Introduction
The concept of artificial intelligence, introduced by McCarthy in 1956 (Cristianini, 2016), has evolved from a theoretical construct to a practical force reshaping educational practice worldwide. In the field of English as a Foreign Language (EFL) teaching, AI-driven tools now support lesson planning, automated grading, personalized feedback, and interactive language practice through chatbots and writing assistants (Chen et al., 2020; Wang et al., 2024). The rapid advancement of generative AI, particularly large language models such as ChatGPT, has accelerated this transformation and placed teachers at the center of a significant pedagogical shift.
Recent systematic reviews indicate that AI integration in English language teaching is expanding, with writing assistance as the most common application, representing over 54% of studies in a recent PRISMA-guided review. AI tools such as grammar checkers, chatbots, and language learning apps have shown promise in improving linguistic accuracy, reducing student anxiety, and offering personalized learning pathways (Du and Daniel, 2024; Yuan and Liu, 2025). However, concerns about academic dishonesty, student overreliance on AI, and diminished critical thinking persist across educational contexts.
While a growing body of research has examined student perceptions and learning outcomes in AI-enhanced classrooms, the teacher perspective remains comparatively underexplored. This gap is significant because teachers are the primary gatekeepers of AI integration. Their beliefs, competencies, and institutional support structures directly shape how AI tools are introduced and function within classrooms (Celik, 2023; Omidvar & Meihami, 2025). Recent evidence indicates that the Technology Acceptance Model (TAM) is the most commonly used theoretical framework in studies of teacher AI adoption, with perceived usefulness and perceived ease of use identified as primary determinants of acceptance. However, emerging research also highlights the need for education-specific acceptance models that account for factors unique to teaching contexts, such as pedagogical beliefs, ethical concerns, and the relational nature of instruction.
The integration of AI into EFL teaching is further complicated by contextual factors. Teachers in resource-constrained settings may face additional barriers related to infrastructure, digital literacy, and access to professional development (OECD, 2018). Recent work from the Western Balkans describes this shift as a “quiet transformation” in which generative AI becomes an invisible yet pervasive presence in students’ academic lives, while many educators encounter a widening digital divide, ethical dilemmas, and unequal access to AI tools (Pireci Sejdiu and Sejdiu, 2025). In the Western Balkans, and in Kosovo in particular, recent institutional initiatives have begun to explore how AI can be embedded into higher education teaching and learning, with an emphasis on developing clear guidelines for responsible use. A recent study of Kosovan university students found that while 85.9% found AI tools easy to use, only 34.1% trusted their feedback, revealing a “trust-usability paradox” that underscores the complexity of AI acceptance (Author, Year). However, no study has yet examined how EFL teachers in these contexts perceive AI tools, what opportunities and challenges they identify, and what professional support they require.
This study addresses these gaps by investigating EFL teachers’ perceptions of AI tools in the English classroom, with a particular focus on Kosovo as an understudied, post-conflict educational context (OECD, 2018; Xue et al., 2024). Building on the Technology Acceptance Model and informed by the AI-TPACK model (Celik, 2023), the study is designed to complement recent research on Kosovan university students’ AI acceptance, which documented a trust–usability paradox in their use of AI tools for English learning (Author, Year).
Three research questions guide the investigation:
•
RQ1: What are EFL teachers’ current AI tool usage practices, including tool preferences, frequency of use, and perceived benefits for language instruction?
•
RQ2: What challenges and concerns do EFL teachers experience or anticipate when integrating AI tools into their teaching?
•
RQ3: What professional development needs and support structures do EFL teachers identify as necessary for effective AI integration?
Together, these questions provide a comprehensive picture of EFL teachers’ positioning in the evolving AI-enhanced educational landscape and allow for a nuanced comparison with student perspectives in the same national context.
Contribution statement
This article makes three main contributions. Empirically, it provides one of the first mixed-methods accounts of EFL teachers’ AI practices, opportunities, and challenges in and beyond Kosovo, thereby adding geographical and educational diversity to a literature still dominated by student-focused studies in East Asian and Gulf contexts. Theoretically, it extends Technology Acceptance Model research by situating perceived usefulness and ease of use within an AI-TPACK perspective, showing how a readiness gap and a dual identity of AI as both a productivity tool and a pedagogical threat emerge when teacher knowledge, ethics, and institutional conditions are considered. Conceptually, it links these teacher findings to previously documented student trust–usability paradoxes in the same national context, offering a more systemic view of AI-mediated EFL education. Practically, it identifies concrete priorities for AI-focused professional development, including scaffolded digital and AI literacy, ethically grounded guidelines for classroom use, and support for teachers as active designers of AI integration rather than passive users of tools.
Literature review
AI integration in EFL education
The integration of AI into language education has produced a robust and rapidly growing body of research. Systematic reviews indicate that AI tools are most commonly used for writing instruction, grammar correction, vocabulary development, and automated assessment (Lo et al., 2024; Memarian and Doleck, 2023). A 2025 systematic review using PRISMA methodology found that 54.55% of studies focused on writing assistance, while challenges such as academic dishonesty and overreliance on AI were reported in 27.27% of the reviewed articles. These findings highlight AI’s current pedagogical role in text-based language skills and signal ongoing concerns about its broader educational impact.
