Applying the AI assessment scale in writing and translation for EFL Open Access

English as a Foreign Language (EFL) instruction has historically demonstrated an aptitude for integrating new technologies into practice and a willingness to experiment with techniques that may provide new ways to improve language acquisition. This is evidenced by the longstanding field of Computer-Assisted Language Learning (CALL), which has been developing since the 1950s (Tafazoli and Golshan, 2014). Since the public release of OpenAI’s ChatGPT in 2022, a major focus of technology in EFL has been on artificial intelligence (AI) and generative artificial intelligence (GenAI). GenAI (including ChatGPT) models are a subset of AI applications and have developed rapidly in recent years, and new models have the ability to produce human-like, conversational text, instantaneous translation into multiple languages and the production of multimedia outputs, including images, video and audio. In this text, we refer to both AI tools more broadly and GenAI tools at specific instances.

However, it is important to note that AI technologies have been in use in the EFL context since prior to 2022. These include AI-powered digital writing assistants such as Grammarly, which can help provide suggestions for textual development (Grammarly, 2022). Such tools have been shown to improve grammatical accuracy, confidence and autonomy, and enhance feedback quality (Barrot, 2020; ONeill and Russell, 2019; Thi and Nikolov, 2021). Other similar technologies include automated paraphrasing tools, which can assist in rephrasing text (Roe and Perkins, 2022) as well as machine translation applications such as DeepL. It remains to be seen whether these technologies will retain a use case in the EFL classroom, or whether they will be supplanted by GenAI tools that can complete all such tasks in a single user interface.

In the broader field of education, the use of GenAI has led to consternation and debate. Research has evolved on the specific nature of risks regarding GenAI and plagiarism, or academic integrity violations (Cotton et al., 2023; Dehouche, 2021; Eaton, 2023; Kumar and Mindzak, 2024; Perkins, 2023; Wilder et al., 2021). Institutions and policies are now showing a shift towards acceptance of the use of GenAI in specific uses for text construction (Perkins and Roe, 2023a, b). The acceptance of certain ethical uses of AI to assist with the writing process is of value to EFL learners, given that writing in a second language causes cognitive stress (Gayed et al., 2022). Despite the potential value of decreasing learner stress, significant existential risks have been attributed to GenAI. In addition to the culturally biased output produced by such models, which may affect the validity of scientific knowledge (Messeri and Crockett, 2024; Roe, 2025), there is concern that the significant demand and techno-optimism associated with GenAI, combined with limited oversight from regulatory bodies or governments, will lead to the rapid depletion of natural resources (Bashir, 2024).

While it is important to acknowledge these highly concerning impacts, educators must also be pragmatic and deal with the issues that are currently arising. There is limited guidance on how to use AI in education; at the same time, many students and teachers are using freely available models without sufficient support and preparation. In the U.S., 20% of all students surveyed were using ChatGPT for schoolwork (Sidoti and Gottfried, 2023). Consequently, there is an urgent need for clearer guidance and frameworks to help learners and teachers use these tools effectively and critically.

One of the frameworks introduced to support educators with this goal was the AI Assessment Scale (AIAS) (Furze, 2024; Perkins et al., 2024, 2025). The AIAS is designed as a flexible, five-point scale intended to guide educators and students on different types of AI use cases in educational assessment. The AIAS is not intended to be “enforced” using a detect and catch pedagogy but instead is meant to be used to guide conversations, assessment redesign and encourage transparent experimentation with GenAI tools where appropriate. However, while the AIAS was originally conceived to help deal with the disruptive changes brought about by GenAI tools in assessment, it can be adapted to other pedagogical use cases. Specifically, the AIAS can be utilised to help guide learners and educators on how to integrate AI at different levels into the classroom. In this conceptual paper, we make the case for this approach by conducting a holistic and comprehensive review of the literature related to GenAI and EFL instruction to frame our argument. Following this, we describe the AIAS in detail and examine what the AIAS is and is not. We then present an approach that can be used to facilitate the classroom use of GenAI in an English teaching context, with a specific focus on Levels 2, 3 and 4 of the AIAS. Our hope is that such an approach can, in future be validated through empirical data collected in the classroom context.

