AI as a reflective coach in graduate ESL practicum: activity theory insights into student-teacher development

ELTE practicums typically involve supervised teaching practice, systematic, or structured observation, and a familiarisation process to the teaching context. These experiences vary greatly in intensity (Gebhard, 2009, p. 250) but are pivotal for reflective practice (RP). Through these experiences, STs are encouraged to examine their teaching practices critically with the goal of continuous improvement. Research has established that RP is a complex, dynamic and highly personalised cyclical process that involves deep thinking, critical analysis and continuous improvement that benefits significantly from collaborative efforts and external input (Godínez Martínez, 2018, p. 443; Korucu-Kis & Demir, 2019, p. 1249). However, RP is often limited by inadequate feedback mechanisms that fail to promote critical selfreflection (Godínez Martínez, 2018, p. 430; Ochieng’ Ong’ondo & Borg, 2011, p. 522). Furthermore, Walsh (2013, p. 12) suggests that many ST and supervisors often lack both the knowledge and training in practising effective RPs.

However, the practicum supervisor plays a key role in guiding ESL STs through structured reflective processes to support the application of what is learned in their courses to the realities of their practicum teaching context (Gebhard, 2009, p. 254), especially in early training stages (Gan & Lee, 2015, p. 252; Ochieng’ Ong’ondo & Borg, 2011, p. 522). However, challenges can include conflicts with mentors (Mohd Ariff Albakri et al., 2017, p. 231), theory-practice gaps (Al-Adwani & Al-Shammari, 2022, p. 100) and time constraints for feedback (Celen & Akcan, 2017, p. 266). Drawing from my experience in running graduate practicum courses for many years, I have also observed these challenges in current RP processes most notably feedback delays.

This research is situated within the higher education and ELTE context and acknowledges the convergence of technology, pedagogy and ESL ST needs. The studypositions itself at the intersection of technology and language education, emphasising the potential of AI in enhancing and potentially transforming ESL teacher training and development. My positionality attempts to value both the depth of academic inquiry and the practical implications of AI in practicum settings.

Generative AI has the potential to provide personalised instruction and feedback, assisting ST in identifying and addressing gaps in their understanding, thereby enhancing their critical thinking and pedagogical skills (Mollick & Mollick, 2022, p. 2). The utilisation of AI as a reflective coach in ESL practicums could support the development of metacognitive abilities, offer nonjudgemental feedback spaces (Mollick & Mollick, 2023a, 2023b, p. 18) and enhance technological awareness. However, it’s crucial to balance AI with human mentorship (Mollick & Mollick, 2023a, 2023b, p. 5) ensuring AI complements rather than replaces human insights. This research study aims to contribute to the discourse on AI’s role in enhancing STs’ RPs and pedagogical decision-making by integrating AI as a reflective coach into ESL practicums.

My search strategy for this narrative literature review was exploratory and adaptive to the evolving research area. Initial searches (Table 1) in traditional databases like EBSCOhost and Education Resources Information Center (ERIC), while yielding numerous papers on reflection and practicums, often conflated practicums with broader PD. While there was a large number of papers on incorporating technology in practicums, the specific focus on AI as a reflective coach in ELTE was barely represented. To address this, I utilised AI-enhanced search tools such as Elicit.com and ScholarAI, employing keywords like ‘AI as Coach’ and ‘ESL Practicum’ extracted from the research title and questions. This refined approach, supplemented by traditional databases and Google Scholar, focused on the literature pertinent to AI’s role in coaching and RP in ELTE.

Employing a snowballing method, I explored the references of key sources to uncover older, significant literature, especially those addressing challenges in practicum supervision, thus ensuring a comprehensive understanding of the topic. My inclusion criteria focused on AI in higher education, specifically in English Language Teaching (ELT) and reflective practices (RP), excluding studies not central to these themes, like those solely on COVID-19 and distance learning. The strategic use of keywords managed with Boolean operators aligned the search with my research objectives. Given the novelty of the field, I also consulted secondary literature from online platforms such as Medium. The literature was then synthesised into five key themes, structured narratively for this review.

