2.1. Research theme: Learning processes and outcomes

Research involving GenAI has followed a predictable trend in that, whenever a new technology or digital tool emerges, one of the very first topics that researchers often rush to investigate is stakeholders’ perceptions of it. Of course, developing a thorough understanding of people’s perceptions is vital. After all, if students or their teachers do not like the tool, do not find it useful, or find it challenging to use, then it is unlikely that they will continue using it for educational purposes in the future. That being said, when it comes to new educational technologies, studies of actual learning tend to come second. Thus, the first theme presented here, learning processes and outcomes, involves investigating how GenAI tools can foster aspects of the learning process, as well as facilitate positive L2 learning outcomes.

As discussed, most studies involving GenAI and learning processes and outcomes to date have focused on the extent to which GenAI tools can be useful for WCF purposes. In their synthesis of 51 WCF studies, Crosthwaite and Sun (Reference Crosthwaite and Sun2025) note that many studies have examined how GenAI tools can lead to improvements in writing quality, as well as topics involving the accuracy of GenAI feedback, and how students engage with the feedback they receive. Relatedly, studies have also examined how teachers interact with and use GenAI for providing feedback (e.g., Han & Li, Reference Han and Li2024). Continued research involving GenAI and WCF will be critical, as this research has many practical implications for teachers and learners in terms of formal and informal assessment. However, we also need studies that go beyond examining feedback, as learning to write extends far beyond seeking and receiving WCF.

Writing, particularly in an L2, is a complex process that involves various cognitive and social dimensions (Hyland, Reference Hyland2022; Williams, Reference Williams2012). Additionally, depending on the stages of one’s L2 proficiency, students may be engaging with writing for very different purposes. At lower levels of proficiency, learners are primarily engaging in learning-to-write activities, in which they are learning basic skills such as spelling, manipulating grammatical forms, and constructing sentence- or paragraph-level discourse in highly controlled situations (see Manchón, Reference Manchón2011). At more advanced stages of proficiency, students are often engaging in writing-to-learn tasks. In this sense, the writing tasks that students engage in may resemble real-life tasks, mimic disciplinary genres, or be reflective in nature, in which the process of writing is used as a mechanism for prompting students to engage in critical thinking and to learn or develop new ideas (see Leow, Reference Leow and Manchón2020). Due to the complexities involved in writing, research is needed that investigates a myriad of dimensions, particularly when it comes to learning processes and outcomes.

As Keefe and Copeland (Reference Keefe and Copeland2011) note, literacy is often viewed as a basic skill that many children learn during elementary schooling; in reality though, it is a far more complex. Traditionally speaking, literacy has involved one’s capacity to read and write, yet today, it is also viewed as the capacity to interpret, comprehend, and critically evaluate information from a variety of print and digital resources, as well as the capacity to manipulate and produce various modes for communicative purposes (Cope & Kalantzis, Reference Cope, Kalantzis, Cope and Kalantzis2015; Pangrazio et al., Reference Pangrazio, Godhe and Ledesma2020). With the development of GenAI technologies, GenAI literacy has also now emerged as a new dimension of literacy in multiple respects (Warschauer et al., Reference Warschauer, Tseng, Yim, Webster, Jacob, Du and Tate2023). For one, as more students continue to use GenAI tools, there is a difference between simply using GenAI tools and knowing how to use them ethically and productively (i.e., prompting chatbots in ways that avoid cheating and using them to support one’s work and learning rather than offloading all cognitive work). Additionally, AI literacy is increasingly a desired skill beyond the classroom, as many employers are seeking individuals who can teach them how to use and integrate AI into the workplace (Ryan, Reference Ryan2025).

As mentioned, GenAI literacy is one of the three major topics of inquiry that has been investigated within L2 writing. With such studies, many have focused on teachers. These studies have often investigated aspects of teachers’ knowledge of GenAI tools, their capacity to use them for pedagogical purposes, and more (e.g., Moorhouse, Reference Moorhouse2024; Moorhouse & Kohnke, Reference Moorhouse and Kohnke2024; Wang et al., Reference Wang, Derakhshan and Ghiasvand2025). Due to the novelty of GenAI as a technology as well, some L2 writing scholars have proposed their own literacy frameworks (e.g., Darvin, Reference Darvin2025; Ou et al., Reference Ou, Khuder, Franzetti and Negretti2024; Wang & Wang, Reference Wang and Wang2025). For example, Ou et al. (Reference Ou, Khuder, Franzetti and Negretti2024) have argued that in order to understand critical GenAI literacy, a four-part framework is required, which includes concepts such as interaction with AI, ethics in academic writing, technical value and limitations, and self-learning. It is notable that this framework—like other researchers’ proposed frameworks—all share similar constructs. When aggregated, these frameworks often attempt to tap into key issues of literacy, such as users’ knowledge of what GenAI tools can and cannot do generally, knowledge of the focal tool’s functions, along with how such tools can be used ethically to support one’s learning.