AI chatbots and large language models have been shown to serve as “non-threatening” practice partners that reduce speaking anxiety and increase willingness to communicate (Du and Daniel, 2024; Zhang et al., 2024). Experimental studies have demonstrated significant gains in writing performance when AI tools are used for formative feedback. For example, Mahapatra (2024) reported a large effect size (η2 = 0.902) when comparing experimental and control groups in an AI-assisted writing intervention, while Song and Song (2023) found statistically significant post-test differences in a randomized controlled trial. At the same time, comparative studies indicate that human feedback still provides superior qualitative nuance, supporting a “hybrid” model that combines AI efficiency with teacher expertise (Steiss et al., 2024; Zeevy Solovey, 2024).
Teacher perceptions of AI in language teaching
Research on teacher perceptions of AI has grown substantially since 2023, though it remains less developed than research focused on students. A systematic review of 29 studies found that EFL teachers generally regard AI tools as helpful for lesson planning, materials development, personalized learning, and providing feedback. Teachers report that AI reduces their workload and improves instructional efficiency, with lesson planning and material creation identified as the most common applications.
However, teachers’ perceptions are not uniformly positive. Multiple studies identify concerns about academic integrity, student overreliance, diminished creativity, and the limitations of AI in fostering critical thinking. A qualitative study of Indonesian EFL teachers found that while most view AI as a supportive tool, they emphasize that it cannot replace the human role in teaching, particularly in building rapport, responding to emotional cues, and adapting to real-time classroom dynamics.
Theoretical frameworks: TAM and AI-TPACK
The Technology Acceptance Model (TAM), developed by Davis (1989), remains the most widely used framework for studying teacher AI adoption. TAM states that perceived usefulness and perceived ease of use are the main determinants of technology acceptance. Recent extensions of TAM for AI in education have added constructs such as trust, self-efficacy, and hedonic motivation (Tram et al., 2024; Cai et al., 2024).
However, several scholars have argued that TAM alone is insufficient for understanding AI adoption in educational settings. The Technological Pedagogical Content Knowledge (TPACK) framework, originally proposed by Mishra and Koehler (2006), provides a complementary perspective by emphasizing the intersection of technological, pedagogical, and content knowledge. Celik (2023) expanded TPACK into the “Intelligent-TPACK” framework, incorporating ethical assessment as a key dimension. This expanded model demonstrated strong fit indices (RMSEA = 0.037, CFI = 0.980) and showed that ethical assessment positively influences technological pedagogical knowledge.
Recent work has further integrated TAM and TPACK. A study by Firdaus et al. (2025) involving 436 Indonesian preservice EFL teachers found that TPACK significantly impacts both AI behavioral intention (β = 0.350) and actual AI use (β = 0.557), while perceived ease of use drives perceived usefulness (β = 0.674). These findings suggest that integrating TAM’s attitudinal constructs with TPACK’s knowledge dimensions provides a more comprehensive account of teacher AI adoption, particularly in EFL contexts where pedagogical beliefs and content knowledge interact with technological affordances.
Teacher professional development for AI integration
The literature consistently identifies professional development as a critical enabler of effective AI integration. A TPACK-based professional development program for novice EFL teachers demonstrated significant improvements in knowledge of AI-powered tools, teaching skills, and self-efficacy (Zulianti et al., 2024). Similarly, Moorhouse et al. (2024) found that structured training helped teachers recognize the need to tailor AI-generated content to student levels, moving beyond uncritical adoption.
Despite these promising interventions, studies consistently report that most teachers lack adequate preparation. Wan et al. (2025) found that even among teachers who actively use AI, institutional clarity and perceived control over student learning significantly influence their integration practices. In the Western Balkans and Kosovo, institutional initiatives are only now beginning to develop frameworks for AI integration in higher education, with a recent focus group concluding that clear and practical guidelines for responsible AI use are needed.
Research gap
While the literature has expanded rapidly, several gaps remain. First, most teacher perception studies originate from East and Southeast Asian contexts, with limited representation from the Western Balkans and under-resourced educational settings. Second, few studies examine the interplay between teachers’ perceived opportunities and challenges within a single investigation, particularly using mixed methods designs that capture both quantitative patterns and qualitative depth. Third, the relationship between teachers’ AI usage practices and their professional development needs has not been systematically explored. This study addresses these gaps by providing a comprehensive examination of EFL teachers’ perceptions, drawing on TAM and AI-TPACK to interpret the findings within a theoretically grounded framework.
Building on this literature, the present study adopts an integrative framework that combines the Technology Acceptance Model and the AI-TPACK perspective (Davis, 1989; Celik, 2023). TAM explains how perceived usefulness, perceived ease of use, and related attitudinal constructs shape teachers’ intentions to use AI tools in their practice (Cai et al., 2024; Liu and Ma, 2024). AI-TPACK highlights the knowledge, ethical awareness, and pedagogical judgment needed to integrate these tools meaningfully into specific teaching contexts, including teachers’ capacity to evaluate AI outputs and make context-sensitive decisions about their use (Celik, 2023). By bringing these perspectives together, the study captures both attitudinal drivers of AI adoption and the broader pedagogical and ethical conditions under which EFL teachers in Kosovo and similar settings can use AI in responsible and educationally productive ways (Omidvar and Meihami, 2025).
Methodology
Research design
This study used a mixed methods approach, combining quantitative survey data with qualitative open-ended responses to provide a comprehensive understanding of EFL teachers’ perceptions of AI tools in the English classroom. The research design was guided by the TAM framework to structure attitudinal measures and the AI-TPACK model to interpret teachers’ knowledge-related responses. This approach was appropriate because teacher perceptions involve both measurable attitudes, which can be captured through Likert scale items, and nuanced experiences that require qualitative exploration.