Graßmann and Schermuly’s (2021, p. 109) concept of AI coaching as a ‘machine-assisted, systematic process’ for setting and achieving professional goals has potential in ELTE. This approach positions AI as a pivotal tool in enhancing RPs and fostering ST autonomy, using data-driven processes to customise and optimise learning experiences.

Mollick and Mollick (2022, 2023a, 2023b) explore how AI chatbots can augment the reflective capacity of STs by providing diverse examples and feedback, thereby deepening their metacognition and improving teaching practices (Mollick & Mollick, 2022, pp. 1–2; 2023a, 2023b, pp. 18–19). Bearman and Ajjawi (2023) promote the use of AI-integrated teaching strategies to develop critical digital skills and sound judgement, emphasising the balance of self-awareness and critical knowledge as essential skills for modern teachers.

Golombek and Johnson (2004, p. 309) highlight the role of narrative inquiry in resolving cognitive dissonance in teacher development, a process further enhanced by technology according to Li and Walsh (2023). AI, in this context, emerges as a promising mediator for addressing cognitive dissonance and advancing professional learning in ESL education.

Demszky et al. (2023) discuss the principles of designing automated feedback tools that balance personalised feedback with teacher autonomy. This aligns with the use of AI tools as reflective coaches, providing non-judgemental, private feedback that supports rather than evaluates teacher development.

This synthesis of key themes from the literature recognises the vital role of feedback, RP, technological integration and the unique challenges in practicum experiences within ELTE. AI emerges as a transformative tool, bridging gaps between theory and practice, enhancing RP and addressing feedback and technological integration challenges. My study aims to empirically explore AI’s role as a reflective coach, contributing novel insights into its impact on pedagogical decision-making and professional growth. This research contributes to addressing a significant need in ELTE and participates actively in the evolving conversation about the role of AI in ELT practicum courses.

This is an exploratory case study (Cohen et al., 2017, location 2496) as it is designed to investigate how STs interpret and find meaning in their experiences using AI as a reflective coach in a graduate ELTE practicum course. As participants interact with the AI tool in different ways, the researcher is required to interpret and search for meanings from their experiences (Creswell & Creswell, 2018, p.4; Merriam & Tisdell, 2015, p. 8).

In my role as the instructor of the graduate practicum course used in this study, it was necessary to address ethical considerations. To ensure impartiality and equal access, I introduced the concept of AI as a reflective coach during a class session with all the students. Following the session, a research assistant (RA), not involved in the practicum course, presented the research project to the STs. This was followed up with an email to all students in the course inviting them to participate in the study. STs who were interested in participating in the study contacted the RA directly, who in turn provided the Informed Consent Form (ICF). The RA responded to any questions about the study and maintained confidentiality of participation until after the final course grades were submitted. To be eligible, participants needed to be actively teaching in the local community as part of their practicum and have access to ChatGPT-3.5, (the free version of an advanced large language model developed by OpenAI), or Copilot (formerly known as Bing Chat). Participants who signed up to participate received the following links from the RA: the initial questionnaire, the AI structured prompt and the PostReflection Questionnaire (PRQ). All interactions and tasks were conducted remotely with the RA acting as an intermediary should participants have any questions. Ethical clearance was obtained by the Department of Educational Research, Lancaster University, and reviewed and cleared by the Saint Mary’s University Research Ethics Board, (Appendix A).

This study involved 26 international graduate ESL STs from a Canadian university, who were offered the opportunity to participate anonymously. Twelve STs volunteered, representing a range of practicum experiences, from teaching refugees in the community with limited support to completing a Certificate in Teaching English to Speakers of Other Languages (CELTA) course with more structured feedback. Despite the same reflection prompt being used by the STs, they used the AI Coach in unique ways, demonstrating its ability to adapt and offer valuable help in a wide range of contexts. Although most participants were fully engaged, incomplete data from some (detailed in Table 2) could impact the study’s findings.