Although literacy-related research continues to emerge, relatively little has been published on L2 writers’ GenAI literacy and how it develops over time in conjunction with other academic literacy skills (e.g., Tan, Reference Tan2025; Lim & Darvin, Reference Lim and Darvin2026). One illustrative example is a study by Tan (Reference Tan2025). In her study, Tan explored how one L2 English graduate student leveraged ChatGPT to support her academic literacies development during the first year of her graduate program. Tan triangulated data such as interviews, survey responses, and logs from ChatGPT. Although the heart of Tan’s article is about academic literacies development, through examining the student’s use of ChatGPT over one year, it also details aspects of the participant’s GenAI literacy over time. In her findings, Tan traces some of the positive aspects of her participant’s literacy development, such as how the participant learned to effectively use the chatbot to support her completion of reading and writing tasks and to learn information about disciplinary conventions in her major. However, Tan also documents some of the negative impacts of chatbots on her academic literacy. For example, the participant voiced concerns that, over time, she felt she was “relying on it [ChatGPT] a little too much”; as a result, the participant provided the self-assessment of “I don’t think my writing skills have really improved that much” over the year (p. 8).

Tan’s study demonstrates the importance of further investigating GenAI literacy, and some of the ways in which GenAI literacy may interact with and support or hinder academic literacies. In particular, we now need studies that investigate questions such as: What do students know and/or believe about GenAI tools when they enter our classrooms? In addition, how do their initial conceptualizations of these tools, along with their practices, change over time? That is, with increased literacy, to what extent are students able to use GenAI as a scaffold or support tool rather than for cognitively offloading critical academic literacy skills? Finally, to what extent can training (e.g., class sessions or workshops) impact different facets of students’ GenAI literacy? Because literacy is a multifaceted construct, researchers might choose to explore a specific aspect of literacy pertaining to one of these questions. For example, with students’ conceptualizations of GenAI tools and how they change over time, researchers might focus on students’ knowledge of what these tools can and cannot do, as well as how students’ perceptions of the ethical uses of GenAI tools evolve over part or all of one semester (see Kessler et al., Reference Kessler, Rostrán Valle, Çekmegeli and Farrell2025 for an example survey on the ethical uses of GenAI tools for L2 writing). In terms of how students’ practices evolve, researchers might look at changes in students’ prompts or prompt-engineering strategies (i.e., how they query a chatbot either verbally or in writing). As both Warzel (Reference Warzel2023) and Pegrum (Reference Pegrum2025) have noted, prompt engineering is an emergent form of thinking; therefore, prompt engineering itself is a form of literacy that needs to be both taught and understood in terms of its development over time.

In terms of executing these studies, Tan (Reference Tan2025) once again provides a useful foundation on which to build, both for researchers who are interested in qualitative or quantitative methods. That is, because literacy itself is a complex skill which may develop slowly, collecting data at multiple time points to track that development is likely necessary. For instance, in qualitative studies, this might involve researchers collecting multiple data sources for triangulation purposes, such as interviews, reflective logs/journals, chatbot logs, writing samples, think alouds, or stimulated recalls. In terms of quantitative studies, researchers might collect one or two of these types of data as well, along with another instrument such as a survey. Surveys or questionnaires, in particular, could be implemented in a pre-/post-test design or even at multiple time points spread across a semester.

Finally, for analyzing and interpreting data, there are various options. As discussed, multiple writing researchers have proposed analytical frameworks for researching and teaching GenAI literacy (e.g., Darvin, Reference Darvin2025; Ou et al., Reference Ou, Khuder, Franzetti and Negretti2024; Wang & Wang, Reference Wang and Wang2025). Some researchers have even published studies in which they have designed and validated questionnaires that attempt to tap into constructs of AI literacy (e.g., Ng et al., Reference Ng, Wu, Leung, Chiu and Chu2024; Wang et al., Reference Wang, Derakhshan and Ghiasvand2025). Researchers are encouraged to explore these options listed here and, ultimately, to determine which methods and frameworks align best with their own interests and also those that meet the needs and profiles of the learners in their contexts.

Apart from literacy, in applied linguistics, there has also been growing recognition of the importance of individual differences (IDs) and emotions. As research has shown, in many cases, IDs (such as anxiety, motivation, willingness to communicate, and more), like emotions (such as joy, frustration, and boredom), have the capacity to either inhibit or facilitate language learning processes and outcomes. Thus, language learners’ and teachers’ IDs and emotions have received a significant amount of attention during the past decade (see Dewaele & Li, Reference Dewaele and Li2018; Li et al., Reference Li, Hiver and Papi2022; Martínez Agudo, Reference Martínez Agudo2018). Due to the critical roles they play in second language acquisition (SLA), we have also begun to see studies emerge that investigate IDs or emotions in relation to L2 writing and GenAI.

