The integration of Generative AI (GenAI) into the curriculum has often been advocated by educators since its emergence in 2022. This reflective study was conducted to evaluate the initiative taken by an English for Academic Purposes (EAP) module offered at Xi’an Jiatong-Liverpool University (XJTLU). The initiative involved incorporating GenAI into the teaching of three basic academic writing skills: doing research, paraphrasing, and summarizing. We first demonstrated how this was embedded in the learning tasks before reflecting upon the successes and challenges through a student survey, teachers’ personal recall, and an examination of reports submitted by students. High perceived usefulness and task engagement levels were observed although there were challenges related to academic integrity and cognitive offloading. Several recommendations were made to address these challenges. 

Key words: GenAI, academic writing, doing research, paraphrasing, summarizing, academic integrity

The ease of which a GenAI tool can provide student writers with information for their academic writing assignments has led to its frequent use or misuse. With the proliferation of GenAI, they now tend to resort to its use in the first place. Prior to the launch of ChatGPT, most students would have started their research with search engines and the digital library resources provided by universities. However, the use of GenAI can easily lead to academic integrity violations – whether willful or unintentional – if students submit writing assignments that are entirely generated by an AI tool. While a GenAI tool may be asked to provide references for their source use, students’ lack of awareness of its potential hallucination, for instance, in the form of source fabrication, may compound the problem. 

Against this backdrop, the importance of raising the student writers’ awareness of the unintended impact brought by GenAI cannot be overstated. The study by Gao et al. (2023) provides context by indicating that conventional plagiarism checkers used by universities cannot reliably detect texts produced by GenAI. From our own experience, these plagiarism checkers may indicate that a text is written by an AI tool to a certain percentage, but they fall short of identifying the AI-generated part of the text. As such, Peres et al. (2023) calls for measures which can ensure that assignments, for example, essays, reports, SWOT analyses, etc., are the original work of students, not the output of GenAI. They also propose training students to use GenAI and having an explicit discussion on how to use it with integrity, transparency, and honesty (see also van Dis et al., 2023). Although GenAI may pose academic integrity concerns, Kong et al. (2024) and Milano et al. (2023) advocate that the education communities incorporate GenAI tools into their curricula by making adjustment to existing pedagogies or designing new ones. XJTLU, as an English as a Medium of Instruction (EMI) institution, adopts the same stance. To implement this policy, since the end of 2023 it has provided both its academic staff and students with free access to its ChatGPT-based GenAI tool, XIPU AI.

 

This reflective study has two main objectives. The first is to showcase an initiative taken by an English for Academic Purposes (EAP) module offered to over 800 Year 2 School of Advanced Technology students in the Academic Year 2024-2025. The initiative was to integrate XIPU AI into the EAP syllabus for teaching three basic academic writing skills: doing research, paraphrasing, and summarizing. The second is to share both the successes and the challenges in the implementation of the initiative through an evaluation of students’ perceptions gathered in a survey, teachers’ personal recall, and a simple analysis of student reports generated by Turnitin plagiarism checker. 

As mentioned by Hirvela and Du (2013), second language (L2) student writers who transition from one language, culture, and rhetorical system to another one, especially in the Anglophone contexts, may find quoting, paraphrasing and summarizing very demanding. As a result, L2 academic writing courses often place a substantial emphasis on these areas of academic skills (ibid). Our module also devotes substantial focus on the development of the last two skills: paraphrasing and summarizing, being aware of the difficulties students face when integrating source materials into their writing.  

Before the start of the semester, we embedded XIPU AI in the existing conventional activities for teaching doing research, paraphrasing, and summarizing. The idea was to train our students to be familiar with the tool, and to use it responsibly so that academic integrity violations could be prevented. Their familiarity and mastery of the tool could also help them do well in their report-writing project, in which the task was to choose and evaluate one of the 7 given types of technology (refer to Figure 1). To evaluate their chosen technology, they would have to search for information on these types of technology and the criteria to be used. 

 

Figure 1 The technical report writing project

The following explains the teaching of these basic academic writing skills by using XIPU AI:

Our EAP lessons often included short and simple lead-in or warm-up activities to encourage students to use XIPU AI so that they would become familiar with the tool when searching for information. Doing this would also prepare them for the subsequent learning tasks.

This example lead-in activity asks students to guess the next word that completes the sentence about ChatGPT and then to compare their answers with those given by XIPU AI (or another GenAI tool). 

 

 

Figure 2 The lead-in activity with GenAI

The task below was designed to teach students how to use prompts such as ‘define’ and WH- and YES/NO questions for researching the topics of their writing project.

 

 

Figure 3 The prompts for research project writing

Students might not be aware that GenAI could be used to paraphrase source information for their research project writing. The figure below demonstrates how we taught them to use XIPU AI for this purpose. Instead of directly asking them to use the AI tool, we first taught them the conventional paraphrasing techniques such as using synonyms and changing sentence structures. This was important because they needed to acquire this skill for their timed-writing in the final exam, where they would have to paraphrase source information from the provided reading text.