Theoretical framework
The study is based on an integrated framework that combines the Technology Acceptance Model with the AI-TPACK perspective on teacher knowledge (Davis, 1989; Celik, 2023). TAM structures the attitudinal components of the questionnaire, especially items related to perceived usefulness, perceived ease of use, confidence, and concerns about AI replacing teaching tasks, which are central constructs in recent AI acceptance studies in language education (Liu and Ma, 2024). These constructs help explain why teachers choose to adopt, adapt, or limit AI use in their practice.
At the same time, the AI-TPACK perspective informs the interpretation of items related to professional development, ethical concerns, and perceived readiness. This framework emphasizes that effective AI integration requires coherent combinations of technological, pedagogical, and content knowledge, along with the ability to evaluate AI-based decisions critically and ethically (Celik, 2023). Viewing the findings through this combined lens allows for moving beyond a narrow focus on acceptance to consider how teachers’ knowledge, values, and institutional conditions shape their AI practices and perceived needs for support (Moorhouse et al., 2024; Omidvar and Meihami, 2025).
Participants
Twenty-eight EFL teachers participated in the study, as shown in
Table 1, representing diverse educational contexts. The sample included teachers from primary (28.6%), secondary (39.3%), undergraduate (17.9%), graduate (10.7%), and IELTS Academic (3.6%) levels. Gender distribution was relatively balanced, with 57.1% female and 42.9% male participants. Age distribution was concentrated in the 25–34 range (42.9%), followed by 35–44 (28.6%) and 45–54 (21.4%). Teaching experience was well distributed, with 32.1% reporting 15 or more years, 28.6% reporting 8–14 years, 21.4% reporting 0–2 years, and 17.9% reporting 3–7 years. The majority (89.3%) reported familiarity with AI tools used in education, while 10.7% were unsure.
Table 1. Participant demographics.
| Characteristic | Category | n |
% |
|---|
| Gender | Female | 16 | 57.1 |
| Male | 12 | 42.9 |
| Age | Under 25 | 1 | 3.6 |
| 25–34 | 12 | 42.9 |
| 35–44 | 8 | 28.6 |
| 45–54 | 6 | 21.4 |
| 55+ | 1 | 3.6 |
| Teaching level | Primary | 8 | 28.6 |
| Secondary | 11 | 39.3 |
| Undergraduate | 5 | 17.9 |
| Graduate | 3 | 10.7 |
| IELTS Academic | 1 | 3.6 |
| Experience | 0–2 years | 6 | 21.4 |
| 3–7 years | 5 | 17.9 |
| 8–14 years | 8 | 28.6 |
| 15+ years | 9 | 32.1 |
| AI familiarity | Yes | 25 | 89.3 |
| Not sure | 3 | 10.7 |
Data collection instrument
A semi-structured questionnaire was developed with four sections. Section 1 collected demographic information (age, gender, teaching level, and years of experience). Section 2 assessed AI familiarity, tool usage, and frequency of use. Section 3 included 10 Likert scale items on a five-point scale (strongly disagree to strongly agree) measuring two constructs: (a) perceived educational value of AI (five items addressing writing improvement, lesson planning efficiency, learning personalization, feedback utility, and student engagement) and (b) teacher attitudes and concerns (five items addressing confidence, role anxiety, training adequacy, content accuracy concerns, and task replacement beliefs). Section 4 included three open-ended questions inviting teachers to describe (a) the most positive impact of AI, (b) the most challenging or concerning aspect, and (c) the type of training or support needed.
The Likert scale items were developed by adapting wording and constructs from recent AI acceptance and AI-TPACK studies in language education, including research on extended Technology Acceptance Models, AI literacy, and Intelligent-TPACK (Almansour, 2024; Celik, 2023; Liu and Ma, 2024). Items on perceived usefulness and ease of use were aligned with validated TAM formulations, while items on training, ethical concerns, and role anxiety were based on emerging frameworks for teacher AI readiness and professional competence (Chiu et al., 2024; Moorhouse et al., 2024). The draft questionnaire was piloted informally with three EFL teachers to check clarity and relevance; minor wording changes were made before full distribution, but no structural changes to the constructs were necessary.
Data analysis
Quantitative data were analyzed using descriptive statistics, including frequencies, percentages, and response distributions. Qualitative responses from the three open-ended questions were analyzed using inductive thematic analysis following Braun and Clarke’s (2006) six-phase approach. Responses were first read and reread for familiarization, then systematically coded. Initial codes were collated into potential themes, which were then reviewed, refined, and named. The mixed methods approach enabled triangulation of findings, providing both statistical breadth and interpretive depth.
Ethical considerations
The study followed institutional ethical guidelines for research involving human participants. A written informed consent statement was obtained from all examinees. Before the distribution of the questionnaires, the examinees were advised on: (1) the information obtained from the questionnaires would be used for scientific purposes only‚ (2) their identity and privacy would be kept fully confidential‚ and (3) they were free to withdraw from the study at any time. Participant privacy was protected through anonymized data collection, and no personally identifying information was stored. Participation was voluntary, and respondents were informed of their right to withdraw at any time.
Results
AI tool usage practices
Regarding frequency of use presented in
Table 2, the largest group of teachers reported using AI “sometimes” (46.4%, n = 13), followed by “often” (35.7%, n = 10), “rarely” (14.3%, n = 4), and “always” (3.6%, n = 1). No participant reported never using AI tools. Combining “often” and “always,” 39.3% of teachers were frequent users, while 60.7% used AI less frequently.
