AI literacy in shadow education: exploring Chinese EFL practitioners’ perceptions and experiences

3.1 Participants

Participants were invited using a combination of purposive and convenience sampling (Cohen et al. 2018). To ensure the relevance and depth of insights, participants met the following inclusion criteria:

• 1) Current engagement in English language tutoring within China’s shadow education sector.

• 2) Experience in incorporating AI tools into their teaching practices, including but not limited to AI-driven language assessment, automated feedback systems, virtual tutoring, and adaptive learning technologies.

• 3) Willingness to participate in semi-structured interviews and provide reflections on their teaching practices.

Twenty potential participants were identified through the authors’ professional networks in the shadow education industry. Following an initial email inquiry regarding their willingness and availability, nine tutors from diverse tutoring contexts, including private tutoring institutions, online platforms, and freelance settings, consented to participate (see Table 1).

Ethical considerations were rigorously observed throughout the study. Participants were informed about the study’s purpose, data collection methods, and rights. Ethical approval was obtained following institutional guidelines, and participants provided informed consent. All data were pseudonymised to maintain confidentiality, and any identifying details were removed before reporting the results.

3.2 Data collection

Data for this study were collected exclusively through one-to-one semi-structured interviews, which were used to explore the complexity of tutors’ perceptions and practices regarding AI literacy in China’s shadow education. The second and fourth authors conducted interviews between August and September 2024 via Tencent Meeting, a widely used online conferencing platform in China. The interview protocol was developed based on a comprehensive review of existing literature on AI literacy in general education and language education. While the core set of questions was theory-driven, interviewers also incorporated probes and follow-up questions in response to participants’ individual reflections and experiences, ensuring a context-sensitive and participant-driven inquiry (Tisdell et al. 2025). Each interview lasted between 45 and 120 mins, resulting in a total recording length of 822 mins, and they were audio-recorded with participants’ informed consent to ensure the accuracy and completeness of the data for subsequent analysis. The recordings were later transcribed verbatim to facilitate rigorous thematic analysis (see Section 3.3).

3.3 Data analysis

The collected interview data were analysed using thematic analysis, following the six-phase framework proposed by Braun and Clarke (2006). First, all interview recordings were transcribed verbatim using the IFLYREC, an online transcription platform. To ensure transcription accuracy and reliability, the transcripts were sent back to the participants for member verification (Lincoln and Guba 1985). Additionally, the second and fourth authors conducted multiple readings of the transcripts to develop a deep understanding of the data. This familiarisation stage (Nowell et al. 2017) allowed for the identification of initial patterns, potential themes, and interconnections within the data.

Second, the second and fourth authors conducted line-by-line coding to identify recurrent ideas, concepts, and practices emerging from the data. An initial coding was developed based on both inductive (data-driven) and deductive (theory-driven) approaches (Braun and Clarke 2019). The inductive approach ensured emergent themes from the data, while the deductive approach allowed for mapping findings onto existing theoretical frameworks related to AI literacy.

Following the initial coding, similar or related codes were grouped into broader themes that encapsulated participants’ core beliefs, attitudes, and practices regarding AI integration in shadow education. The thematic categorisation helped uncover key pedagogical concerns, challenges, and perceived benefits of AI-assisted teaching.

Moreover, group checking and discussions were conducted to enhance the credibility and validity of the findings. The first and second authors independently reviewed and validated the identified themes by comparing them with participants’ original transcripts to ensure that the interpretations accurately reflected their perspectives and practices. Discrepancies were discussed and resolved through consensus to minimise potential researcher bias (Creswell and Poth 2016).

Fifth, the refined themes (see Appendix A) were clearly defined and labelled to ensure they captured the essence of participants’ perspectives and practices. Each theme was supported by interview excerpts, demonstrating how the practitioners constructed and negotiated their AI literacy in different teaching contexts. Additionally, themes were linked to existing literature on AI-assisted language education and professional development to provide a contextualised interpretation of the findings.