As with other disciplines, AI has had a large impact on EFL teaching and learning (Alshumaimeri and Alshememry, 2024). As these technologies have developed, so too have stakeholders’ assessments of what students need to function in an EFL context rife with digital tools. For this reason, there has been an increased focus among practitioners and researchers on promoting various forms of literacy, beginning with technological literacy and progressing through media, digital and finally to AI literacy today (Fan and Zhang, 2024; Roe et al., 2025b). When adapting the AIAS to support EFL instruction, it is important to first explore the evidence base for how, why and when AI may support EFL learning and teaching.

There is a large and developing body of literature on AI and GenAI in EFL education. Systematic reviews have identified that most studies in this area focus on writing skills (Alshumaimeri and Alshememry, 2024), while comparatively less focus has been given to other macro-skills, such as reading, listening and speaking (Lo et al., 2024). Additionally, most studies focus on one specific GenAI tool, ChatGPT and on text rather than multimodal outputs, which is perhaps unsurprising given the first-mover advantage that ChatGPT holds. Despite this advantage, it is anticipated that future applications will explore other multimodal GenAI outputs and a range of other AI applications, such as NotebookLM, Claude, or Gemini. Furthermore, there is evidence that multiple digital tools can benefit student learning in an EFL context, including Paperpal, Quillbot and WordTune (Marzuki et al., 2023). Furthermore, as synthetic media and deepfake applications advance in terms of ease of use and cost-effectiveness, more attention may be given to this area. Some have argued that the integration of AI into education is inevitable (Koraishi, 2023), although this is a hotly debated issue. However, there is widespread recognition that AI is not a wholly positive technology, leading to the common metaphor of AI as a “double-edged sword” (Derakhshan and Ghiasvand, 2024).

Research has both postulated and validated several positive outcomes for learners regarding interventions related to GenAI tools in classroom settings. A meta-analysis of 64 studies demonstrated that ChatGPT interventions significantly enhanced academic performance, affective-motivational states and higher-order thinking propensities while reducing mental effort, although it had no significant impact on self-efficacy (Deng et al., 2024). A review of 21 experimental and quasi-experimental studies found that EFL instruction with the use of AI chatbots was more effective than that without them (Wu and Li, 2024). Among the many tasks that ChatGPT can achieve in relation to language learning, Kohnke et al. (2023) identified correcting and explaining mistakes, creating multiple genres of text, developing quizzes, giving definitions and several more, indicating a wide and broad range of tasks that GenAI can contribute to. Similarly, other authors highlight that these kinds of tools may benefit vocabulary acquisition, pronunciation and student engagement (Algraini, 2024; Mohamed, 2024). Use of GenAI tools such as ChatGPT may also benefit motivation for EFL learners (Huang and Mizumoto, 2024; Yuan and Liu, 2025), improve willingness to engage in conversation (Yang et al., 2022), boost overall engagement (Wang and Xue, 2024), improve students’ abilities to enter into a “flow state” (Zhang et al., 2021) and enable EFL students to set higher goals (Guo and Li, 2024) and express their ideas in English (Zhao, 2022) and collaborate (Teng, 2024). GenAI technologies may assist with language learning beyond the classroom through Informal Digital Learning of English (IDLE) (Liu et al., 2024) and visually embodied GenAI chatbots may enhance the emotional aspects of language learning using human-like avatars (Wang et al., 2024).