Table 2. Frequency of AI tool use.
| Frequency | n |
% |
|---|
| Always | 1 | 3.6 |
| Often | 10 | 35.7 |
| Sometimes | 13 | 46.4 |
| Rarely | 4 | 14.3 |
| Never | 0 | 0.0 |
Tool preferences in
Table 3 showed a clear dominance of ChatGPT, used by 89.3% (n = 25) of participants. AI-based quiz and test generators were the second most popular category (57.1%, n = 16), followed by Grammarly (50.0%, n = 14) and Quillbot (39.3%, n = 11). Text-to-speech and speech-to-text tools were used by 28.6% (n = 8) of participants. Other tools, including Gemini, NotebookLM, Canva, Flippity, and Perplexity, received isolated mentions.
Table 3. AI tools used by teachers.
| AI Tool | n |
% |
|---|
| ChatGPT | 25 | 89.3 |
| AI based quiz/test generators | 16 | 57.1 |
| Grammarly | 14 | 50.0 |
| Quillbot | 11 | 39.3 |
| Text to speech/speech to text | 8 | 28.6 |
| Other (Gemini, Canva, Perplexity, etc.) | 5 | 17.9 |
Perceived educational value of AI
Likert scale responses in
Table 4 regarding the educational value of AI revealed a nuanced pattern. Teachers showed the strongest agreement on lesson planning efficiency and learning personalization, while responses on writing improvement and student engagement were more moderate.
Table 4. Teachers’ perceived educational value of AI (Likert Scale Responses).
| Statement | SA | A | N | D | SD |
|---|
| AI makes lesson planning more efficient | 12 (42.9%) | 7 (25.0%) | 5 (17.9%) | 1 (3.6%) | 3 (10.7%) |
| AI can help personalize learning | 9 (32.1%) | 11 (39.3%) | 4 (14.3%) | 0 (0.0%) | 4 (14.3%) |
| AI provides useful feedback for learners | 5 (17.9%) | 12 (42.9%) | 6 (21.4%) | 5 (17.9%) | 0 (0.0%) |
| AI encourages student engagement | 7 (25.0%) | 8 (28.6%) | 8 (28.6%) | 3 (10.7%) | 2 (7.1%) |
| AI helps students improve writing | 2 (7.1%) | 9 (32.1%) | 13 (46.4%) | 3 (10.7%) | 1 (3.6%) |
Lesson planning efficiency received the highest combined agreement (67.9%), followed by learning personalization (71.4%), feedback utility (60.7%), and student engagement (53.6%). Writing improvement had the most neutral responses (46.4%), with only 39.3% agreement. This pattern suggests that teachers view AI as more useful for behind-the-scenes planning and individualization than for directly improving specific language skills.
Teacher attitudes and concerns
The second set of Likert items in
Table 5, captured teachers’ confidence, anxieties, and beliefs about AI’s impact on their professional role.
Table 5. Teacher attitudes and concerns (Likert Scale Responses).
| Statement | SA | A | N | D | SD |
|---|
| I feel confident using AI tools | 6 (21.4%) | 13 (46.4%) | 6 (21.4%) | 3 (10.7%) | 0 (0.0%) |
| I am concerned about AI content accuracy | 5 (17.9%) | 13 (46.4%) | 5 (17.9%) | 2 (7.1%) | 3 (10.7%) |
| I have received enough training | 5 (17.9%) | 4 (14.3%) | 6 (21.4%) | 10 (35.7%) | 3 (10.7%) |
| AI could replace some teaching tasks | 3 (10.7%) | 9 (32.1%) | 7 (25.0%) | 6 (21.4%) | 3 (10.7%) |
| I worry AI will reduce my role | 3 (10.7%) | 1 (3.6%) | 6 (21.4%) | 14 (50.0%) | 4 (14.3%) |
Most teachers (67.9%) expressed confidence in using AI tools, consistent with the high familiarity rate (89.3%). A substantial majority (64.3%) were concerned about the accuracy of AI-generated content. Only 32.1% agreed they had received adequate training, while 46.4% disagreed, revealing what we term the “readiness gap.” Regarding professional displacement, 64.3% of teachers were not concerned that AI would reduce their role, suggesting resilience in professional identity. However, 42.9% agreed that AI could replace some teaching tasks, indicating a pragmatic acknowledgment of AI’s potential rather than an existential threat.
To examine whether confidence varied by career stage, teaching experience was cross-tabulated with confidence in using AI tools.
To examine whether AI confidence clustered at specific career stages, a simple cross-tabulation of teaching experience by confidence in using AI tools was created (
Table 6). The results showed that high confidence (agree or strongly agree) was distributed relatively evenly across all experience groups, ranging from 55.6% among teachers with 15 or more years of experience to 100% among those with 3–7 years of experience. Notably, teachers with 15 or more years of experience reported confidence levels comparable to those with 0–2 or 3–7 years of experience, indicating that willingness to work with AI is not limited to early career teachers. Given the small sample size (N = 28), no inferential tests were conducted, but this descriptive pattern aligns with studies reporting limited or inconsistent demographic effects on teacher AI acceptance (Alharbi, 2024; Slamet, 2024).
Table 6. Teaching experience by confidence in using AI tools.
| Years of teaching experience | Low confidence (SD/D) | Neutral (N) | High confidence (A/SA) | Total |
|---|
| 0–2 years | n = 0 | n = 2 | n = 4 | n = 6 |
| 3–7 years | n = 0 | n = 0 | n = 5 | n = 5 |
| 8–14 years | n = 1 | n = 2 | n = 5 | n = 8 |
| 15+ years | n = 2 | n = 2 | n = 5 | n = 9 |
| Total
| n = 3 | n = 6 | n = 19 | N = 28 |
Reported challenges
The multiple-choice question on challenges revealed four main concerns (
Table 7). Ethical issues (cheating, misinformation) and student overreliance on AI were reported by 64.3% of teachers. Limited access to appropriate technology was cited by 46.4%, and lack of training by 42.9%. Time constraints in lesson planning or adaptation were mentioned by 14.3%.