Sixth, the reported data were translated into English by the first and second authors, who have extensive academic translation experience. To ensure translation fidelity, the translated transcripts were cross-checked against the original Chinese texts and refined through a back-translation process. In addition, a peer-review procedure was conducted by the third author, who is a professional Chinese-English translator (van Nes et al. 2010).

3.4 Trustworthiness

Reflexivity protocols were implemented to critically examine our positionality and minimise biases and potential influences throughout the research process, thereby ensuring transparency, ethical integrity, and credibility in this study (Lincoln and Guba 1985). Reflective memos were utilised during both data collection and analysis to document our ideas, thoughts, assumptions, and evolving interpretations. Regular group discussions among the four authors were held to challenge assumptions and refine interpretations. Participants were invited to review transcripts and preliminary findings to further enhance credibility and validate our analysis. These practices helped identify and mitigate potential biases, ensuring that personal perspectives did not unduly influence the interpretation of the collected data.

4.1 Attitudes towards AI application in teaching

Participants expressed varying attitudes towards the use of AI in teaching. Some held positive views, perceiving AI as a valuable assistant tool. For example, Tang highlighted AI’s ability to retrieve suitable teaching materials and provide a rich corpus to enhance the quality of instruction, stating:

“Current AI applications are already extensive, and future advancements could enable AI to handle more complex tasks, which could potentially assist a lot of teaching activities.” (Interview, Tang)

Others echoed this perspective:

“I think it is quite positive. As for its role, I believe it can largely serve as a powerful aid for teachers.” (Interview, An)

“I think AI plays the role of a ‘military strategist’.” (Interview, Yuan)

These perspectives indicate that some tutors view AI as an assistant, while also recognising that it remains human-led. Yuan, for instance, likened AI to a ‘military strategist’ because he saw its potential to enrich teaching content behind the scenes, although it could not directly interact with students. Most participants regarded AI positively, acknowledging its potential to enhance efficiency, reduce workload, and support student learning.

However, some tutors expressed caution towards AI integration in teaching, primarily due to institutional constraints and strong goal-oriented expectations from parents. As Guo explained:

“It’s not that AI couldn’t be useful, but my company has rigid instructional structures. For example, in a six-to-eight-person class, I must start by introducing vocabulary, then practice short sentences, then move on to listening, explain exercises and so on. There’s no flexibility to just ‘insert AI’ into this process. Besides, classes are recorded, and it is hard to say if they will comply with institutional regulations. Unless the institution says, we are now using AI in class. You simply follow the existing structure.” (Interview, Guo)

This statement highlights how predetermined lesson structures may limit tutors’ autonomy to experiment with AI in teaching. Guo further emphasised that the primary goal in her institution was to improve students’ test scores, and tutors risked receiving complaints if they introduced innovative methods:

“Students care about how you can improve their grades, and it seems they are not concerned about whether you use some novel method. Parents pay tuition fees expecting structured lessons. If I use AI for discussion-based activities and spend 10 minutes letting students talk, I could easily get complaints. If that happens, the teacher is fully responsible, and there’s no way around it.” (Interview, Guo)

Participants (e.g., Guo and Lu) also described a disconnect between decision-makers and end-users in AI adoption. They noted that school administrators sometimes purchase AI products that do not align with teachers’ or students’ actual needs, resulting in a mismatch between design and practical application. As Guo pointed out, teachers are often expected to adapt to AI products rather than use those specifically designed to support teachers’ instructional needs.

Some tutors also described a transition in their attitudes towards AI. Wang, for example, initially resisted AI due to several factors, including his belief that his age made it more challenging to adapt to new technologies. He also felt that his background in the humanities created a natural resistance to technical tools, which he perceived AI tools as complex and inconvenient. The costs and accessibility of specific AI tools further discouraged him. However, after trying some AI tools, he recognised their practical benefits, particularly in lesson planning and assisting with grading.

