Enhancing EAP students’ academic writing and AI literacies: practice at a transnational university in China

3.1 Prompt writing for EAP AW

A key aim of EAP AW courses is for students to gain an awareness of different genres and discourse patterns. Students often enter AW courses with a “one size fits all” approach to writing tasks which does not take genre conventions or task appropriateness into consideration. Students also tend to be unfamiliar with process writing as many high school writing assessments are short, timed writing tasks conducted in exam conditions. Students may therefore attempt to apply familiar strategies to the range of academic writing tasks they are expected to complete at university. In the same way that writing an essay without planning, developing or generating ideas is unlikely to lead to the production of a task-appropriate response, copying an AW writing prompt into a GenAI tool without careful prompting will likely lead to an output which is unsatisfactory (Ngo and Hastie 2025). Effective prompting, as with academic writing, requires students to consider the audience, purpose and style of a desired text (Rowland 2023). Given the commonalities between AW and AI prompting, we aimed to utilise activities using GenAI tools to explore these competencies with students. These activities were chosen to align with the “know and understand AI” and “use and apply AI” aspects in the framework by Ng et al. (2021).

3.2 Enacting EAP AW assessment and feedback literacies

A further challenge we had experienced was in helping students understand learning outcomes and rubrics, or enhancing their assessment literacies more broadly (Rust et al. 2005). Due to a lack of familiarity with EAP rubrics, our students initially produced writing in the style of a high school essay or an IELTS writing task. After exposure to and engagement with AW rubrics and learning outcomes, students produced more task-appropriate texts. However, before properly engaging with marking rubrics and learning outcomes, there is a danger that students may assume a GenAI tool can automatically produce a task-specific response of appropriate quality. We aimed to encourage dialogue between students, artefacts, and GenAI to enhance students’ assessment literacies. These dialogues around standards and rubrics were chosen to align with the “know and understand AI” and “use and apply AI” aspects in the framework by Ng et al. (2021).

3.3 Critical thinking for EAP AW

Our EAP AW module has significant expectations for the level of criticality and autonomy students are required to demonstrate in a UK style HE course. Students who are used to a more teacher-led style of teaching and feedback can find adapting to a more student-led style of education challenging initially. There is the danger that students could overly rely on GenAI to mitigate some of these challenges, but we also recognised that GenAI could potentially help with reducing some of these challenges, if its use is scaffolded carefully. We aimed to promote student agency by supporting their use of GenAI and providing carefully structured support involving teacher modelling and comparison tasks. Previous studies within EAP settings have suggested that students enjoy critically analysing GenAI output and comparing these to their own compositions (Kostka and Toncelli 2023). We took this as a starting point for building students’ evaluative judgement around both their own AW texts (Tai et al. 2018) and GenAI output (Bearman et al. 2023). These activities aimed to support the third aspect of the framework by Ng et al. (2021): evaluate and create AI.

3.4 Exploring academic integrity within EAP AW

Our EAP AW course is usually students’ first exposure to the concept of academic integrity. In particular, the foundation year is the first time when students have been asked to perform source-based writing tasks in which they use and synthesise texts by different authors. EAP courses are the place where many students are exposed to ideas around academic integrity and are particularly important for establishing habits which students draw on throughout their academic lives (Perkins et al. 2020). Given the ease with which GenAI tools can produce large amounts of coherent and seemingly authentic text, concerns have been raised about the issue of “AI-giarism”. This is where students use GenAI to write entire essays and present these as their own work (Chan 2024). Plagiarism is a difficult concept to define and is influenced by cultural and moral perspectives around authorship and ownership (Howard 1995). Determining how much a GenAI tool has been used in a student submission and how much usage constitutes plagiarism is a challenging task, and institutional policies and approaches will differ (Luo 2024). The reasons why students perform any kind of plagiarism are complex, and intention to deceive is not always the primary motivation (Hu and Lei 2012). As such, we wanted to avoid taking a punitive approach to GenAI and academic integrity and aimed to have an open and inclusive dialogue with students involving the discussion of case studies and our own modelling of appropriate uses of GenAI. This focus on GenAI and academic integrity aligned with the fourth level of the AI literacies model by Ng et al. (2021): AI ethics.

4.1 Understanding GenAI: opportunities and pitfalls for academic writing

The first practice was to guide students to know and understand AI in the AW context. Before starting university, students may have some experience using different GenAI platforms, including ChatGPT, for daily and academic purposes. However, few have used it for writing an academic essay or may not have used GenAI ethically or effectively (Črček and Patekar 2023).