When it comes to GenAI and IDs, scholars have recently explored topics such as motivation (e.g., Teng, Reference Teng2025) and metacognition (e.g., Mizumoto, Reference Mizumoto2023; Xu et al., Reference Xu, Qiao, Cheng, Liu and Zhao2025), among others. Teng (Reference Teng2025), for example, conducted a study examining how motivation developed among 261 L2 English learners in China, as students engaged either with peers or GenAI for writing support. Adopting a longitudinal design with a survey administered at three different time points over one year, Teng reported that GenAI use appeared to increase learners’ motivation, particularly early on; however, learners’ motivation then leveled off and plateaued over time. As such, Teng notes the capacity of GenAI tools to enhance student motivation in certain phases of the writing process.

In turning from IDs to emotions, we have also seen publications that examine the emotions that students experience as they interact with chatbots (e.g., Xie et al., Reference Xie, Ou, Jiang, Kessler, Casal and Marino2027; Zhao et al., Reference Zhao, Zhou, Liang and Lian2025; Zou et al., Reference Zou, Reinders and Amjad2025b). For instance, Xie et al. (Reference Xie, Ou, Jiang, Kessler, Casal and Marino2027) conducted a case study with a group of four university-level English learners in China, as they engaged with GenAI while composing a multimodal picture book. In the study, students were required to prompt a chatbot to produce visuals that could be included in their picture books. Using data such as an observation, screen recordings, stimulated recall, and a focus group interview, the authors recount the fluctuations in students’ emotions. Although the focal students occasionally experienced joy when the chatbot produced what they wanted, the students mostly experienced a range of negative emotions such as frustration and, eventually, hopelessness, as a result of continually prompting the chatbot without receiving their desired results.

Crucially, these two studies by Teng (Reference Teng2025) and Xie et al. (Reference Xie, Ou, Jiang, Kessler, Casal and Marino2027) showcase the important roles of IDs and emotions in SLA, as well as the potential interactions between them, including how negative emotions have the capacity to demotivate students when engaging with activities. As a result, we need additional research that explores IDs and emotions independently, as well as research that probes the potential interactions between them. For instance, we need studies that investigate questions such as: When engaging with GenAI to complete a task, what emotions (e.g., boredom, frustration, joy) do students report experiencing (and why)? How does the use of GenAI impact the ID of X (such as anxiety, identity, metacognition, motivation, or willingness to communicate)? Finally, when engaging with GenAI, to what extent does X emotion impact Y ID (or vice versa)?

Methodologically, there are multiple ways to approach these research questions. For qualitative studies, case studies are a logical choice, either for investigating the emotions of individuals or pairs/groups of students as they collaborate with GenAI tools for a given task (e.g., during a singular class activity or an activity that extends over a brief period, as exemplified by Xie et al., Reference Xie, Ou, Jiang, Kessler, Casal and Marino2027). Additionally, emotions—like IDs—have been shown to fluctuate and evolve over time. Therefore, studies are needed in which researchers adopt more longitudinal, ethnographic approaches in the classroom. This could be achieved by tracing how individuals or pairs/groups experience specific IDs and/or emotions over a semester and the extent to which this impacts students’ engagement with activities. For accomplishing this, data could be collected in the form of observations (by the researcher/teacher), field notes, audio recordings of student interactions, and semi-structured interviews with students.

In terms of quantitative designs, a survey might be employed as a means of tracking changes in an ID or emotion. Once again, the time period could be brief (e.g., before and after a singular activity), or it could extend over a lengthier period. This might involve giving a pre-/post-survey or administering a survey on three or more occasions (as demonstrated in Teng, Reference Teng2025). With quantitative studies involving IDs, in particular, there are questionnaires that have been proposed for evaluating aspects of IDs such as anxiety (e.g., Horwitz et al., Reference Horwitz, Horwitz and Cope1986; also see Lee & Ye, Reference Lee and Ye2023), metacognition (e.g., Sun et al., Reference Sun, Zhang and Carter2021; Zhang & Qin, Reference Zhang, Qin, Haukås, Bjørke and Dypedahl2018), and motivation (e.g., Dörnyei & Taguchi, Reference Dörnyei and Taguchi2010; also see Papi, Reference Papi2010). Thus, researchers might adapt and employ these questionnaires to suit their needs. With quantitative studies as well, apart from describing how GenAI use impacts learners’ processes, it will be particularly insightful if researchers can link IDs or emotions to specific learning outcomes. For instance, some of the aforementioned studies attempt to link IDs such as metacognition with writing performance (e.g., Sun et al., Reference Sun, Zhang and Carter2021). In this regard, it would be useful for researchers to combine questionnaires with elicited examples of students’ written output and performance (e.g., via a timed writing task). For analyzing these data, depending on the instruments and types of variables involved, researchers might consider conducting regressions, structural equation modeling, or other statistical analyses to examine the extent to which IDs or emotions influence or predict certain learning outcomes, including any interactions among variables.