 

Figure 4 Paraphrasing task

Similar to paraphrasing, students were also taught how to summarize by using the conventional methods such as locating main ideas in paragraphs and reading the introduction or the concluding paragraph of a longer text. After that, they were allowed to use XIPU AI for the task, and were reminded of the need to check and cite the source for the generated summary. Figure 5 below is an example of summarizing a short paragraph while Figure 6 is for summarizing a long article of approximately 1200 words, where students were then asked to rewrite the GenAI-generated summary to make the sentence structure less repetitive and more varied.

 

 

Figure 6 Task: summarizing a paragraph 

 

Figure 6 Task: summarizing an article and rewriting the GenAI-generated summary

In addition to the above three basic academic writing skills, there was one final activity designed to teach students to work with the source information provided by XIPU AI and to avoid academic integrity violations (refer to Figure 7).

 

 

Figure 7 Task: academic integrity discussion

Our critical reflection was conducted in the following section by using a combination of the following quantitative and qualitative measures. 

We measured the success of the initiative through a short survey which was administered to the students after three weeks’ exposure to the above GenAI-embedded tasks. Each statement in the survey used XIPU AI to represent GenAI; however, in class students were allowed to use other tools if they preferred. Table 1 summarizes the survey responses given by 54 students.

 

 

Table 1 The summary of student responses (N=54)

Statement 1 (Doing research): I enjoyed using XIPU AI to find information at the beginning of the EAP class.

The wording ‘to find information’ instead of ‘to do research’ was deliberately used in the statement in order to avoid confusion that could potentially be caused by the diction ‘research’. From the table, it could be seen the majority of the students (42 out of 54 or 78%) enjoyed using XIPU AI to find information for the lead-in activities with 16 students expressing strong agreement. It could be due to the fact that most lead-in activities in the module are discipline-specific topics such as Artificial Intelligence (AI) and Internet of Things (IoT). This requires students to understand certain specific knowledge before they can better take part in each lead-in activity or the ensuing activities. While 8 chose to be neutral with unknown reasons, 4 students in total expressed disagreement possibly caused by their dissatisfaction towards XIPU AI alone, not GenAI. Two of these students, in fact, mentioned in class its weakness in understanding and answering their questions as compared to other GenAI tools such as ChatGPT and Kimi. However, this might have been due to their wrong use of prompts when searching for information. 

Just as there was a high level of enjoyment in finding information, the majority (34 students or 63%) also showed a similarly high usefulness level of using XIPU AI to research their writing topics and make their final choice of technology for evaluation. This was probably because they knew they could rely on XIPU AI or GenAI for their research. However, 17 students did not express any agreement or disagreement. They were probably aware of the limitation of GenAI tools as they do not usually provide references that were required in their writing assignments, and instead, they would still need to use other resources such as Internet search engines or the University Library database. 

Paraphrasing and summarizing by XIPU AI were found to be useful by the majority of the respondents (39 and 41 respectively). This was expected because effective paraphrase was demonstrated to them in class and they were asked to try doing it by using the GenAI tool. The same was done for summarizing a short text (about 150 words) or a long text (about 1200 words). However, there were 12 students (22%) who chose ‘Neutral’ for the two academic skills, possibly thinking that any GenAI tools, not just XIPU AI, could perform the two important tasks equally well. Several students (a total of 3 for paraphrasing and 1 for summarizing) did not find XIPU AI useful, and again, this may have been caused by their perceived limitations of the tool or their preference for using other GenAI tools, believing them to be more superior. 

Overall, the survey results indicated that most students perceived the integration of GenAI in teaching the three basic academic writing skills to be highly enjoyable and useful. 

We also measured the initiative through the personal recounts and experiences shared by the teachers on the module during meetings. It was observed by most teachers that the GenAI-embedded activities had generally engaged students with the learning materials better as it was a novelty for students to use GenAI in class. This was not surprising because lead-in activities in EAP lessons typically take the form of discussion among a small group of students and, without GenAI, their discussion quality would be low and the shared knowledge would be quite limited, especially on discipline-specific topics. In contrast, with the insightful answers given by the AI tool, students became more confident in sharing the generated information and more willing to listen to one another. In terms of paraphrasing and summarizing, we witnessed the surprise on most students’ faces upon discovering the ability of GenAI to perform the two tasks, although the above survey results also showed a small percentage of students who would rather not take a stance concerning its usefulness. 

Despite the high levels of perceived enjoyment, usefulness, and engagement, some challenges regarding the use of GenAI for teaching these basic academic writing skills were noted in the final Writing Coursework submitted by students.

Uncited information, which is an indication that students have used GenAI-generated text, was found in a small number of reports. We examined 75 out of the over 800 submitted reports through random sampling and found that these reports contained some degree of information that was not considered as common knowledge and therefore needed to be cited. Being sensitive information that needs ethical approval, we could not disclose any details here. One most plausible reason for this occurrence was that these student writers may have missed their classes when the GenAI-embedded activities were taught. Another reason could be due to their misbelief. During the individual feedback session, we asked several students why some information in their introduction section was uncited. Their answers were surprising as they thought that the acquired knowledge they had researched and summarized through GenAI was theirs; therefore, it did not need to be cited. However, compared with the vast number of reports which had addressed the writing task with proper citations and paraphrases of source use, we could argue that the initiative was already a considerable success.  