The students were first briefly introduced to the GenAI technology, including its history, how it works and common applications. After that, the XIPU AI platform, a GPT-based GenAI tool created by the institution, was introduced to the students, including the access, interface and basic functions. Meanwhile, students were asked to log in to the platform and interact with it so as to gain some hands-on experience, especially those who had not previously used any GenAI platforms. The EAP lecturer offered basic tech support during the process and invited the students to share what they thought of the platform when it came to problem-solving. A guide made by the IT support team was shared with the students to help them further explore the tool, such as its advanced functions and usage limitations.

After that, it was important for the students to understand what opportunities GenAI could offer for helping with their EAP writing. First and foremost, students should be aware of the purpose of using GenAI for academic writing. Specifically, they should be clear why they were trained to use GenAI for their writing and how GenAI could help them apart from what they could learn from the course and the writing assistance they could obtain from some non-GenAI tools such as Grammarly and Marking Mate. This is because many studies found that motivation or willingness is a significant influencing factor of users’ adoption intention of GenAI for educational purposes (Chan and Hu 2023). To help students explore the potential benefits, we provided a few possible areas which they could try asking XIPU AI to help with. These areas included brainstorming ideas on a writing topic, structuring a paragraph or essay, receiving feedback and improving language (Pack and Maloney 2024). In the controlled interaction practice, the students were provided with some sample texts and prompts in order for XIPU AI to generate more desirable outputs, and then they were asked to compare their own thoughts to the outputs of XIPU AI and those of Grammarly or other similar tools to discuss the advantages of using GenAI to help with the writing process.

Additionally, the students were also asked to reflect on the downsides of using the GenAI tool for academic writing compared to producing the writing completely by themselves or with the help of classmates or teachers. The possible downsides may include inaccurate information (off-topic responses to the writing task), fake or “hallucinated” references, and language under or over the students’ targeted level. More specifically, one way for them to notice the restrictions or challenges of using GenAI for academic writing was to compare the GenAI outputs with what was taught in the EAP course. For example, the students were tasked with checking whether the structure of a paragraph generated by XIPU AI contained all the structural components taught in the class and whether the paragraph was cohesively organised by using cohesive devices they learned from the course. At this stage, with the relatively simple sample prompts, students might find that the text produced by GenAI did not meet the requirements of the course, and thus they should be the “gatekeeper” and evaluate carefully prior to the uptake of any GenAI-created content.

At the end of the semester, after the students used and learned more about GenAI, they reported that they perceived GenAI useful and intelligent for academic writing to a bigger extent than before. Meanwhile, they became more cautious about whether its outputs were aligned with the course expectations.

4.2 Prompt engineering for process-based writing

The students were also trained how to craft effective prompts for different stages of the writing process to save time and generate more desirable outputs (Ngo and Hastie 2025; Walter 2024). The writing process was divided into brainstorming on the essay topic, reading and analysing sources, making an outline, writing the draft, receiving feedback, revising the draft (Flower and Hayes 1981). This was aligned with the process-based writing instruction in the existing EAP writing curriculum, so the integration of AI literacy training on how to engineer prompts to help students in the above various steps was seamless.

Before the students were taught how to craft prompts specific to receiving assistance in the academic writing process, they were given some general tips on prompting GenAI. We encouraged the students to treat GenAI as a naïve kid, and thus they should be clear and specific when giving a prompt, because vague prompts may lead to vague answers. Some general tips we provided are as follows:

1. Provide XIPU AI with a specific context or background of what you are working on and what you need to help with.

2. Narrow down requests and questions to focus XIPU AI on the task at hand. Avoid vague prompts such as “improve this paragraph for me” and “identify problems of my essay”. Provide some example outputs if needed.

3. You may ask follow-up questions since XIPU AI sometimes needs clarification. It may be necessary to have a back-and-forth dialogue with XIPU AI.

4. Limit the word count of your prompts to 150–250 words since lengthy prompts could confuse XIPU AI. Similarly, you may need to ask XIPU AI to review one paragraph of your essay at a time.

5. You may need to adjust and refine your prompt (e.g., rephrasing it) based on XIPU AI’s response if your initial prompt does not work.