Tutors’ perceptions greatly shaped their instructional strategies and approaches to technology integration. Research suggests that teachers are more likely to adopt technological tools when they perceive them as beneficial to teaching and learning (Deng et al. 2014). In this study, not all participants immediately embraced AI upon first encountering it, but most acknowledged that its development and increasing prevalence in education are irreversible trends. Some participants (e.g., Fei and Yu) emphasised the importance of embracing AI to stay relevant, warning that ignoring this shift could result in being left behind. Additionally, Yu noted that basic and repetitive teaching tasks could be easily replaced by AI, posing potential risks to the role of teachers. This awareness prompted her to explore ways to adapt her pedagogy by focusing on creativity and human-centred teaching skills.

Despite the perceived benefits, participants also raised concerns about AI’s pedagogical integration, particularly regarding regulatory constraints and the risk of overreliance on AI. Yu, for instance, expressed concerns that large language models could be misused, as students sometimes fail to validate AI-generated content, which could negatively impact their knowledge acquisition and skill development. She warned that excessive reliance on AI might lead to declines in students’ critical thinking, analytical skills, and language proficiency, as AI-generated content often lacks specificity. To mitigate these risks, she stressed the need for training and management measures to regulate AI use, ensuring that students and educators can use AI tools effectively and responsibly.

Additionally, some tutors exhibited a teacher-centred attitude toward AI, as reflected in the following statement:

“We need to understand that AI is a tool to support our teaching but not a replacement for our role. In the teaching process, it’s important to strengthen interaction with students and pay attention to their learning needs. In general, tutors hope their students do well and they often feel guilty if this goal is not met. However, AI is not accountable for outcomes. AI only responds to tutors’ requests. That is why tutors must rely on their own judgement and take ownership of their responsibilities.” (Interview, An)

An’s statement underscores the belief that while AI can assist educators, the ultimate responsibility for student outcomes rests with teachers. Her perspective highlights the importance of maintaining human interaction in the classroom and using AI ethically within a pedagogically sound framework.

4.2 Understanding and fundamental knowledge of AI

Most participants indicated that their understanding of AI was primarily derived from informal sources such as social media, peer recommendations, and self-directed online research. They also highlighted a lack of institutional training on integrating AI into teaching and learning practices (e.g., Fei). A recurring theme among participants was the necessity for tutors to understand AI’s fundamental principles and operational mechanisms. As two participants articulated:

“Tutors should understand the basic principles of AI and know how it works to use it effectively. Tutors should also have data awareness and recognise that AI’s localisation requires local data.” (Interview, Fang)

“Tutors should first understand the underlying principles of AI and avoid misinterpreting it. Then, they should master its practical applications in teaching, such as combining it with voice recording technologies.” (Interview, Lu)

Wang echoed the view that tutors need only a basic understanding of AI functionality. However, he observed that most English tutors likely lack a clear concept of AI literacy. Some tutors (e.g., Fang, An, and Wang) also emphasised the importance of enhancing human-AI collaboration by learning how to craft precise prompts to optimise AI-generated output. For example:

“Teachers need to be proficient in using AI tools provided by the company, and they must learn how to communicate effectively with AI. By writing good prompts, they can get more accurate results and make the tool serve our teaching better. For example, when using an essay grading system, they need to give clear instructions to receive valuable feedback.” (Interview, Fang)

“If I use certain prompts, the output can be much better, because sometimes it is just a small detail, even a single word that can make a big difference in the result.” (Interview, Wang)

These insights reflect tutors’ understanding of how large language models respond to human input and underscore educators’ need to develop skills in prompt engineering. Additionally, Lu suggested that effective AI integration requires interdisciplinary knowledge, encompassing both technical AI functions and pedagogical applications. This includes the ability to critically evaluate AI-generated content and apply AI-driven methods in pedagogically sound ways. While some tutors acknowledged their ability to use AI in practice, they admitted a lack of deep understanding of its underlying technical logic. One tutor illustrated a contrast between technical staff and English tutors in their approaches to AI integration:

“Compared to frontline teachers, our technical staff have shown more interest in and ability to apply AI. Some of our programmers started researching AI early on, which is why we were able to integrate speech recognition into our app for post-lesson pronunciation correction. As a live-streamed English tutor, my approach is different. The technical staff aim for precision, and they focus on refining prompts and training AI tools to improve accuracy. I prefer using AI for the first draft and then applying my own edits drawing on my ten-year teaching experience. I just need it as a good starting point, and I do not want to spend too much time fine-tuning the AI.” (Interview, Fei)

This account highlights the differing priorities between technical staff and tutors. While technical teams concentrate on optimising AI performance, tutors leverage AI primarily as a support tool for instructional purposes rather than engaging deeply with its technical intricacies.

4.3 AI application and evaluation

Participants identified several key factors influencing their adoption and selection of AI tools, including accessibility (e.g., Tang), content quality (e.g., Yuan), perceived usefulness of functions (e.g., Fei), ease of use (e.g., Guo), and affordability (e.g., Yuan).

Most tutors reported utilising AI to assist with time-intensive teaching activities, particularly in developing instructional materials. Interview data suggest that AI primarily supports curriculum development, lesson planning, instruction, assessment, and reflection. For instance, Fang and Tang described how they used AI to generate extensive teaching materials:

“I share my teaching objectives and the content I plan to cover with the GenAI, and it might be able to generate a relatively structured lesson plan for me. Then, I could make some personal adjustments within that framework, which would make lesson preparation more efficient.” (Interview, Fang)

An and Fang noted that while they rely on GenAI to generate lesson plans and PPT drafts, they still need to personalise the content to align with pedagogical objectives and student needs.

Regarding curriculum development, Yuan described AI as an essential tool for retrieving resources and generating example sentences tailored to students’ proficiency levels. AI also facilitates the design of teaching activities, such as generating practice questions and creating engaging speaking exercises:

“AI saves me a lot of time for lesson preparation and helps me find suitable materials and example sentences. It also provides new teaching methods, such as designing engaging speaking activities.” (Interview, Yuan)

Beyond lesson planning, participants recognised AI’s potential to enhance teaching content by providing a richer corpus of linguistic resources and improving language expression quality (e.g., Yuan and Wang). Tang emphasised the AI’s role in speaking instruction, highlighting its ability to generate speaking demonstration texts for IELTS preparation. Guo observed that some teachers use GenAI for interactive speaking practices, supporting their language skill development.

Despite AI’s advantages, participants acknowledged its limitations. An and Fang pointed out that AI-generated content often requires revision, as it does not necessarily address specific teaching experiences or student characteristics. A further concern was that novice teachers might over-rely on AI-generated lesson plans, potentially introducing errors into their teaching. Wang similarly noted that while AI is useful for gathering background knowledge on unfamiliar speaking topics, its effectiveness in classroom teaching remains limited:

“When it comes to using AI in the classroom, it might not be that effective. In class, students want to learn knowledge of the language, and test-taking strategies. These are areas where AI cannot provide much help.” (Interview, Wang)

These findings suggest that while tutors generally perceive AI as a valuable tool for enhancing educational practices, they also advocate for a more structured approach to AI integration. For instance, Fang proposed the development of an ‘AI Pedagogy’ to establish best practices for AI-assisted teaching:

“AI has brought many opportunities for English tutors. However, there are also challenges. I feel that many existing teaching methods are still outdated. I’m not saying they’re without value, but they are somewhat old-fashioned because they don’t include the AI dimension, what we might call ‘AI pedagogy’. It’s still unclear how we should approach and integrate AI into our teaching, but this is worth discussing and exploring.” (Interview, Fang)

Overall, the study underscores the need for structured AI literacy training for tutors, critical evaluation skills for AI-generated content, and further discussion on best practices for AI-assisted pedagogy.