As AI-powered language learning tools may be able to adapt to individual needs (Mohamed, 2024), personalisation of learning materials, for example, text reading passages adapted to learner profiles, is possible, reducing both the burden on material design for the teacher (Koraishi, 2023) and providing increased speed and quality of learner feedback (Dai and Liu, 2024). Research has also explored how GenAI and ChatGPT are viewed by EFL teachers, with Slamet (2024) finding that teachers express a positive perception of ChatGPT to benefit student access to linguistic resources, and Ulla et al. (2023) finding similar positive intentions to use ChatGPT for lesson planning and resource creation. In writing and feedback, ChatGPT has been shown to have higher reliability coefficients for marking EFL work than human raters (Li et al., 2024).

Despite findings in EFL research demonstrating a multitude of positive benefits for learners across multiple skills and areas of language acquisition, there is a consensus in the literature that AI should not be implemented in an unstructured way, and tools such as chatbots can, in some cases, constrain rather than improve language learning (Jeon, 2024). Furthermore, the literature suggests that AI use requires teachers to engage in professional training to be prepared to embed AI into EFL where appropriate (Jiang, 2022), which must be done with the awareness that such technologies can provoke strong emotional reactions from teachers and learners (Shen and Guo, 2024; Yang and Zhao, 2024; Yin et al., 2024). Consequently, a critical approach to implementing GenAI in EFL practice is essential.

GenAI models, such as ChatGPT, may also lack cultural sensitivity (Werdiningsih et al., 2024) and present a worldview aligned with those in the training data, which by its nature, focuses on Western cultural norms. The novel nature of AI chatbots may generate fear and anxiety among EFL learners (Yang and Zhao, 2024). Therefore, teachers must play a pivotal role in guiding technology in the classroom (Mohamed, 2024) while aiming to develop learners’ critical literacy (Darwin et al., 2024). Additional concerns to be aware of when introducing AI to the EFL classroom include the possibility of impacting creativity and spreading false information (Derakhshan and Ghiasvand, 2024) as well as overreliance on the tools (Darvishi et al., 2024; Gao et al., 2024; Yuan et al., 2024) or the dishonest usage of AI-powered writing tools (Roe et al., 2023). Accessibility and inclusion are concerns, as free versions of GenAI tools have user limits when compared to the paid version of the same tools (Nizzolino, 2024); therefore, instructors must focus on providing equal learning opportunities (Kohnke, 2023) and mitigating the pressure for learners to subscribe to premium models (Yuan et al., 2024). These myriad challenges and risks highlight the importance of a structured, cautious approach for integrating AI into EFL.

Substantial research exists on AI integration in translation pedagogy, including AI-mediated translation (Amano et al., 2025) and human-AI collaboration trust in professional translation workflows (Rivas Ginel and Moorkens, 2025). These studies demonstrate that structured AI integration can support translation but require critical evaluation of both the quality of outputs as well as the broader impact on society.

The AIAS was designed as a simple and practical tool to help educators and students deal with the sudden emergence of GenAI tools and their astonishing capabilities in accomplishing a variety of assessment tasks. This was a result of the fact that a small proportion of students could use GenAI to subvert the validity of assessment and achieve a passing level without putting in the required effort and time themselves (Thompson et al., 2023) and a broader view that, as disruptive technologies develop, assessment would have to fundamentally change. The view that assessment in higher education must be re-evaluated has now received widespread attention in the literature (Bearman et al., 2024; Mao et al., 2024; Rasul et al., 2024; Thanh et al., 2023; Thompson et al., 2023; Xia et al., 2024). This has become even more critical as the abilities of GenAI tools to tackle assessment have grown even further since their first release, including demonstrating the ability to pass high-stakes medical and legal examinations (Head and Willis, 2024; Newton et al., 2024; Newton and Xiromeriti, 2024).