After the students were familiar with the general protocol of prompting and interacting with GenAI, the writing-process-based prompt engineering was implemented, which foregrounds student-centred learning, open and transparent communication practices, and students’ “hands-on” experience using relevant AI tools (Ngo and Hastie 2025). For the different stages of the writing process, the students were asked to brainstorm what GenAI could help them with specifically, and then the lecturer elicited feedback from the students and provided some suggested areas. After that, the students tried creating their own prompts to receive outputs from XIPU AI for assistance in the suggested and other areas. For example, in the stage of reading and analysing sources, students were tasked with practising using XIPU AI to help with their source comprehension and analysis and to identify the main advantages and disadvantages in relation to the essay topic illustrated in one of the provided source articles. Sample prompts were not provided at first so that students could independently craft prompts based on the general tips of prompt engineering, review the results and refine their prompts. Subsequently, the students shared their dialogue with GenAI among themselves and agreed on the most effective prompt(s), and through peer learning, they learned different ways of prompting XIPU AI for the same purpose which could lead to a variety of responses. After that, the lecturer demonstrated with a relatively holistic and effective prompt which yielded a satisfactory output. Then, the students were asked to improve their prompts, if needed, by following the key features of the sample prompt to generate new outputs. By observing and modelling prompts for process-based writing, students were more likely to identify the features of effective interactions with GenAI or prompt engineering at different points of the writing process. After the semester-long training, the learners thought the provided prompt banks were useful but still hoped to enhance their prompt engineering skills for more advanced and spontaneous interaction with GenAI in order to receive desirable outputs.

4.3 Enhancing assessment literacies

The third aspect of our practice was elevating learners’ AW assessment literacies. In order for students to use GenAI more effectively for their academic writing, it is important that they are clear about what the final writing product should look like, which means they need to be familiar with the task specifications and assessment criteria. In that way, students would be able to create targeted prompts which instruct GenAI to generate outputs that are consistent with the course learning outcomes and standards for academic writing (Rust et al. 2005). In addition, with the course-specific AW assessment literacies, students are more likely to consistently make informed evaluative judgement of GenAI outputs (Bearman et al. 2023).

One practice we conducted was providing a feedback checklist, designed based on the marking rubric, for students to make self-evaluation of their drafts and asking them to craft prompts by integrating the specific evaluation items included in the checklist. First, it is essential that students understand the substance of each checklist item. The lecturer checked the students’ understanding by asking questions and invited them to refer back to the teaching materials. For example, one of the statements in the checklist was “You have used a variety of cohesive features within and between paragraphs”. The students should know what “cohesive features” mean and how these can be used within and between paragraphs to enhance textual cohesion, which was covered in the course materials. If the students are not clear about the instructions, they should review the class content or ask for clarification, because they would need to evaluate GenAI’s feedback on this aspect in the later stage.

After that, we asked the students to reflect on and identify the needs of using GenAI in providing feedback. They were asked to consider on which checklist items they would prompt GenAI to provide feedback and on which they would trust their self-evaluation more or it would be more efficient to manually make evaluation and corrections based on what they learnt. During this process, the students engaged with the assessment criteria more in-depth by reviewing the evaluative statements in the checklist and anticipating GenAI’s ability of providing certain feedback. Then, the students were invited to engineer prompts for receiving feedback on the items they chose. Some sample prompts were provided to show how students could incorporate the checklist statements into their prompts. They were also asked to assess whether the feedback provided was aligned with the specific requirements in the marking rubric and the teaching materials and how they would address the GenAI feedback.

Another practice was guiding the students to use GenAI to help them comprehend and respond to teacher feedback on their drafts (Carless and Boud 2018). First, the lecturer gave the students some time to read through the individual feedback to try to understand it and make a plan of revising their drafts accordingly. During this process, some students had difficulty comprehending and interpreting the teacher feedback or thinking of ways to address the feedback to improve their writing (Winstone et al. 2017). To help with this, the lecturer asked the students to prompt GenAI to explain the feedback they found challenging to understand or unpack the terminology used in the feedback. For example, “lack of unity” in the feedback seemed abstract and unfamiliar to some students, and they used GenAI for providing further explanation in relation to their own writing by asking for specific examples in their drafts to demonstrate the “lack of unity”. Due to practicality, teachers may not be able to provide sufficient examples for a feedback comment, but asking GenAI to search writing text consistent to the teacher feedback facilitates students’ understanding of the feedback and provides reference as for how they could make revisions. After fully comprehending the feedback, students also used GenAI to provide suggestions on revision, especially when they were not sure how to make changes according to the feedback. When doing this, the students were reminded to take control over the whole revision process and evaluate GenAI outputs before adopting any of its suggestions. The students were also encouraged to use GenAI to receive extra feedback to supplement the teacher feedback on their revisions.

4.4 Critical thinking

In addition, we trained the students to conduct critical evaluation of GenAI outputs for AW. Facing such a powerful tool, learners are likely to become dependent on it, especially for assignments or assessments that they are required to complete after class at their own pace. However, over relying on the tool for producing writing may not be helpful for EAP learners, especially EFL learners, to learn the target language and make progress (Barrot 2023). Therefore, we offered learning opportunities where the students could enhance their critical use of GenAI by analysing, reflecting on and improving the GenAI writing outputs.