4.4 Considerations of AI ethics

During the interview, participants discussed various ethical implications of AI, including inaccuracy, privacy, and anxiety about job security. First, participants reported instances of inaccuracies in GenAI responses. Fang, for example, expressed concerns about the reliability of GenAI:

“It is typical for AI to make mistakes. When it does not understand something, it may also produce hallucinations. If you are conducting research or dealing with important issues, having some inputs with hallucinations and biases is not good.” (Interview, Fang)

Due to the absence of a precise accountability mechanism for AI tools, Guo raised concerns about potential biases in AI-driven classroom assessment tools. She noted that such tools often prioritise quantitative indicators and rely on surface-level data to predict learning outcomes, potentially overlooking critical qualitative aspects such as learners’ backgrounds and psychological states. Beyond content evaluation, participants emphasised that ethical considerations are an integral component of AI literacy. Most participants stressed the importance of teachers making informed decisions regarding when and how to integrate AI into their pedagogical practices.

Second, participants also voiced concerns about the privacy and potential misuse of personal data. Guo, for instance, highlighted issues related to registration forms and consent checkboxes that require users to provide personal information, such as phone numbers and identity documents, without clearly specifying how this data is used. She further raised concerns about technologies such as facial recognition, webcams, and video calls, which, although seemingly innocuous, could be exploited by unauthorised surveillance software. Fei echoed these concerns, particularly regarding data security and privacy, stating:

“Regarding data privacy, such as student records or even my own voice, there are concerns. I think the increasing use of AI in daily life and its growing precision can lead to potential privacy breaches. Platforms like ByteDance and Didi, which integrate AI, highlight this risk. Sometimes, after discussing a topic with a friend or conducting a search, I receive highly targeted recommendations from shopping apps. Or even before I have fully formed my thoughts, related suggestions appear. This targeted precision is disconcerting.” (Interview, Fei)

Third, participants expressed apprehension regarding the impact of AI on job security in English language education. Yu observed that routine, repetitive tasks such as grading assignments and providing feedback were at high risk of automation. She argued that to remain relevant, teachers must enhance their pedagogical approaches by integrating AI while emphasising creativity, which is an aspect that AI cannot replicate.

Fei discussed how AI-powered foreign language education disrupted the live-streamed English teaching industry. She cited examples such as Pocket Tutor and AI Foreign English Teacher, which simulate native North American teachers’ voices and offer students cost-effective alternatives for daily language practice. While these AI-driven platforms enhance affordability and accessibility, they also present a competitive challenge to traditional live instruction.

Despite these concerns, many participants acknowledged the limitations of AI. Some (e.g., An, Yuan, Lu, and Wang) noted that while AI can support education, it lacks essential human attributes such as emotion and responsibility. As An stated:

“The drive of responsibility makes teachers accountable, and I feel very happy when students perform well, and I feel sorry when they encounter difficulties. I share meaningful connections with students, which is an important part of my teaching life.” (Interview, An)

Similarly, Yuan emphasised:

“A teacher’s ability to empathise, understand emotions, and provide emotional support remains a key characteristic that cannot be replaced by AI.” (Interview, Yuan)

4.5 Professional development

Participants unanimously agreed on the importance of maintaining an open mindset and continually enhancing their AI literacy. However, opinions varied regarding the necessity of formal organisational training. Additionally, some participants identified external support, particularly government policies, and third-party educational content, as crucial to fostering comprehensive AI literacy development.