The AIAS, in its original incarnation, offered a five-point scale that was intended to guide educators and students on how to structure AI and GenAI use within assessments, while maintaining transparency and academic integrity, based on the principles of clear, simple and two-way communication, and thoughtful assessment redesign rather than an unenforceable set of “rules”. This is important given the gap between learners' and teachers’ opinions on how to use AI (Smolansky et al., 2023). The lowest level of the AIAS was intended to cover tasks where security was vital and no external tools were allowed (No AI), while the fifth level (Full AI) was intended to allow learners to make full use of the affordances of the new technology to meet specific assessment goals. The AIAS has been adopted by universities and K-12 institutions, and promoted by governmental agencies as a potential tool to assist educators in dealing with the implications of GenAI (Lodge, 2024). It has been translated into multiple languages and modified significantly for various educational contexts (Perkins et al., 2025). Furthermore, empirical research has demonstrated that implementing the AIAS in a tertiary context can stimulate a shift in pedagogical practices on behalf of assessors and promote academic integrity (Furze et al., 2024).

At the time of publication, the AIAS has been updated to replace the fourth level (AI and Human Evaluation) with the previous fifth level (Full AI) and the addition of a new level (AI Exploration) (Perkins et al., 2025). The rationale behind this change is that since the original framework was released, GenAI has become so commonplace and so advanced that there may now be a need to actively promote and even expect the use of such tools in assessment in an exploratory manner. To this end, Full AI (in which all use of AI is permitted) is now replaced with “AI Exploration,” which may involve assessment design that necessitates engagement with GenAI to meet learning outcomes. Stylistically, changes were also made to the colour scheme to distinguish it from a “traffic-light system,” which may implicitly have assumptions that lower levels of the scale, previously coded red, suggested a negative connotation and that higher levels of the scale had more positive connotations. The current version of the AIAS is shown in Figure 1.

Despite being a tool primarily geared towards assisting with the redesign and redevelopment of assessment in the age of GenAI, the AIAS can equally be used as a framework for structuring course outcomes, learning activities, or pedagogical tasks. For example, the original AIAS has been adapted to provide guidance on how to develop learners' English for Academic Purposes (EAP) skills by redefining the levels to align with common tasks found in this context (Roe et al., 2024b). Again, this approach demonstrates the flexibility of the AIAS as a tool that can be adapted to different learning environments, yet supports a clear set of principles underpinning the integration of GenAI into educational assessment. These are that it must be built on a clear set of transparent expectations and rules, that it must be adaptable as technology changes and that it must be critically and carefully implemented.

A review of the literature demonstrates that there are potential benefits to GenAI, which can be used in EFL, not just for writing, but in contributing to a number of different skills and supporting many different tasks. To achieve this, it is important to cultivate AI literacy and facilitate informed engagement (Yan et al., 2024) through correctly scaffolded instructions. This relates to broader calls in EFL for teachers and students to be able to recognise the limitations of GenAI and use it safely and with integrity (Kohnke et al., 2023). As a result, we suggest that the AIAS can be adapted for in-class EFL use, not just for structuring assessments for and of learning, but also for broader task-based approaches. While we believe that all levels of the AIAS may be useful in differing instruction activities, we contend that Levels 2, 3 and 4 have special applicability to common tasks in an EFL classroom context. Below, we demonstrate how this may work in practice and the types of EFL tasks that may show alignment with these levels of the AIAS.

Each level selected can be used to give a visual guide and clear explanation to the learner regarding the applicability of AI to the task they are being asked to complete (either inside or outside the classroom). It is important to note that this conceptual approach requires an introduction to the meaning and purpose of the AIAS to students. We suggest that prior to beginning an AIAS framed activity, the framework and rationale are introduced to students. The AIAS should not be described as a set of rules that cannot be broken, nor a method for “catching out” students who, for example, use Level 3 techniques for a Level 2 task. The overarching aim is to guide learners to appropriate use based on mutual understanding and careful task design.

Level 2 of the AIAS represents “AI Planning”. This level is particularly suitable for guiding common EFL tasks, including pre-task activities. For example, if preparing a discussion task based on reading for a group of upper-intermediate learners, then Level 2 may be selected and communicated to the learners. In this context, learners would not be able to use AI “on-the-fly” during the group discussion but may be encouraged to use tools such as ChatGPT, Claude, or NotebookLM to help them understand the reading through clarification of content, support with explaining new vocabulary, or summarising key ideas. GenAI tools can be particularly beneficial and scaffold language learning in these tasks. At the same time, learners will need to be given adequate instruction on the limitations and possibilities of tools that generate inaccurate information. Learners will need to pay close attention to assessing the quality of the output, thus developing evaluative judgement, a key skill for the age of AI (Bearman et al., 2024).