One example learning activity was to guide the students to critically interact with and reflect on the GenAI’s responses to a prompt which required it to detect non-academic vocabulary, such as vague and colloquial words, and revise them into academic vocabulary. First, the lecturer reviewed the knowledge about academic writing style and common informal language with the students. Then, the students chose a possible prompt from a provided prompt bank for identifying informal words in their draft paragraph (Appendix A). After that, they input the prompt into XIPU AI and received a response which they were asked to analyse by considering: 1) whether the informal words recognised by XIPU AI were the same as or similar to what we reviewed beforehand or consistent to the ten rules of academic writing style taught in the class, and 2) whether XIPU AI was helpful in making the draft paragraph more academic and enhancing the overall quality. At the end of this step, some students reported that XIPU AI was not holistic enough to identify all the informal language mentioned in the EAP class and that XIPU AI detected some academic vocabulary in the original draft as not academic.

Subsequently, in order for XIPU AI to produce more effective outputs aligned with the course requirements, we guided the students to refine their prompts by adding some examples or types of informal words that were taught in the class (Appendix B). Next, they were asked to discuss with their peers whether the responses were different from the previous step and whether the outputs were more aligned with the teaching materials. A similar procedure was used for the students to prompt XIPU AI to revise the identified informal language, ask follow-up questions and critically evaluate the GenAI outputs. At the end of the task, students were required to make a decision by themselves whether they would revise their paragraph based on XIPU AI’s advice and to provide rationale. This practice created an opportunity for the students to learn that there may be a gap between the GenAI knowledge base and the course content especially when it is not trained to be consistent with the teaching materials and that they would need to “teach” GenAI what is covered in the course and review the GenAI responses carefully and critically to produce an end-product that meets the needs of the writing assessment. In this way, the students applied what they learnt to assess the information provided by GenAI, which involved them in exercising their evaluative judgement (Bearman et al. 2023).

This aspect of the AI literacies training seemed successful since many learners stated that they usually evaluated the GenAI outputs before they decided on whether to adopt them in the writing drafts.

4.5 Academic integrity

The fifth aspect is related to AI ethics where learners should respect academic integrity when using GenAI for AW. When we were exploring what GenAI could do for the students, we made it clear that they should not use GenAI to “write” a paragraph or an essay for them, not even chunks (e.g., sentences) of text. We adopted a relatively strict approach since students should be able to apply what they learnt from the EAP classes to their writing, independently developing their writing skills and critical thinking ability. Specifically, ethical learning practices were showcased through case analysis, writing sample analysis and use of “Turnitin”, a tool for checking plagiarism and collusion.

For the case analysis, students were provided with various scenarios where learners use GenAI appropriately or inappropriately and were invited to discuss with others whether the users in the cases were aware of academic integrity and why or why not (Ngo and Hastie 2025; Rowland 2023). Generally speaking, if the users have control over the core ideas of the writing and maintain the “ownership” of the writing product, then it should be considered as using GenAI with integrity. A potential analogy would be thinking of GenAI as a more intelligent writing tutor than “Grammarly” and “Marking Mate” which could become a helper for, rather than the author of, students’ academic writing (Kim et al. 2024).

Regarding writing analysis, a sample paragraph generated by GenAI and the other one by the module were shown to the students, and they were asked to discuss which one they thought were produced by GenAI and also whether GenAI was reliable or met the standards of the course or academic integrity in that case. One of the most noteworthy academic integrity issues was that GenAI seemed to fabricate data, usually with vague citations (e.g., “research shows that…”).

Another way to check the academic integrity was to use “Turnitin”. At the time when the students were trained to use GenAI for academic writing, the “AI detection” function on “Turnitin” was still under development, which sometimes could be unstable or unreliable, and was not officially used by the institution to penalise unethical or other inappropriate use of GenAI in students’ assessments. Despite that, it was still helpful to make the students aware of the potential risk of being caught if they do not follow rules about academic integrity. Moreover, in the GenAI-created sample paragraph mentioned above, Turnitin did highlight a few chunks of text that were plagiarised from other sources, and these may not be closely related to the main idea of the paragraph. The Turnitin report was given to the students to analyse and reflect on the pitfalls of GenAI and the importance of being the overseer and owner of their own writing products.

By the end of the training, most learners were fully aware of the concept of authorship, claiming that the essays should be originally written by themselves and that they should have control over the core ideas and textual structure.