Table 7. Challenges reported by teachers.
| Challenge | n |
% |
|---|
| Ethical concerns (cheating, misinformation) | 18 | 64.3 |
| Student over reliance on AI tools | 18 | 64.3 |
| Limited access to appropriate technology | 13 | 46.4 |
| Lack of training | 12 | 42.9 |
| Time constraints in lesson planning | 4 | 14.3 |
Qualitative findings: positive impacts
Thematic analysis of the open-ended question on positive impacts identified four primary themes.
Theme 1: Time Efficiency in Lesson Planning.
The most frequently mentioned benefit was saving time on lesson preparation and materials development. Teachers described AI as a tool that significantly reduces the workload of creating lesson plans, tests, worksheets, and differentiated activities. One teacher stated, “Fast ways to prepare lesson plans, tests, etc.” Another explained, “I do not use AI in the classroom, except rarely, but I use it in the preparation of my lesson plans. And this is incredibly time-saving.” A third noted, “Reducing lesson planning time and having more time to focus on students and activities.”
Theme 2: Personalization and Differentiation.
Teachers valued AI’s ability to tailor content to different student needs and levels. One respondent wrote, “AI has helped personalize student learning activities and significantly reduced the workload involved in planning and creating materials. It allows me to tailor lessons to different learning styles and provide immediate feedback.” Another stated, “Having the opportunity to create different tasks to meet the needs of various students, especially those with learning needs or neurodivergent students.”
Theme 3: Student Engagement.
Several teachers noted that AI increased engagement and motivation. One described how AI “helps me come up with creative ideas that make learning fun and an unforgettable experience for students.” Another mentioned “better engagement” as the primary impact, while a third cited “personalized learning paths, instant feedback and support, and increased student engagement.”
Theme 4: Creative Instructional Ideas.
A smaller but notable group of teachers valued AI as a source of instructional creativity. One teacher noted that AI “provides creative and interactive ideas,” while another described using AI for “specific lesson planning and more creative approaches in practice sessions.”
Qualitative findings: challenges and concerns
Thematic analysis of the open-ended question on challenges revealed five themes.
Theme 1: Student Over reliance and Loss of Critical Thinking.
The most prominent concern was that students become excessively dependent on AI tools, which erodes their capacity for independent thinking and authentic language production. One teacher stated, “The most concerning issue is students’ overreliance on AI-generated content. Many are becoming too dependent on these tools, which affects their creativity, critical thinking, and attention span.” Another wrote, “Ensuring students do not become overreliant on it and still develop their own critical thinking skills.”
Theme 2: Accuracy and Trustworthiness of AI Content.
Teachers expressed concern about the reliability of AI generated information and the challenge of verifying its accuracy. One respondent stated: “Trying to confirm the accuracy of information provided from AI.” Another observed: “Evaluating the AI responses and adapting it.”
Theme 3: Erosion of Creativity.
A distinct theme emerged regarding the concern that AI diminishes both student and teacher creativity. One teacher wrote, “Losing both students’ and teachers’ creativity.” Another expressed fear that AI was “making us passive teachers.”.”
Theme 4: Academic Integrity.
Several teachers raised ethical concerns about cheating and plagiarism. One teacher noted, “Students use ChatGPT to write papers assigned as homework.” Another identified concerns about academic integrity and cheating, overreliance and eroded critical thinking, and equity and access disparities.
Theme 5: Infrastructure and Access Barriers.
Teachers in some contexts reported practical barriers. One stated, “Sometimes we do not have access to the internet.” Another cited limited resources as a primary obstacle.
Qualitative findings: training and support needs
Thematic analysis of the training question identified four themes.
Theme 1: Hands on, Practical Training.
The most common request was for practical, experiential training sessions. Teachers asked for “hands-on training with clear, practical guidelines,” “workshops regarding the use of AI,” and training that takes a “hands-on approach to training for specific AI tools.”
Theme 2: Ethical and Responsible Use Training.
Teachers sought guidance on navigating ethical dimensions. One respondent requested “training related to ethical considerations of AI use” along with “an overview of different tools for different purposes.” Another asked for “practical integration strategies (lesson planning and curriculum alignment), prompt engineering skills, ethical use, and policy development.”
Theme 3: Tool specific and Contextualized Training.
Teachers expressed a need for training tailored to specific tools and local contexts. One teacher asked for guidance on how to use websites like MagicSchool AI and Quizziz AI to generate materials and resources. Another requested training that considers the needs of local-level learners.
Theme 4: Foundational Technology Training.
Some teachers identified a need for more basic technology integration skills before advancing to AI. One teacher stated, “We as teachers need basic training on incorporating technology in the classroom first, and then we can expand our goals to AI tools.”
Discussion
AI as a lesson planning tool: pragmatic adoption
The findings of this study confirm and expand on recent research documenting teachers’ pragmatic adoption of AI tools. The dominance of ChatGPT (89.3%) reflects global patterns reported in recent systematic reviews, where ChatGPT and similar generative AI tools have become the primary entry point for educators exploring AI-assisted instruction. The emphasis on lesson planning efficiency as the top perceived benefit aligns with findings from studies in Oman, Indonesia, and North Macedonia, where teachers consistently identify materials development and lesson preparation as the most valued applications of AI.