Of course, these are but a few examples of studies that might be conducted; however, the results of such studies could shed important light on GenAI’s role in mediating IDs, emotions, and L2 learning processes and outcomes.

2.2. Research theme: Student use and interactions

While the first theme covered various studies that delved into learning processes and outcomes, the second theme addresses student use and interactions. Specifically, this theme pertains to various aspects of how learners engage with GenAI tools independently and with each other. As discussed, most of the research to date exploring how students use or engage with chatbots has focused on WCF, including how students react to the feedback that they receive and the extent to which they find that feedback to be useful. However, there is now a body of literature emerging in which scholars have investigated student-initiated uses of GenAI tools (e.g., Hoomanfard & Shamsi, Reference Hoomanfard and Shamsi2025; Hwang et al., Reference Hwang, Chang and Sun2025; Kessler et al., Reference Kessler, Rostrán Valle, Çekmegeli and Farrell2025; Kohnke et al., Reference Kohnke, Zou and Su2025). For instance, in Kessler et al. (Reference Kessler, Rostrán Valle, Çekmegeli and Farrell2025), we adopted mixed methods to investigate how 287 university students enrolled in 5 different foreign languages reported using GenAI tools for L2 writing purposes. Surprisingly, in sharp contrast to various reports about widespread student use (e.g., College Board, 2025), only 14.6% of the students in our sample reported being users of GenAI tools; many also believed GenAI use to be unethical in higher education settings, unless used for specific purposes. Still, we identified 12 distinct ways in which those 14.6% of student users reported leveraging GenAI for L2 writing purposes. Of particular note is that only 1 of the 12 uses was for WCF. For instance, students reported (and subsequently demonstrated via individual case studies) using GenAI for learning how to use words appropriately in context, generating ideas, learning collocations, creating outlines, and more.

As this study and others have demonstrated, there are a myriad of ways in which students are currently engaging with GenAI tools to support their L2 writing efforts both inside and outside of the classroom. Because of this, we now need studies that explore various aspects of student use and interactions. As will be discussed, we need an array of studies that investigate how learners interact with GenAI tools individually and collaboratively, along with how students process and react to chatbot-related outputs during those interactions.

Over the past two decades, collaborative writing (CW) has been a popular topic of inquiry within L2 writing. Not only do multiple theories of SLA and human development support its use (e.g., Interactionist Approaches, Sociocultural Theory), but there is now a deep well of literature that describes its various benefits to learners (e.g., Li & Zhang, Reference Li and Zhang2023b; Storch, Reference Storch2019; Zhang et al., Reference Zhang, Gibbons and Li2021; Zhang & Plonsky, Reference Zhang and Plonsky2020). Research also suggests that many L2 writing teachers now recognize these benefits and regularly adopt CW activities in their classrooms (e.g., Kessler & Casal, Reference Kessler and Casal2024). Of course, despite its widespread adoption, the CW literature is also filled with complexities—that is, CW is not simply good for L2 learning whenever and however it is implemented. For instance, both students’ proficiencies and group size (e.g., dyads vs. groups) must be considered prior to pairing students; additionally, the types of activities and digital tools given to learners can shape the interactions they engage in. Thus, even though CW has been researched extensively, there is still a considerable amount of research that needs to be done both on novel topics (see Li & Zhang, Reference Li and Zhang2023a) and in terms of replication (see Kessler, Reference Kessler2025).

In addition, as GenAI tools continue to make their way into classrooms, we need studies that examine students’ individual and collaborative engagement with GenAI for L2 writing purposes. This is because many students are now engaging in writing activities with GenAI as their sole writing partner or “peer.” Some researchers have also suggested that in instances where it may be challenging to find human interlocutors, GenAI may be a promising alternative for students (e.g., Cheng et al., Reference Cheng, Li and Lee2025; Su et al., Reference Su, Lin and Lai2023). However, many teachers are also asking their students to work collaboratively in class with both a human peer and a GenAI tool (e.g., in a pair or group activity).