Most participants underscored the critical role of ongoing AI literacy development in effectively integrating AI into teaching. Yuan, for instance, emphasised the need for teachers to engage with AI technologies proactively. Despite initial challenges, he believed that consistent practice could significantly improve their proficiency:

“Teachers should maintain an open mindset to understand AI applications in teaching. Although there may be difficulties, continuous learning and practice can enhance our AI capabilities.” (Interview, Yuan)

Some participants expressed differing perspectives on the value of structured AI training programmes within institutions. Several participants recognised the benefits of formal training. Lu, for example, viewed the training offered by his organisation as advantageous because it provided targeted, practical examples tailored to specific teaching contexts. He also argued that institutional training could foster a culture of innovation regarding AI use in education:

“Because AI usage varies by subject, organisational training can effectively summarise best practices for specific subjects. It provides teachers with domain-specific practices they can directly implement. Additionally, sharing innovative ideas within the organisation can foster an environment conducive to AI innovation and openness among teachers.” (Interview, Lu)

Similarly, An noted that attending AI training sessions helped her recognise AI’s potential in her teaching:

“Before the training, I primarily used AI tools for paraphrasing without systematically considering their teaching applications. After attending the training, I became aware of AI’s potential in areas such as providing writing feedback and oral instruction, directly benefiting my teaching. I found this very inspiring.” (Interview, An)

Conversely, some participants questioned the necessity and efficacy of formal training. Guo, for instance, contended that practical necessity would naturally drive teachers to develop AI literacy independently:

“If AI use becomes mandatory, people can learn quickly without formal training. Just like when touchscreen smartphones appeared, no extensive training was necessary. Teachers do not need in-depth expertise; basic operational skills suffice.” (Interview, Guo)

Wang preferred informal, self-directed learning methods, such as short instructional videos and content shared on Xiaohongshu (RedNote, a Chinese social networking platform). He also cautioned against prematurely institutionalising AI training programs, suggesting that some might be commercially motivated:

“Training might not be particularly necessary right now. Prompt-writing skills and using AI tools can be self-taught through platforms like Xiaohongshu. Formal training systems haven’t yet matured.” (Interview, Wang)

Beyond institutional AI training, participants highlighted the critical role of external support, particularly from government agencies and third-party content creators, in advancing AI literacy among educators. Lu advocated for increased government support in the form of policies, funding, and regulatory frameworks to ensure the responsible integration of AI into education:

“Governments should enhance policy support, funding, and regulation for AI applications in education. Clear standards need to be established to guide and regulate the use of AI within educational contexts.” (Interview, Lu)

Participants also emphasised the value of third-party educational content in supporting professional development. Concise, accessible instructional materials were seen as effective in demonstrating AI applications to teachers. Wang, for instance, suggested:

“Short, clear, practical demonstration videos could effectively communicate AI’s usefulness, illustrating its benefits clearly and succinctly.” (Interview, Wang)

5.1 Practitioners’ perceptions of AI: a human-centred perspective

Regarding their perceptions of AI, all participants emphasised the importance of adopting a human-centred mindset when selecting and using AI tools. This aligns with UNESCO’s (2024) framework, which advocates for AI to be assistive rather than replace or undermine teachers’ agency and accountability in pedagogical decision-making (Mouta et al. 2025). While the practitioners highlighted their irreplaceable role in making instructional decisions and providing emotional support, some also expressed concerns about job insecurity. For instance, GenAI tools, acting as dialogic partners (Liu 2024; Wegerif and Major 2019), can offer rich feedback on students’ written assignments and oral practice. These tools provide cost-effective alternatives for learners, potentially reducing course enrolments and profits in the shadow education sector. However, this concern was reframed as an external stimulus that motivated practitioners to embrace the opportunities presented by AI and develop AI-integrated pedagogical approaches.

Participants paid little attention to the broader systemic and societal implications of AI use and educators’ social responsibilities. It is crucial that practitioners understand the risks and challenges associated with AI in teaching and recognise how AI may exacerbate social injustice, inequality, and structural exclusion (Dilek et al. 2025). This reflects Bali’s (2023) concept of critical AI literacy, highlighting the importance of ethical and equitable awareness in AI integration. Similarly, UNESCO (2024) stresses that teachers need to adopt a human-centred mindset and uphold their social responsibility to promote inclusivity and cultural pluralism for all students, regardless of background (e.g., gender, ethnicity, socio-economic status, and ability).