Data collection was conducted anonymously over 4 weeks in November/December 2023, involving up to three interactions per participant with ChatGPT-3.5 or Copilot (formerly Bing Chat). These AI tools were selected for their universal accessibility. Initially, STs completed a questionnaire detailing their previous experience with these tools, then engaged with a structured metacognitive prompt adapted from Mollick and Mollick (2023a, 2023b, p. 18) designed for AI-mediated post-class reflections (Appendix B). Creating a chain-of-thought—’a series of intermediate reasoning steps’ (Wei et al., 2023, p. 1)—prompt ‘significantly improves the ability of large language models to perform complex reasoning’. In this study, the prompt allowed the AI tool to assume the role of a coach and aimed to ensure as much as possible consistency in AI responses. The AI coach afforded STs time to think more deeply on their practice and in turn ‘nurture students’ ability to learn independently and critically’ (Mollick & Mollick, 2023a, 2023b, p. 19). Conversations were coded as 1.x, 2.x and 3.x, where ‘x’ denotes an anonymous participant, and structured around specific, guided questions posed by the AI (Appendix C). The AI’s role was to guide and prompt reflection, encouraging STs to articulate and analyse their teaching practices and challenges independently.

After each interaction with the AI coach, STs completed a PRQ. The purpose of the PRQ was for STs to reflect on their experience. As it was not possible to organise semi-structured interviews, STs had the option to answer additional questions related to Activity Theory in the context of the study and to explore their experience more deeply. The STs were asked to answer these questions after their third interaction with the AI, although quite a few of the STs answered them after each interaction. The primary focus of the research was to analyse the STs’ AI interactions and questionnaire responses in a way that would ensure participant anonymity.

The study has several limitations. The inability to conduct semi-structured interviews or track STs individually limited the opportunities for gaining deeper, personalised insights into their AI experiences. The diversity in teaching contexts and modes, such as teaching online and in-person classes across varied learner demographics, presented challenges in standardising the findings. Moreover, complexity arose from the participants’ differing ELT experience levels, including those completing a CELTA course as part of their practicum. While prioritising participant anonymity was essential to minimise instructor bias, this approach restricted the depth of data collection. Despite these limitations, the study effectively met its objective of exploring AI’s role as a reflective coach to increase metacognition, demonstrating its adaptability and potential in varied practicum teaching contexts.

A thematic analysis was conducted on qualitative data from questionnaires, AI interactions and PRQs. Initially, a qualitative summary of participants’ backgrounds and attitudes towards AI as a tool for reflection provided context for understanding their perspectives. The Reflexive Thematic Analysis (RTA) approach, Byrne (2021), was adopted to analyse STs’ interactions with the AI Coach, emphasising reflexivity in the analytical process. This entailed a six-phase process, as proposed by Braun and Clarke (2012, 2013, 2014, 2020, as cited by Byrne, 2021, p. 1398), involving continuous reflection on my own assumptions, perspectives and biases to better understand how they might influence my interpretation of the data. An inductive approach allowed themes to emerge naturally from the data, without reliance on preconceived theoretical frameworks. The analysis was recursive and iterative (Byrne, 2021, p. 1397), involving the continuous coding of data in Microsoft Word and Excel. This continuous process of analysing and interpreting the similarities and subtle differences in the data allowed themes to emerge.

While the six-phase RTA provided a structure for analysing the conversations with the AI Coach, an experiential orientation to the final section of the PRQ (Questions 11–18) was taken, as this valued STs’ own reflections and feelings about their experiences with the AI. This placed a specific emphasis on understanding the STs’ subjective experiences and interpretations, acknowledging their teaching contexts and reflections, (Byrne, 2021, p. 1396). A detailed question-by-question analysis was conducted, integrating the Activity System framework to capture the STs’ experiences. The themes were synthesised and subsequently analysed through the lens of the research questions, themes from the literature review and Activity Theory principles.

A crucial part of the analysis was to acknowledge the broader social, cultural and historical contexts of the data. These reflexive, inductive and context-sensitive analytical methods, underpinned by Activity Theory, sought to identify patterns in how STs critically evaluated and evolved their teaching approaches.

The goal was to find trends in the way STs critically assessed and improved their teaching practice by using thoughtful, flexible and context-aware methods of analysis, which were based on Activity Theory.