A second example in which Level 2 of the AIAS may be beneficial for EFL instruction is written communication. In the AIAS adaptation for EAP, we argued that GenAI tools may help learners to become familiar with the drafting process and generating their own feedback on text that they have produced (Roe et al., 2024b). In relation to EFL more generally, GenAI may be used in a Level 2 series of activities to provide assistive input on written text production. This is an area where GenAI tools excel; providing quick, user-friendly feedback on how to improve the grammatical accuracy, style, flow and coherence of written text, and explaining the reasons behind these. Educators may also encourage the use of prompting techniques to encourage personalisation of the output – for example, asking the GenAI tool to make text easier to understand for an EFL learner. Depending on the first language of the learner, GenAI tools may even be able to have a dialogue in both languages, thus supporting translanguaging approaches. In translation-focused EFL classes, Level 2 may include AI-assisted pre-translation tasks such as terminology research or clarifying culture-specific references.

AIAS Level 3 refers to “AI Collaboration”. There is a strong alignment between EFL and the AIAS in this context. Although this collaboration between human and GenAI tool is targeted toward the completion of assessment tasks, it can also be extended and adapted to learning tasks in the EFL classroom. Indeed, a strong research base suggests that collaborative approaches are effective for implementing AI within an EFL context. For example, Lee (2024) highlights that using ChatGPT as a writing assistant collaboratively in EFL can help students feel reassured due to the ongoing, instantaneous feedback it can provide, while simultaneously helping to foster patience and involvement with the learning task. Furthermore, Guo and Wang (2024) asserted that a collaborative approach for teachers with ChatGPT can offer effective feedback for student writing. This approach is aligned with empirical studies that have shown that students are more open to AI-supported feedback from teachers than AI feedback alone (Roe et al., 2024a).

For Level 3 focused activities, learners can be encouraged to co-create texts or even speech output (for multimodal GenAI tools) using the target language, phrases, or structures. For higher-level learners, attention can be drawn to challenges sometimes found in GenAI output, such as “wordy” or overly verbose responses, as well as hallucinated or fabricated content (Kohnke et al., 2023). Such collaborative approaches have already been reported in literature, showing that such techniques can improve L2 writing proficiency (Wiboolyasarin et al., 2024) and academic writing proficiency (Nguyen et al., 2023). Collaborative tasks may also include group work or language games; for example, prompting an AI tool to generate quiz questions or resources for other students. Another possible application for the formative assessment of learning may be to ask learners to submit an unedited piece of writing for peer evaluation and then another, subsequently edited with GenAI. Translation training can also be supported at Level 3, for example, by drafting a translation with AI assistance and then post-editing and critiquing the output.

Level 4 of the AIAS (Full AI) encourages learners to make use of any available tools that they see fit when tackling an assessment task. Again, this all-inclusive approach may be beneficial at times in language classrooms, although it requires careful thought and planning to mitigate the known risks of AI technologies. These technologies can be used to support EFL learners in developing confident voices and styles in a new language. Level 4 activities could encourage users to see how effective an application like ChatGPT is for tackling a complex writing or reading test (encouraging the learners to “test” the tool) or could involve asking learners to generate their own self-study resources using GenAI applications. Full AI tasks are useful for encouraging an experimental and creative approach to not only assessment, but also language learning in general. On the other hand, this should be initiated in a way that encourages students to be judicious and cautious in utilising new technologies and guides them toward trusted applications. A further example of a Full AI task may be for a group of learners to produce a poster, presentation slides, or other resources using whichever AI technologies they wish to use (e.g. image, text, or entire presentation slide set generation). Assessments here focus on both linguistic elements of the output as well as the demonstration of more technical skills; however, this approach assumes the requisite level of AI literacy and familiarity with use. A Full AI task should only be explored after students have demonstrated that they have a clear understanding of the limitations, ethical considerations and practical techniques for using GenAI in this setting.