This study also revealed that not all practitioners held positive attitudes towards AI, with both external factors (e.g., institutional constraints, exam-oriented goals, and limited access to AI technology) and internal factors (e.g., personal beliefs) influencing their perspectives. For example, Guo reported that her institution enforced standardised teaching practices to meet parental and student expectations for high test performance, leaving little room for experimentation with new technologies. Professional autonomy was constrained in such cases, as standardised curricula and teaching approaches – proven effective across multiple settings – were prioritised. Given shadow education’s market-driven, customer-centred, and profit-oriented nature (Xiong et al. 2022), institutions often favour standardisation over pedagogical innovation. Additionally, Wang’s initial reluctance to use GenAI stemmed from limited access and self-perceived barriers (e.g., age and non-technical background). This aligns with Teng’s (2019) argument that teacher agency is shaped by “personal disposition and internal beliefs” (p. 203). However, Wang’s attitude became more positive after using GenAI for lesson planning, as he experienced its practical benefits (e.g., time efficiency and access to richer materials), which encouraged him to explore further applications.

5.2 Practitioners’ experiences of AI: limited classroom integration

Generative AI has the potential to support teachers in both routine and creative tasks (Firaina and Sulisworo 2023). This study provides insights into tutors’ perceptions of AI integration in teaching and their perceived AI literacy. Overall, tutors appreciated AI tools for their ability to assist with lesson planning, the development of customised teaching resources, and the provision of timely feedback – findings consistent with previous literature (e.g., Jia et al. 2022).

However, while most English tutors in shadow education recognised the potential of AI, its application was primarily confined to activities outside the classroom, such as lesson preparation and assessment, with limited use in live instructional settings. This finding underscores the complexities of AI integration in classroom practice, which is often shaped by institutional policies and the expectations of parents and students. Furthermore, many AI tools are developed without sufficient input from educators, leading to a misalignment between technological affordances and classroom realities (Luckin 2017). This, coupled with concerns about algorithmic bias and data-driven inequities (Baker et al. 2019), highlights the need for a more critical and inclusive approach to AI adoption (UNESCO 2024).

5.3 Developing practitioners’ AI literacy: a need for holistic support

For effective implementation of AI in educational settings, it is essential to provide educational practitioners with comprehensive training that covers both the technical aspects and the ethical implications associated with its uses in teaching and learning (Yeh 2025). This study has emphasised the need for a more holistic approach. As Mishra and Koehler (2006) argue in their TPACK framework, effective technology integration requires an understanding of the dynamic relationships between content knowledge, pedagogical strategies, and technological tools. Therefore, standalone technical training is insufficient; AI integration should be context-sensitive, subject-specific, and strategically applied. For meaningful AI integration, practitioners require a more critically reflective approach to AI literacy – one that empowers them to use AI tools effectively and examine their ethical, pedagogical, and societal implications through a critical co-discovery approach (Dilek et al. 2025).

The findings of this study have important implications for practice and policy. First, tutors require sustained, context-specific professional development that fosters critical AI literacy, moving beyond technical training to include pedagogical relevance and ethical reflection. Second, educational technology developers should engage in inclusive co-design with educators to ensure that AI tools align with real classroom needs, remain accessible, and uphold ethical standards (Baker et al. 2019). Finally, educational institutions should invest in long-term, integrated strategies that embed AI literacy within both pre-service teacher education and ongoing professional learning, as emphasised by Dilek et al. (2025) and UNESCO (2024). By fostering a more comprehensive and critically informed AI literacy framework, educators can enhance their professional autonomy and contribute to a more inclusive, ethical, and pedagogically sound integration of AI in education.