This pattern of pragmatic, task-oriented adoption aligns with the TAM framework, which predicts that perceived usefulness drives technology acceptance (Davis, 1989). Teachers in this study clearly found AI useful for reducing workload and saving time, both direct indicators of perceived usefulness. The relatively lower agreement on writing improvement (39.3%) compared to lesson planning (67.9%) suggests that teachers view AI as more effective for their own professional tasks than for directly improving students’ language skills, a distinction with important implications for how AI is positioned in teacher training programs.
The readiness gap: willing but unprepared
One of the most significant findings is the discrepancy between teachers’ confidence in using AI (67.9% agreement) and their perception of having received adequate training (only 32.1% agreement). This 35.8 percentage point “readiness gap” suggests that many teachers are engaging with AI tools through self-directed exploration rather than structured professional development. This finding aligns with Wan et al.’s (2025) typology of “cautious integrators,” who use AI selectively but feel constrained by insufficient institutional guidance.
The readiness gap also relates to research on the AI-TPACK framework, which emphasizes that effective AI integration requires not only technological skills but also the ability to align AI tools with pedagogical goals and content knowledge (Celik, 2023). Teachers without structured training may develop technological knowledge separately from pedagogical integration strategies, which can result in superficial AI use that does not achieve its full educational potential. The finding that some teachers requested “basic training about the incorporation of technology in the classroom first” before moving on to AI-specific skills suggests that professional development programs should be scaffolded, addressing foundational digital literacy before introducing AI-specific competencies.
Extending TAM through an AI-TPACK lens
From a theoretical standpoint, the readiness gap identified in this study highlights the limitations of relying solely on Technology Acceptance Model constructs to understand AI adoption among teachers (Davis, 1989). Teachers in this sample report high perceived usefulness and ease of use, yet also describe significant deficits in training and uncertainty about how to align AI tools with their pedagogical goals and ethical responsibilities – a pattern noted in other recent teacher-focused studies (Slamet, 2024; Derakhshan and Ghiasvand, 2024). This finding supports the argument that TAM should be complemented by AI-TPACK, in which technological knowledge is explicitly connected to pedagogical strategies, content demands, and ethical evaluation (Celik, 2023). In other words, positive attitudes toward AI are not sufficient; teachers need structured opportunities to develop integrated AI-TPACK so their everyday classroom decisions about AI use are guided by robust pedagogical and ethical reasoning rather than experimentation alone (Moorhouse et al., 2024; Nazaretsky et al., 2022).
The dual identity: productivity tool vs. pedagogical threat
The qualitative findings reveal what we describe as a “dual identity” in teachers’ relationship with AI. On one hand, teachers enthusiastically embrace AI as a productivity tool that saves time, generates materials, and supports differentiated instruction. On the other hand, they express deep concern about its impact on creativity, critical thinking, and the relational aspects of teaching. This tension is also evident in the quantitative data: while 64.3% of teachers do not worry about AI reducing their professional role, 64.3% express concern about AI content accuracy, and an equal proportion identify student overreliance as a major challenge.
This dual identity aligns with the “facilitator–cautious integrator” continuum identified by Wan et al. (2025) and with Derakhshan and Ghiasvand’s (2024) finding that Iranian EFL teachers view AI as both a “catalyst” and a potential threat to ethical teaching practice. The concern about “making us passive teachers,” expressed by one respondent, reflects a deeper anxiety about pedagogical deskilling – a concern that extends beyond the TAM framework’s focus on ease of use and perceived usefulness. This finding supports recent calls to incorporate teacher agency and professional identity into technology acceptance models for educational AI.
Ethical concerns as a primary barrier
Ethical concerns, such as cheating and misinformation, and student overreliance emerged as the most frequently cited challenges, each reported by 64.3% of teachers. This finding is consistent with the international literature. A systematic review of EFL teacher perspectives found that the potential for AI to promote cheating and plagiarism is one of the most frequently cited concerns worldwide. Similarly, Daud et al. (2025) reported that overreliance on AI was identified as a challenge in 27.27% of reviewed studies. These concerns are also evident in broader higher education debates, where educators question whether they are grading students or grading ChatGPT, and note that conventional plagiarism detection tools are increasingly ineffective against sophisticated AI-generated prose (Pireci Sejdiu and Sejdiu, 2025).
The prominence of ethical concerns has significant theoretical implications. In the traditional TAM model, ethical considerations are not explicitly included as a construct. However, the current findings indicate that ethical concerns act as a moderating variable that can weaken the relationship between perceived usefulness and actual adoption. For example, a teacher may find AI genuinely useful for lesson planning but limit its use in student-facing activities due to concerns about academic integrity. This interpretation aligns with Author’s (Year) finding of a “trust usability paradox” among Kosovan EFL students, where high ease of use coexisted with low trust, and extends it to the teacher context, where trust concerns focus not on personal use but on the downstream effects of student AI dependency.
Convergence and divergence with student perceptions
The findings of this teacher study can be considered alongside evidence from Kosovan EFL students, who also report intensive use of AI tools for language learning and a complex mix of enthusiasm and concern. Both groups identify grammar, writing, and vocabulary as the main skill domains supported by AI tools, and both emphasize time savings, instant feedback, and ease of access as central benefits. In this sense, teachers and students share a broadly pragmatic orientation toward AI, treating it as a convenient assistant for structurally focused tasks rather than a transformative replacement for human instruction.