The analysis of the initial survey, AI conversations and PRQs (Q1-10) revealed that using ChatGPT/Copilot as a metacognitive tool for self-reflection provided personalised feedback, an opportunity to organise their thoughts, helped STs ask relevant questions about their teaching, offered useful teaching suggestions and facilitated self-reflection. Moreover, some STs stated that the AI coach was successful in guiding deeper reflection and demonstrated potential in supplementing traditional feedback methods. They appreciated its ‘timely and non-judgemental feedback’, which made them feel comfortable and supported in their interaction.

Additionally, some teachers were surprised by the AI’s empathetic communication, which created a sense of emotional connection and rapport:

‘I feel comfortable asking questions to ChatGPT because it will never make me feel embarrassed, no matter how easy my question is’ (PRQ 7).

‘I was quite surprised with how “feeling” the AI felt, and barring the lack of person-to-person conversation, I thought using the AI was quite beneficial overall’ (PRQ 8).

Conversely, some participants raised questions about AI’s ability to provide more insightful and theoretically informed feedback or to offer alternative solutions to their problems. One respondent anticipated more insightful and theoretically grounded feedback, expressing concern about the AI’s affirmative nature without offering alternative solutions. Similarly, some questioned Al’s ability to understand fully the cultural dynamics at play in the classroom. Finally, another respondent stated that the AI did not offer any new or innovative ideas that they had not already considered or encountered elsewhere.

The following themes, (Table 3–7) were derived from the STs’ conversations with the AI (Appendix C) and responses to questions 11–18 of the PRQ. Activity System diagrams, (Figures 2–6) provide a visual representation of each theme, and their relevance to the research questions is highlighted. The inclusion of STs’ comments illustrates how each theme is grounded in their actual experiences and reflections.

Figure 2:

Theme: Maximising Instructional Time & Enhancing Reflective Practice (RP)

Figure 3:

Theme: Reflective Practice (RP), Self-Awareness (SA), and Professional Development (PD).

Figure 4:

Theme: Addressing Teaching Challenges and Support for Graduate ESL Student-Teachers (ST).

Figure 5:

Theme: Professional Development (PD) and Confidence in Pedagogical Abilities

Figure 6:

Theme: Expectations for Future Use and Development

The aggregation of findings provides a summary of how STs used and evaluated their interactions with the AI coach in their respective teaching contexts. In the following section, I will discuss these perspectives in relation to the research questions, examine their alignment with Activity Theory and explore their broader implications within the ELTE.

When synthesising the insights from my research, it is evident that the responses to both research questions are deeply interconnected, each shedding light on the use of AI in ELTE.

My first research question explored how an AI Coach influences the depth and quality of graduate ESL STs’ RPs. Here, I found that the AI Coach significantly enhanced RPs, providing a structured and data-driven approach to self-evaluation and pedagogical refinement. This aligns closely with the responses to my second research question, which focused on the AI Coach’s mediation of effectiveness and challenges in critical evaluations of teaching actions. In this context, my findings revealed that the AI Coach not only facilitated a deeper understanding of teaching practices but also supported STs in effectively navigating and addressing pedagogical challenges.

The convergence of these insights highlights a broader, more comprehensive role of AI in ELTE. It emphasises the AI Coach’s capability not just as a passive tool, but as an active mediator that can significantly shape the teaching and learning process. In both RPs and critical evaluations, the AI Coach emerges as a crucial component in enhancing the RPs of ESL teachers.

Therefore, this study contributes to the understanding of AI’s role in ELTE by providing empirical evidence of its impact on both RPs and critical evaluations. This dual role of AI offers a new perspective on how technology can be integrated into a range of contexts and settings to foster not only instructional effectiveness but also ongoing professional growth and development in the field of ELT.

My research highlights how AI can potentially support and improve the teaching skills and self-awareness of graduate STs in different ESL contexts. The integration of AI coaching tools into practicum programmes offers significant benefits. These include personalised, conversational reflection and analytical insights, a point illustrated by one participant’s experience: ‘It’s very beneficial to be reminded of your initial goals and hearing it from “someone” helps much more than engaging with a written note’ (ST, PRQ 8).