For example, in Michel et al. (Reference Michel, Bazhutkina, Abel and Strobl2025), the authors argue that teachers must find creative ways to integrate GenAI into their CW assignments since GenAI is a powerful technology that has the capacity to positively influence learning outcomes. In their study, Michel et al. explored how four dyads in an L2 German class engaged with GenAI during a CW activity, which required students to compare their own writing with GenAI-produced models; the students used a rubric with statements to help guide them through the comparison and evaluation process. Analyzing data such as audio recordings of dyads’ interactions and screen recordings of students’ writing and revision behaviors, the researchers provide various insights into how well the dyads collaborated, what the students discussed during their interactions, and ultimately, how students reworked their papers to try to improve the overall text quality. Michel et al.’s study is but one of multiple early attempts to explore the intersections of CW and GenAI, but it is important because it suggests the potential of GenAI to be used to foster language-related episodes (LREs) (i.e., discussions about language use). It also highlights some of the complexities of CW scholarship, including the need for more research. For instance, while some dyads engaged in numerous LREs, one dyad demonstrated a lack of interaction with the chatbot, and another higher-proficiency dyad focused more on the task itself rather than language-related matters. This study highlights the need not only for mixed methods but also for future studies that compare and contrast student and teacher perception data.

As such, more research is needed that explores several topics that have previously been the focus of computer-mediated CW studies, only this time with GenAI as the technology in focus. For instance, some key questions include: Do learners prefer engaging in GenAI-supported writing activities individually or collaboratively? (Do their preferences change based on: the writing task/the students’ proficiency levels/or on other factors?) Additionally, how does individual vs. collaborative engagement with GenAI impact students’ written output in terms of (lexical and syntactic) complexity, accuracy, fluency, and text quality? When engaging with GenAI-supported CW activities, how do factors such as proficiency level, group size (e.g., dyads, groups of three), or other pairing considerations influence the nature of students’ interactions (e.g., the number of LREs produced and resolved)? And finally, how do higher proficiency students in graduate-level courses individually and/or collaboratively engage with GenAI tools for research writing purposes or composing other academic genres?

In terms of executing such studies, fortunately, there is an extensive body of CW literature. Therefore, there are plenty of options for all researchers, regardless of whether they are interested in quantitative, qualitative, or mixed methods. Due to the number of available options, instead of recommending specific study designs here, I encourage readers to review two key resources. The first resource is Storch’s (Reference Storch2019) research timeline on CW, which appears in this journal. Storch’s timeline is an informative account of early, influential CW studies, which includes brief descriptions of the studies’ designs and findings. The second resource is Zhang et al. (Reference Zhang, Gibbons and Li2021), which is a systematic review of 113 computer-mediated CW studies. This synthesis covers key features of studies’ designs, including the contexts, methods, analytical frameworks, and data analyses used in prior studies. Readers are encouraged to review these two resources—particularly Zhang et al.’s synthesis on computer-mediated CW—as many of the studies that are discussed can be modified to include GenAI as the focal technology.

As some of the prior tasks have suggested, with GenAI’s emergence, numerous benefits have been documented; however, there also have been major concerns. Plagiarism and cheating are often the primary issues discussed across academia and by the public more broadly, yet another concern—particularly from linguists and language educators—has been how GenAI tools might contain inherent biases toward speakers of different Global Englishes and toward those who are learning non-English languages and their varieties (e.g., Jeon et al., Reference Jeon, Lee and Coronel-molina2024; Kang & Hirschi, Reference Kang and Hirschi2025; Lee et al., Reference Lee, Jeon, McKinley and Rose2025). This concern stems from the types of data that many large-language models are presently built upon, which (at the time of writing this article) often lack significant training data from non-English languages and which also reflect Standard American English (SAE) preferences; thus, as scholars have noted, many of the chatbots that are available have the potential to exhibit the same problematic biases (Bommasani et al., Reference Bommasani, Hudson, Adeli, Altman, Arora, von Arx, Demszky, Bernstein, Bohg, Bosselut, Brunskill and Brynjolfsson2022), and there tends to be tremendous variability in performance across languages (Swinehart et al., Reference Swinehart, Nguyen and Yeh2025). As Kang and Hirschi (Reference Kang and Hirschi2025) further remark, “many current AI models seem to replicate, and at times, even amplify or distort social biases found in human language in unexpected ways… AI systems do not process speech with diverse characteristics equitably” (p. 70). Obviously, there are many languages spoken around the world, and there are also many diverse varieties of both English and other languages. Therefore, for those who are learning an L2 that is not English, or for those who do not speak a dominant variety, the interactions they have with chatbots have the potential to be problematic. For instance, issues may arise when learners are seeking WCF. That is, if the training data are biased, then learners may be corrected unnecessarily.

