, and master. While responses were consistently positive, the moderate variability reflects some differences in individual experiences or comfort levels with the technology. When compared across gender and class format to examine whether students’ experiences with generative AI varied, the results were reasonably conclusive. The analysis included four groups: male and female students in online and in-person academic writing classes. Mean scores ranged narrowly from 4.21 to 4.24, with an SD between 0.61 and 0.73. These consistently high values suggest that students, regardless of gender or learning environment, generally found generative AI intuitive and manageable, with minimal variation across groups. A summary of this data can be seen in Table 4.

The relationship between both the PU and PEOU constructs is central to the purpose of the TAM. Theoretically, the easier people believe a technology is to use, the more useful people tend to perceive it. In terms of this study, this relationship was examined by applying a Pearson’s correlation coefficient. For each participant, mean scores were computed across four PEOU items and seven PU items, respectively. Analysis was then conducted to assess the linear association between these two constructs. The result revealed a moderately strong, positive correlation (r = .61). This finding supports core assumptions of the TAM, where increased ease of use tends to enhance perceived usefulness and thereby strengthen overall acceptance of the technology.

There appear to be a number of other differing factors which affect both PU and PEOU. First of all, questions 16 and 17 explored the influence of students’ peers and / or instructors. These responses were grouped under the SI construct. The data generated by Pearsons correlation test shows it plays a meaningful role in shaping both PU and PEOU. The correlation between SI and PU is moderately strong (r = .42), indicating that students who observe peers or instructors effectively using AI are more likely to perceive it as academically beneficial. This suggests that usefulness is not evaluated in isolation rather socially reinforced through community exposure and shared experiences. SI also shows a smaller but still noteworthy correlation with PEOU (r = .33), implying that seeing others navigate AI tools with ease can alleviate uncertainty and boost personal confidence in their usability. Together, these connections highlight how students’ perceptions of value and simplicity are shaped not just by the tool itself, but by the behaviors and attitudes of those around them.

Questions 15 and 18 were designed to elicit answers based on the emotions of using generative AI. One such emotion was feeling guilty for using it (Q15). The data shows a meaningful negative relationship with their perceptions of primary constructs. The correlation with PU is r = -.31, indicating that students who experience higher levels of guilt tend to see generative AI tools as less beneficial to their academic writing. This suggests that ethical discomfort can diminish the perceived value of AI, even when the tools have practical benefits. The correlation with PEOU is weaker (r = -.18), but still notable, implying that guilt may slightly erode students’ sense of comfort or fluency with AI technology. These findings highlight how affective and moral concerns can suppress technology acceptance, particularly when students perceive AI use as conflicting with personal or institutional values. Contrastingly, question 18 explored the emotion of enjoyment and reveals a meaningful positive relationship with both PU and PEOU. The correlation between enjoyment and PU was r = .46, indicating that students who find AI tools engaging and pleasurable are notably more likely to perceive them as academically beneficial. Similarly, enjoyment showed a moderate correlation with PEOU (r = .35), suggesting that emotionally positive experiences during AI-assisted writing contribute to the perception that these tools are intuitive and easy to use. These findings highlight the role of affective engagement in reinforcing technology acceptance; when students genuinely enjoy using generative AI, they are more likely to embrace its value and feel confident in their ability to navigate it.

Moving forward, it is clear AI is here to stay. An understanding of how and why or why not students choose to use this technology in the future has significant pedagogical implications. One issue that could influence this is how much trust learners place in the technology and consequently how much they will use it in the future (BI). Trust in the data results provided by generative AI, measured by question 11, shows a meaningful positive relationship with BI (Question 12). The correlation between trust and BI is r = .41, indicating that students who trust AI tools to produce accurate, ethical, and reliable content are significantly more likely to intend to use them. This suggests that trust acts as a psychological enabler: when students feel confident in the integrity and reliability of AI, they are more willing to integrate it into their studies The strength of this correlation places trust alongside other key predictors like perceived usefulness and social influence, reinforcing its role as a foundational factor in technology acceptance.

Future implications such as the possibility of AI replacing the need for academic writing courses were also examined. Students show moderate agreement with this notion. This belief, examined by Question 14, correlates positively with both PU (r = .39) and BI (r = .36), suggesting that students who anticipate future AI-driven displacement of traditional instruction are more likely to view AI as beneficial and intend to use it. However, this does not necessarily mean students want AI to replace classes. It reflects a projection of technological impact, not a preference. As discussed earlier, many students may see AI as a powerful supplement or alternative, especially for tasks like grammar correction, idea generation, or personalized feedback, but still value human instruction for critical thinking, creativity, and ethical guidance.