EFL learners often spend more time on low-level writing tasks such as word choice and translation, thus giving less time to spend on higher-level structural tasks such as coherent organisation (Gayed et al., 2022); thus, assigning a Full AI task may encourage learners to cognitively offload such lower-order tasks and spend more time engaging with higher-order concerns. Simultaneously, for more proficient EFL learners, the critical evaluation of GenAI outputs can form part of a Full AI learning task. New 'speech-to-speech' models such as OpenAI's Advanced Voice Mode and Google's Gemini 2.0 Advanced Voice Mode have the capability to not only speak in multiple languages but also to respond idiomatically and with varied dialects. These technologies can be used to support EFL learners in developing confident voices and styles in a new language. For translation contexts, Level 4 tasks might include generating a full translation with AI and then critically evaluating its adequacy, fluency and cultural relevance.

All levels of the AIAS can be adapted to the assessment of learning in the GenAI era, and it is essential to have a balance of tasks – including the use of Level 1 “No AI” assessments where assurance of learning, and task validity are paramount. Simultaneously, using a structured approach to collaborate with students, foster AI literacy and encourage open dialogue on the place that AI and GenAI tools have in EFL education. From this perspective, we argue that Levels 2, 3 and 4 of the AIAS are the most suitable for such activities.

A summary of the different suggestions for implementation in the EFL context is presented in Table 1.

Adapting the AIAS to EFL contexts represents an opportunity to enhance the structure and guidance for both language teachers and students. The AIAS has been implemented successfully in higher education at both the institutional and individual levels and has shown promise in guiding the use of AI technology in assessment and learning. While all levels of the AIAS may be used for assessment throughout a course, learning activities that support language development may also apply the AIAS, providing a clear approach and scaffolding for learners’ knowledge and experience with AI while developing critical digital literacy. Within the AIAS levels, we contend that Levels 2, 3 and 4 show particular promise for this purpose. The AIAS framework’s graduated approach to AI integration may be particularly suited to younger generations, given their greater familiarity with technology in studies and in their lives outside of formal education (Polakova and Ivenz, 2024). Institutions should provide structured PD modules, peer-mentoring, sandbox environments for no-stakes testing of ideas and clear policy frameworks so teachers can build Critical AI Literacy (Roe et al., 2025a, b) and confidence in the use of GenAI tools.

However, there are concerns regarding unequal resource access, which could contribute to a digital divide. In deciding whether to implement an AIAS-type set of activities in the EFL classroom, the ultimate decision must be made contextually and with due consideration of equity, learner needs and the overall approach to teaching; technology should be implemented judiciously and only when it is beneficial to overall learning. In our review of the GenAI and EFL literature, several authors reported the need for a collaborative approach with AI tools; for example, Kartal (2024) and Lee (2024) highlighted the value and need for collaboration. Consequently, Level 3 of the AIAS may be one of the highest-potential forms of AI implementation in an EFL context to scaffold idea generation and language production.

To mitigate inequities, courses can adopt offline AI tools, open-source platforms and pooled institutional subscriptions to ensure broader access. To address access inequities, we recommend: (1) institutional pooling of premium AI tool subscriptions distributed through learning management systems, (2) development of open-source alternatives like locally-hosted language models, (3) creation of offline AI simulation exercises that teach prompting and evaluation skills without live tools and (4) partnering with technology companies for educational licencing agreements. These strategies would reduce broader inequities possible between the global north and global south as GenAI continues to shape global assessment strategies (Perkins and Roe, 2025).