One example of this is showcased by Lee et al. (Reference Lee, Jeon, McKinley and Rose2025). In their case study, the authors explored the use of different prompts within ChatGPT, explicitly asking the chatbot to consider Global Englishes principles based on the background of the study’s focal student, who was a speaker/writer of Korean English. The researchers’ goal was to explore whether the chatbot could integrate a Global Englishes perspective by valuing principles like linguistic diversity and communicative effectiveness, or whether the chatbot would default to providing feedback based on SAE and native-speaker norms. In their findings, Lee et al. reported that ChatGPT often resorted to native-speaker norms, focusing on standard grammar and vocabulary. Despite explicitly being prompted to do so, the chatbot failed to consider the student’s background and English variety. The authors provide specific examples of grammatical and lexical items that ChatGPT incorrectly marked as being errors and subsequently replaced with standardized forms.

The study by Lee et al. (Reference Lee, Jeon, McKinley and Rose2025) is a powerful example of the dangers of GenAI to negatively impact speakers of English around the world. It also further illustrates how limited training data (in a specific variety of English or in non-English languages) not only have the capacity to provide users with incorrect responses but also, as Lee and colleagues go on to state, to “eras[e] multilingual and multicultural aspects of the writer …” (p. 94). Due to these potential dangers, we now need future studies that explore aspects of GenAI student use and interactions, involving either a) how GenAI supports or hinders L2 learners of languages other than English (and their varieties), or b) how GenAI supports or hinders L2 writers of various Global Englishes.

These lines of inquiry are necessary since, as discussed, much of what we currently know about GenAI and L2 writing involves English as the target language, primarily from studies involving American or British English. In general, we need more studies that involve languages other than English. These studies could be on any topic pertaining to how L2 writers use or interact with GenAI chatbots, including WCF and how students process and respond to that feedback. Studies might employ original designs, or they might also be replications of existing studies but with English replaced by a different target language (e.g., To what extent do the findings of X-study hold true if English is replaced by X-language?).

For studies involving Global Englishes or varieties of other languages, we need research that explores what happens when learners of different varieties interact with chatbots. For example, questions might be posed such as: What kinds of feedback do students of X-variety receive when seeking feedback on their writing (in terms of grammar, content, style, etc.)? How do they perceive that feedback? For instance, do they perceive the feedback as being (in)correct? And to what extent do these interactions impact aspects of their emotions, identities, motivation, and/or willingness to engage with chatbots? Finally, do learners have their own strategies for prompting or engaging with chatbots when learning a specific variety (and if so, what are they)?

In terms of executing these studies, Lee et al. (Reference Lee, Jeon, McKinley and Rose2025) offer a clear template of how researchers might attempt to prompt a chatbot to consider specific criteria (e.g., linguistic diversity and communicative effectiveness) when responding to learners’ or researchers’ inquiries. However, it is also important to note that it is highly unlikely that students will prompt chatbots with such sophisticated, research-backed prompts when using chatbots on their own. As such, in addition to studies that explore the extent to which prompt engineering might be used to make chatbots sensitive to specific varieties, we also need more naturalistic studies involving students and how they use GenAI. For instance, quantitative studies that are conducted inside a classroom might leverage data such as chatlogs to record the types of prompts students give and the types of responses and feedback they receive, in addition to using surveys to investigate how those interactions are perceived and/or influence different factors (e.g., aspects of awareness, motivation). In terms of qualitative methods, researchers might conduct similar studies inside the classroom but by leveraging data such as observations, chatlogs, and semi-structured interviews; conversely, such studies could also be conducted outside of the classroom via screen-recordings, chatlogs, semi-structured interviews, and self-report data such as reflective journals, as a means of exploring how students engage with chatbots on their own beyond school.

Once again, there are a limited number of studies in this area. Due to the paucity of research, any contributions that attempt to speak to the intersections of GenAI, Global Englishes (or non-English languages), and L2 writing are welcomed.

2.3. Research theme: Teaching

While the first two themes addressed tasks pertaining to learning processes and outcomes and student use and interactions, the final theme addresses teaching. As the name implies, this theme is related to how educators might strategically leverage GenAI tools for specific purposes. Across applied linguistics, we have seen numerous studies in which researchers have advocated for and described how GenAI can be used by teachers. For instance, researchers have discussed how GenAI tools can be used for creating reading materials (e.g., Lin et al., Reference Lin, Jiang and Lei2025), for developing accessible and personalized speech scoring systems (e.g., Shin et al., Reference Shin, Lee and Kim2025), and even for evaluating chatbots’ general capabilities for supporting teachers of languages other than English (e.g., Swinehart et al., Reference Swinehart, Nguyen and Yeh2025). However, as readers will note, none of these aforementioned areas are chiefly concerned with L2 writing instruction. That is, apart from leveraging GenAI for providing WCF, far fewer researchers have examined the applications and utility of GenAI for other areas of L2 writing instruction.

Because of this, we need studies that explore various aspects of teaching with GenAI in the classroom. As will be discussed in the tasks that follow, we need studies that investigate how teachers might leverage GenAI tools to support two key areas that are currently prominent within L2 writing pedagogy. These two areas involve genre-based instruction (GBI) and digital multimodal composing (DMC).