In practicum settings, the ‘anytime feedback’ and deeper reflection facilitated by an AI coach can significantly enhance the learning experience for STs and develop metacognition. This aligns with the overall ‘object’ of the study, as identified in the overarching activity system. The flexibility to access the AI Coach on various devices allows STs to engage in more profound reflection on their practice and receive effective, actionable and timely feedback.

However, it should be stressed that despite the integration of AI, the role of the supervisor remains crucial. In my future practicum courses, utilising ChatGPT, I will request that all STs share the summary table from their reflection, generated by the AI Coach, and a link to the conversation, so that I can review the conversation when necessary. If used for the duration of the practicum, the metacognitive development of the ST would be monitored more easily and effectively, affording the supervisor time to intervene when required. This approach allows supervisors to focus on key aspects of the ST’s reflection and provide necessary input. Termed as ‘augmentation of traditional coaching with AI’ (Graßmann & Schermuly, 2021, p. 119), this method also aligns with the overall ‘object’ of the study by combining the benefits of both traditional and AI-based approaches, allowing the practicum supervisor to have a more responsive and timely feedback mechanism.

In conclusion, integrating AI as a reflective coach into ESL practicum courses holds significant promise for enhancing learning experiences and fostering the development of informed and reflective language teachers. As one participant noted:

‘The future role of AI as a Reflective Coach in shaping teaching practices in our community appears to be promising and transformative. As technology continues to advance, AI can play a huge role in enhancing professional development and refining teaching methodologies’ (ST, PRQ 5).

This sentiment aligns with the study’s findings, emphasising the potential for AI to become a pivotal component in the continuous PD of ESL teachers while also supporting supervisors in effectively managing the demands of practicum programmes.

The exploration of AI in ELTE within this study has not only provided valuable insights but also highlighted several potential avenues for future research. An area of particular interest is the long-term impact of AI as a reflective coach in ELTE. Longitudinal studies, tracking the development of STs over time, would be instrumental in offering a more comprehensive understanding of how sustained use of AI influences teaching practices and pedagogical development.

Another area for future inquiry could involve examining the dynamics between STs and their supervisors in the context of utilising AI as a reflective coach. This aligns with the practicum challenges identified in the literature of the works of Al-Adwani and Al-Shammari (2022), Celen and Akcan (2017) and Mohd Ariff Albakri et al. (2017), which touch upon the complexities of practicum relationships. One of the participants aptly noted, ‘It may help tutors [supervisors] to promote self-awareness of the teaching practices and the different strategies to overcome challenges. The tool forces you to really think about your own experience and how things can be improved’ (ST, PRQ 12), suggesting that AI can potentially make interactions with supervisors more reflective and focused.

Customisation and personalisation of AI tools in ELTE also emerge as possible areas for further exploration. Feedback from some participants included critical comments such as the AI ‘leads teachers to find problems themselves’ and would have preferred ‘more advice during the reflection coaching’ (ST, PRQ 2). While the purpose of the prompt was to provide a metacognitive exercise to encourage STs to think deeply about their practice, as opposed to simply providing them with answers to their teaching questions, future research could explore how AI tools might be adapted to offer a balanced mix of guided reflection, critical feedback and suggestions for teaching. This may require more than one chain-of-thought prompt (Wei et al., 2023, p. 1), and more skill is required of the ST to question the AI more and move away from the given prompt. Lastly, as AI becomes increasingly integrated into ELTE settings, ethical considerations, especially when working with international STs, such as potential bias (Western, gender-based, language), or cultural acceptability (Jamal, 2023, p. 142), must be rigorously examined to ensure responsible and effective use of the technology.

These are just a few possible areas of study that would extend the understanding of AI’s capabilities and its role in enhancing RPs, critical evaluations and the professional transformation of STs in ELTE. Finally, while this study utilised free and readily available AI tools, ChatGPT-3.5 and Copilot (formerly Bing Chat), the effectiveness of harnessing the power of subscription AIs, such as ChatGPT-4 and Gemini Advanced in these areas provides additional possibilities.