The fact that the framework is flexible, customisable and can be adapted to different tasks suggests that it may be able to effectively meet the needs of EFL learners. Rezai et al. (2024) asserted that AI's capacity to improve vocabulary, grammatical accuracy and pronunciation while providing immediate feedback is relevant. These features are especially relevant to tasks introduced under Level 2 (AI Planning) and Level 3 (AI Collaboration) of the AIAS framework.

However, several important caveats should inform the implementation of the AIAS framework in EFL contexts. Mohamed's (2024) observation that technology cannot replace human interaction serves as a reminder that the framework should not be seen as a replacement for traditional language teaching and learning approaches, nor for pedagogies that have stood the test of time. Particularly in translation, issues of adequacy, fluency and cultural equity should form part of the evaluation of AI outputs, ensuring that learners understand risks such as hallucinated content, cultural loss and bias in bilingual corpora. Ultimately, we must take a careful, cautious and critical approach to experimenting with new technologies in an educational context and issues of equity, inclusion, and data privacy must be considered. Furthermore, this conceptual approach assumes that there is institutional support for such a programme, as well as the resources, time and space available for experimentation. One of the limitations of such an approach is that educational institutions may offer varying degrees of support for AI usage, and in order to adopt an equitable approach, the resources must be shared equally among educators and students. Institutional or program requirements, time requirements and student and educator engagement and perception of AI are also likely to be constraining factors in adopting an innovative, AIAS-driven set of learning activities. Therefore, the AIAS may only be applicable in EFL within a narrow range of contexts.

In this conceptual paper, we argue that a structured, scalar approach can benefit learners and educators by providing a clear and transparent method to ensure two-way communication on how to use new technologies for specific tasks. Our analysis of the EFL literature has demonstrated key areas of alignment between the affordances of AI technology and EFL learning and the structure provided by Levels 2, 3 and 4 of the AIAS. This framework can help develop approaches to include AI in the classroom, without surrendering its use in every situation, and without completely banning such tools, given their multiple affordances for learning. Simultaneously, learners may be able to develop AI literacy skills while benefiting from improved language acquisition in certain AIAS-informed tasks. The framework’s graduated approach to AI integration allows for the scaffolding of language skills and digital literacy.

When making the decision as to whether GenAI tools might be integrated into the EFL classroom, educators and institutions both have responsibilities for their learners. Individual educators have a responsibility in considering the equity and accessibility concerns of the class, the AI technologies to be used and their benefit and the overarching concerns regarding such technologies (for example, their energy use, cultural biases and potential inaccuracies, to name a few). Institutions bear the responsibility of ensuring that if AI is to be included in EFL education, educators are provided with professional development to help develop their understanding and use of new and emerging technologies. We note that there are significant issues in which using GenAI models in the classroom is impractical or inappropriate; this is particularly the case in low-resource contexts, or in contexts that may face cultural or institutional barriers to adoption, and so the AIAS EFL may not be suitable for all.

As we adapt the AIAS and structure its use for a variety of different contexts, including classroom activities and assessment redesign, we view the empirical validation of the framework as a logical and crucial next step in our research agenda. Empirical studies exploring such applications in a variety of cultural and institutional contexts could help uncover insights that would provide directions for future adaptations of the framework and validate its worth as a tool for integrating newly emerging technologies into pedagogy. Future research should include classroom pilots across varied EFL contexts, including under-resourced settings, using simple classroom interventions, short teacher–student surveys and pre-post learning measures to confirm effects on planning quality, post-edit accuracy and evaluative judgement.

As AI technologies continue to expand and become embedded in daily life and educational contexts, simple and effective tools and frameworks that may help guide pedagogical practices will become increasingly important. While the AIAS, as a fundamentally assessment-oriented framework, was originally designed for higher education and K-12 contexts at large, there are clear areas of alignment with non-assessment-focused learning activities in the EFL context. Therefore, the AIAS may be worthy of consideration by educators and institutions as one piece of the puzzle in addressing how to use AI in EFL education.

A preprint of this article appeared on the arXiv preprint server prior to peer review.