GBI is a popular approach to L2 writing, which, as Hyland (Reference Hyland2007) explains, enables teachers to ground their pedagogies in authentic texts that students will encounter in their everyday lives. Although there are multiple approaches or schools of thought when it comes to executing GBI, regardless of the approach, Tardy et al. (Reference Tardy, Caplan and Johns2023) remark that one of the primary goals is to move instructors away from teaching prescriptive, templated forms of writing (e.g., five-paragraph essays), and instead, to turn writing instruction into something that is “dynamic, situated, goal-oriented, and responsive to readers and communities” (p. 1).

Given the aims of GBI, as a pedagogy, it is often used with L2 learners of higher proficiencies (e.g., in English for Academic/Specific Purposes [EAP/ESP] courses), when students have the linguistic capabilities to engage with academic or professional genres. Notably, when it comes to L2 academic writing and GenAI, research has shown that instructors have been busy experimenting with GenAI in EAP classrooms for teaching important writing skills such as paraphrasing (e.g., Xu & Zheng, Reference Xu and Zheng2025). Additionally, many L2 graduate students already appear to possess complex strategies for using GenAI to support their academic research and writing efforts (e.g., Hoomanfard & Shamsi, Reference Hoomanfard and Shamsi2025; Zou et al., Reference Zou, Kong and Lee2025a). Despite the prominence of GenAI tools in many graduate students’ lives and academia more broadly, there have been few attempts to explore the applications of GenAI for GBI and for supporting the teaching and learning of different genres in the L2 writing classroom.

One notable study in this regard is Kim and Lu (Reference Kim and Lu2024). In their study, the authors examined the potential of using ChatGPT for automating rhetorical move-step analysis—a common type of linguistic analysis that is often taught to L2 writers in EAP and ESP courses. By experimenting with various prompts, Kim and Lu analyzed ChatGPT’s accuracy in annotating research article introductions from various social science disciplines (e.g., applied linguistics, economics, political science, psychology). Through experimentation and prompt refinement, the researchers demonstrated that, in some cases, ChatGPT was able to tag the rhetorical moves with approximately 90% accuracy. As Kim and Lu subsequently note, these findings have implications not only for researchers but also for instructors and GBI. They argue that GenAI tools might be used by L2 writing teachers for genre analysis activities as a means of helping novice writers tag and analyze unfamiliar texts. Thus, future studies are needed that investigate the potential applications of GenAI tools for facilitating GBI and teaching different academic and professional genres.

While Kim and Lu’s (Reference Kim and Lu2024) study is an excellent demonstration of the capabilities of GenAI tools for move-step analysis, future studies are needed that are situated in classroom contexts with teachers and students. Research questions might be posed, such as: To what extent do teachers find GenAI useful for facilitating GBI? To what extent can GenAI be used to facilitate rhetorical move analysis activities for X-genre, without completely offloading all of students’ cognitive work (i.e., supporting learning, rather than replacing it)? By extension, do students find GenAI to be accessible and useful for such purposes? And more broadly, to what extent can GenAI be used to help foster genre knowledge and awareness (e.g., of audience, moves-steps, and/or common lexico-grammatical features)?

For executing these studies, since classroom-based studies are needed, quantitative, qualitative, or mixed methods could be adopted depending on the size of one’s class and the number of sections that are available for study. For the aforementioned questions, researchers interested in quantitative or mixed methods might adopt a quasi-experimental design with one or two classrooms. Since studying and learning a new genre typically takes more than one class period (with time varying widely from genre-to-genre), multiple GenAI genre-analysis activities might be integrated into a syllabus over the span of one or multiple weeks. Studies with two sections of a class could have an experimental group who adopts GenAI for certain activities, while the comparison group engages in traditional manual analyses. Conversely, if only one teacher and section are available for study, researchers might explore a standalone activity with GenAI and evaluate the teacher’s and students’ experiences.

In terms of data collection and analysis, quantitative or mixed methods studies might analyze samples of students’ writing before and after a treatment. Alternatively, such studies could analyze pre-/post-surveys before and after the activities to evaluate aspects such as changes in students’ genre knowledge or students’ perceptions of the utility and usability of GenAI for learning about the focal genre. In terms of qualitative studies, multimodal visualization reflections might be a useful source of data, particularly for understanding aspects of students’ genre awareness (e.g., Kessler, Reference Kessler2024b; Negretti & McGrath, Reference Negretti and McGrath2018). Finally, since teachers are key to this particular research task, data such as semi-structured interviews could be useful for gaining insights.