However, there are also notable differences. Students report a pronounced trust–usability paradox: 85.9 percent find AI tools easy to use, yet only 34.1 percent trust the feedback they receive. Teachers in the present study, by contrast, show comparatively high confidence in using AI tools but focus their distrust on downstream pedagogical risks, such as student overreliance, academic integrity, and the erosion of creativity, rather than on their own immediate interaction with the tools. While students frame trust primarily in terms of the reliability of error correction and content accuracy, teachers extend trust concerns to broader questions of ethical responsibility, classroom control, and the long-term development of critical thinking.
These differences suggest that trust operates at two distinct levels in AI-mediated language education. At the learner level, trust concerns focus on the correctness and usefulness of specific feedback episodes. At the teacher level, trust concerns encompass systemic issues, including assessment validity, equitable access, and the preservation of human roles and pedagogical values. Integrating these perspectives indicates that future theoretical models should treat trust not as a single construct but as a layered phenomenon, with separate components for micro-level interactional trust in feedback and macro-level institutional and pedagogical trust in AI’s role within the curriculum.
Alignments and gaps in teacher–student perceptions
To clarify these convergences and divergences,
Table 8 summarizes key similarities and differences between Kosovan EFL teachers in the present study and Kosovan EFL students reported in Author (Year) regarding tools used, purposes, perceived benefits, concerns, and trust–usability patterns.
Table 8. Alignment and differences between Kosovan EFL teachers and students.
| Dimension | EFL teachers (present study) | Kosovan EFL students (Author, Year) |
|---|
| Main AI tools used
| ChatGPT dominant (89.3%); AI based quiz and test generators; Grammarly; Quillbot. | Google Translate and ChatGPT equally dominant (68.2% each); Grammarly and Quillbot widely used. |
| Primary purposes
| Lesson planning, materials creation, quiz and test generation, preparation of differentiated tasks. | Vocabulary, grammar, and writing support; translation; exam and assignment preparation; feedback on mistakes. |
| Most valued benefits
| Time efficiency in lesson planning; workload reduction; support for personalization and differentiation; creative ideas for activities. | Instant feedback and correction; time efficiency and accessibility; personalized and adaptive learning; vocabulary expansion; enjoyment and engagement. |
| Main concerns
| Student over reliance on AI; loss of creativity and critical thinking; academic integrity and cheating; accuracy of AI content; infrastructure and access barriers. | Accuracy and reliability of feedback; overdependence on AI; limited contextual and cultural understanding; ethical and integrity issues. |
| Trust and usability pattern
| High confidence in using AI; strong concerns about accuracy and ethical use in the classroom; fear of pedagogical deskilling (“passive teachers”). | Very high perceived ease of use (85.9%) but low trust in AI feedback (34.1%); empirically documented trust–usability paradox. |
| Perceived role of AI in teaching/learning
| Viewed as powerful support tool for planning and differentiation, not a replacement for human teaching; anxiety about role erosion is limited but present. | Viewed as helpful assistant for language improvement, but not a substitute for teachers, especially for ideas, experiences, and oral communication skills. |
Implications for professional development
The findings indicate that many teachers already use AI tools for lesson planning, materials development, and differentiation, yet report limited formal training and ongoing uncertainty about ethical and pedagogical use. This pattern suggests that professional development should go beyond basic tool demonstrations toward systematic development of AI literacy and AI-TPACK. Recent studies show that targeted programs can improve teachers’ knowledge of AI-powered tools, their ability to design AI-supported lessons, and their self-efficacy in using these tools (Enhancing Novice EFL Teachers’ Competency in AI-Powered Tools Through a TPACK-Based Professional Development Program, 2024; Effectiveness of a Professional Development Program for AI Literacy, 2024). These results reinforce the need for structured, sustained training rather than leaving teachers to experiment individually.
In this study, teachers explicitly requested practical, context-sensitive support, including concrete examples of prompts, guidance on integrating AI into existing curricula, and strategies for addressing academic integrity and overreliance among students. These needs align with emerging frameworks that conceptualize professional generative AI competence as a combination of technological proficiency, pedagogical alignment, ethical awareness, and preparation of students for AI-rich environments (Moorhouse et al., 2024; Chapelle, 2025). Professional development in the Kosovan and broader EFL context would therefore benefit from combining hands-on workshops focused on specific teaching scenarios with explicit discussion of ethical issues, local policy expectations, and ways to maintain creativity and critical thinking when AI is present. Designing professional development around authentic teacher tasks, such as co-creating writing prompts with AI, revising AI-generated materials, or modeling critical evaluation of AI outputs in class, can help teachers develop integrated AI-TPACK rather than isolated technical skills.
The readiness gap identified in this study suggests that professional development should be tailored to career stage and existing AI experience. The cross-tabulation showed that high confidence in using AI tools was not limited to early-career teachers, indicating that both novice and experienced teachers can be leaders in AI integration. Meanwhile, international research on AI readiness and AI literacy among teachers highlights that confidence, perceived usefulness, and access to resources interact with institutional support and clear guidelines (Luckin et al., 2022; Al-Abdullatif, 2024). Institutions should therefore design multi-level professional learning pathways that combine foundational AI literacy with advanced modules on assessment, feedback, and student guidance, and recognize teachers as co-designers of responsible AI use in language education rather than passive recipients of technology.
Limitations and future research
This study has several limitations that should be considered when interpreting the findings. First, the teacher sample was obtained through convenience sampling and was relatively small, which limits statistical power and the ability to detect more fine-grained patterns in attitudes and practices. Participants were self-selected EFL teachers who were already at least somewhat familiar with AI tools, likely overrepresenting early adopters and those interested in experimenting with new technologies. As in other AI in education studies using nonprobability samples, the patterns reported here should therefore be treated as analytically informative rather than statistically representative of all EFL teachers in Kosovo or beyond (Crompton & Burke, 2023; Tram et al., 2024).