For analyzing these data, as mentioned, there are multiple approaches to GBI (e.g., ESP, Systemic Functional Linguistics). Although they are not mutually exclusive, those who research genre often operationalize and analyze constructs differently based on the approach being taken. Because of this, data analyses and frameworks will vary widely. Interested readers are encouraged to see resources such as Hyland and Shaw (Reference Hyland and Shaw2016) and Kessler and Polio (Reference Kessler and Polio2024), which cover conducting genre-based research in applied linguistics from multiple perspectives.

Apart from GBI, another popular topic during the past decade has been DMC (e.g., Jiang & Hafner, Reference Jiang and Hafner2024; Kessler, Reference Kessler2024a; Li, Reference Li2022; Lim & Kessler, Reference Lim and Kessler2024; Zhang et al., Reference Zhang, Akoto and Li2023). DMC involves creating meaning by leveraging digital tools and manipulating linguistic text in addition to one or more non-linguistic mode (e.g., aural, gestural, visual, or spatial). As research has demonstrated, the use of digital, multimodal genres and activities is now ubiquitous throughout many higher education contexts (Lim & Polio, Reference Lim and Polio2020). As a result, multimodal literacy is considered a key competency for achieving academic and professional success (Grapin & Llosa, Reference Grapin and Llosa2020; New London Group, 1996).

While prior studies have explored a plethora of digital tools for completing multimodal tasks, in recent years, we have seen researchers begin to explore how GenAI can be integrated into the DMC composing process (e.g., Jiang & Lai, Reference Jiang and Lai2025; Kang & Yi, Reference Kang and Yi2023; Tan et al., Reference Tan, Xu and Wang2025; Xie et al., Reference Xie, Ou, Jiang, Kessler, Casal and Marino2027). For example, in Jiang and Lai (Reference Jiang and Lai2025), the authors conducted a mixed-methods study examining how GenAI use differentially affected the quality of students’ DMC products. In the study, 110 L2 English learners in Hong Kong (ages 11–14) worked in groups of 3–5 over several weeks to collaboratively create a digital video documentary. These 110 students were divided into 2 main groups: those permitted to use a GenAI tool called Poe (n = 14 groups) and those who were not (n = 10). In addition to analyzing students’ video documentaries, the researchers collected screen recordings of the students’ composing processes, conducted observations of in-class activities, and conducted pre-/post-study focus group interviews with students. Jiang and Lai reported that the GenAI-assisted groups outperformed the non-GenAI groups in multiple areas, including total score on the grading rubric, with a large effect size. In terms of qualitative data, analyses of the focus group interviews suggested that the GenAI tool served as an important scaffold for students when they were struggling with generating ideas and structuring their videos, particularly when compared to those students who did not use GenAI.

The findings of Jiang and Lai’s (Reference Jiang and Lai2025) study are noteworthy in that they speak to one of the main challenges that students and teachers often voice when it comes to engaging in DMC. Specifically, this challenge is that high levels of digital literacy may be required to engage in certain types of DMC tasks (such as composing digital videos), which, in turn, can create various technical/usability issues that teachers need to spend time resolving. Additionally, from a student’s perspective, it can take considerable time to find certain non-linguistic modes (e.g., audio, images) to integrate into their compositions. Thus, the use of GenAI in this study and others speaks to GenAI’s capacity to facilitate the teaching and execution of DMC tasks. However, since this line of research is still in its infancy, additional studies are needed.

In particular, research questions might be posed such as: To what extent do teachers find GenAI useful for generating materials to be used in DMC projects? To what extent can GenAI facilitate aspects of students’ composing processes when engaging in DMC (once again, as a means of supporting learning rather than completely offloading all of students’ cognitive work)? Subsequently, how does GenAI use impact aspects of students’ DMC products? And finally, to what extent is GenAI useful for completing X-type of DMC task when compared to Y-type of DMC task? (That is, is GenAI perceived as being more useful for composing some multimodal genres over others)?

Methodologically speaking, classroom-based studies are needed. Additionally, quantitative, qualitative, or mixed methods could be adopted, depending on the size of one’s class and the number of sections that are available. For multi-section studies, Jiang and Lai’s (Reference Jiang and Lai2025) study provides a template for how students in different classes could be divided into GenAI and non-GenAI groups to examine the impact of the technology on specific product-related outcomes (e.g., composition quality, complexity, accuracy, and fluency). Apart from quantitative and mixed methods studies, qualitative studies are also needed. Any of the research questions posed above, for instance, could be investigated by adopting a combination of qualitative data such as observations, screen recordings (of students’ composing processes), stimulated recalls (of the composing process), and/or interviews. Once again, data analyses will vary widely based on the questions being posed and the methods involved; however, with qualitative studies, by triangulating multiple data sources, insights could be provided into aspects of teachers’ and students’ perceptions, along with their actions (e.g., composing processes and decision-making when it comes to leveraging and remixing different modes).