Second, the study is situated in Kosovo, a post-conflict, under-resourced educational context, which is both a strength and a limitation. It provides much-needed evidence from a setting largely absent in current AI in language education research, but the specific institutional conditions, digital infrastructure, and policy environment may differ significantly from those in other regions (OECD, 2018; Xue et al., 2024). At the same time, Kosovo is part of a broader regional conversation in which generative AI is viewed as both a disruptor and an enabler in higher education, with calls for inclusive policies and critical AI literacies that balance innovation and integrity (Pireci Sejdiu and Sejdiu, 2025). Therefore, transferability of the findings to other countries should be based on reasoned comparison rather than direct generalization, considering similarities and differences in teacher education systems, access to AI tools, and national regulations.
Third, the design captures teacher perceptions at a single point in time and relies on self-report questionnaire data, without classroom observations or objective measures of AI use. Self-report data are vulnerable to social desirability and recall bias, and cross sectional snapshots cannot show how teachers’ AI practices and levels of trust evolve as tools, policies, and training opportunities change (Memarian and Doleck, 2023; Yang and Li, 2024). The study also analysed teacher and student data in separate projects, which allowed for a detailed focus in each article but limited the ability to examine teacher–student dynamics within the same institutional settings.
Future research
Future research should address these limitations in three main ways. First, larger scale studies using stratified or probability based sampling are needed to examine whether the patterns identified here hold across different regions, school levels, and institutional types, both within Kosovo and in other Western Balkan and post conflict contexts (Kundu and Bej, 2025; Jiang, 2022). Such designs would permit more robust modelling of how factors such as institutional support, AI literacy, and subject specialization shape teachers’ AI practices and concerns. Second, mixed teacher–student designs that collect data from both groups in the same institutions would make it possible to analyse how teacher attitudes, classroom practices, and institutional messages interact with student acceptance patterns such as the trust–usability paradox documented in Author (Year). This type of design would extend the comparative insights outlined in the present Discussion and offer a more systemic view of AI mediated EFL ecologies.
Third, longitudinal and intervention studies are needed to trace how AI related perceptions and competencies develop over time as teachers receive targeted professional development or as institutional policies change. Longitudinal work could examine whether the readiness gap identified here narrows when teachers participate in AI literacy or AI-TPACK programs (Wang et al., 2024; Moorhouse et al., 2024), while classroom based interventions could test specific strategies for integrating AI in ways that protect creativity, critical thinking, and academic integrity. Combining survey data with classroom observations, artefact analysis of AI assisted materials, and possibly learning analytics would further strengthen the evidence base for designing responsible AI integration in EFL teaching across diverse contexts.
Conclusion
This study offers a detailed account of EFL teachers’ perceptions of AI tools in the English classroom, revealing a complex picture of pragmatic adoption tempered by significant concerns. Teachers overwhelmingly value AI for lesson planning efficiency and learning personalization, with ChatGPT serving as the dominant tool across educational levels. However, ethical concerns, student overreliance, accuracy issues, and insufficient training create substantial barriers to deeper integration.
The identification of a “readiness gap” between teachers’ confidence and the adequacy of their training is a key finding with direct policy implications. Institutions and professional development providers should prioritize scaffolded, experiential training programs that start with foundational digital literacy and progress to AI-specific pedagogical integration, incorporating ethical frameworks throughout. The dual identity revealed in teachers’ qualitative responses – simultaneously embracing and resisting AI – suggests that effective training must address not only technical skills but also professional identity, pedagogical beliefs, and critical evaluation competencies.
From a theoretical perspective, the findings support extending the TAM framework for educational AI contexts by incorporating trust, ethical concerns, and training adequacy as explicit constructs. The AI-TPACK model offers a promising complement by ensuring that technological knowledge is integrated with pedagogical and content expertise rather than treated in isolation.
At the theoretical level, this study contributes to ongoing discussions about modeling AI adoption in education by integrating Technology Acceptance Model constructs with an AI-TPACK perspective (Davis, 1989; Celik, 2023). The findings indicate that high perceived usefulness and ease of use can coexist with significant gaps in training, ethical guidance, and pedagogical integration, which traditional TAM formulations do not fully capture (Al-Abdullatif, 2024; Author, Year). Framing this readiness gap and the dual identity of AI as both a productivity tool and a pedagogical threat through AI-TPACK underscores the need to treat teacher knowledge, ethical judgment, and institutional conditions as integral components of AI acceptance rather than as external constraints (Celik, 2023; Omidvar and Meihami, 2025). This integrated perspective can inform future models of teacher AI adoption that more accurately reflect the complexity of classroom decision making in AI-mediated EFL contexts.
Future research should use larger samples to allow for inferential statistical analysis, adopt longitudinal designs to track attitudinal change, and include classroom observation to link perceptions with practice. Cross-contextual comparisons, especially between well-resourced and resource-constrained educational systems, would further clarify how contextual factors influence AI adoption among EFL teachers. Comparative studies examining student and teacher perceptions within the same context would be especially valuable for understanding the alignment or disconnect between these stakeholder groups.
Ethical approval statement
This study received ethical approval from the Research Ethics Committee of the University “Ukshin Hoti” Prizren, Kosovo, under reference number 02–253.
Conflict of interest
The authors declare that there are no conflicts of interest regarding the publication of this study.