As mentioned in Section 2, we first taught the conventional techniques of doing research, paraphrasing and summarizing before allowing students to use GenAI. However, during marking, we noticed instances from the students’ reports which seems to suggest that their paraphrases were completely done by a GenAI tool. It is important to note that to safeguard academic integrity, the module also implemented an additional measure: a Source Integration Chart (SIC), where students must present the original information alongside its paraphrase in a table to be attached to their report as an appendix. This was probably caused by the many deadlines they had so that they instinctively resorted to the most convenient solution offered by GenAI to address the paraphrasing task when working with source materials. 

Such cognitive offloading, defined as ‘the externalization of cognitive processes, often involving tools or external agents, such as notes, calculators, or digital tools like AI, to reduce cognitive load’, may result in a decline in cognitive engagement and skill development (Risko and Gilbert, 2016, p.4). Yamada (2003) also asserts that inferential thinking is necessary to good paraphrasing since student writers will need to make connections between the reading sources and their writing either deductively (drawing a conclusion based on statements) or analogically (noting similarities between two areas of knowledge). In our context, in addition to the possible impact on reduced engagement in critical thinking activities, of a particular concern would be the students’ language skill development that might be negatively affected by the ease and convenience provided by GenAI once they are accustomed to using it. In other words, they would have missed the opportunity to practice paraphrasing, given that it is a demanding skill that requires higher language proficiency. 

In summary, the initiative to integrate GenAI into the teaching of doing research, paraphrasing, and summarizing skills was found to be highly useful by students. Upon conducting a further critical reflection, the initiative was a considerable success evident from the observed high engagement level with the activities and the high number of reports with proper paraphrases and source acknowledgements, which indicates students’ adherence to academic integrity policy. Despite the successes, some challenges from the unintended effects such as students’ misbelief about citation for GenAI-generated information and cognitive offloading that might inhibit the development of language skills emerged and need to be given due attention. 

This study has several limitations. Firstly, it only employed a simple percentage analysis of the survey results. However, this was perceived sufficient and was a necessary part of a critical reflection for gauging the impact of our initiative. Another limitation was that the success in high compliance with academic integrity policy when working with source materials could arguably be attributed to the requirement for SIC in the writing project. Nevertheless, we contend that the teaching of the basic academic writing skills by using GenAI was a fundamental step, while the SIC served as a subsequent measure that contributed to to the successful outcome. 

A number of recommendations can be made from this critical reflection. First of all, to further educate our students on responsible use of GenAI and to mitigate its potential negative effects, it should continue to be integrated into the EAP syllabus. Repeated coverage of these basic academic writing skills in each semester could be useful in order to enable students to use GenAI responsibly and ethically. The second recommendation is to design a task with prompts that generate both sources and information, but with a subsequent task of verifying the sources through the University Library database or a search engine. Finally, future tasks should include teaching more specific GenAI tools that assist with literature reviews, such as Scopus AI and others. GenAI tools are here to stay, and ongoing efforts to educate student writers are necessary to maximize their potential while minimizing their negative impacts.

We would like to acknowledge both the direct or indirect contributions made by the EAP111 module leader Sam Doran and the other team members, without which this critical reflection would not have been possible. 

 

 

 

 

Gao, C. A., Howard, F. M., Markov, N. S., Dyer, E. C., Ramesh, S., Luo, Y., & Pearson, A. T. (2023). Comparing scientific abstracts generated by ChatGPT to real abstracts with detectors and blinded human reviewers. npj Digital Medicine, 6, 75, 1–5. https://doi.org/10.1038/s41746-023-00819-6  

Hirvela, A., & Du, Q. (2013). “Why am I paraphrasing?”: Undergraduate ESL writers' engagement with source-based academic writing and reading. Journal of English for Academic Purposes. 12(2), 87-98. https://doi.org/10.1016/j.jeap.2012.11.005 

Kong, S., Lee, J. C., & Tsang, O. (2023). A pedagogical design for self-regulated learning in academic writing using text-based generative artificial intelligence tools: 6-P pedagogy of plan, prompt, preview, produce, peer-review, portfolio-tracking. Research and Practice in Technology Enhanced Learning. 19, 030. https://doi.org/10.58459/rptel.2024.19030

Milano, S., McGrane, J. A., & Leonelli, S. (2023). Large language models challenge the future of higher education. Nature Machine Intelligence, 5(4), 333–334. https://doi.org/10.1038/s42256-023-00644-2

Peres, R., Schreier, M., Schweidel, D., & Sorescu, A. (2023). On ChatGPT and beyond: How generative artificial intelligence may affect research, teaching, and practice. International Journal of Research in Marketing, 40(2), 269–275. https://doi.org/10.1016/j.ijresmar.2023.03.001

Risko, E. F., & Gilbert, S. J. (2016). Cognitive Offloading. Trends in Cognitive Science. 20, 676–688. https://doi.org/10.1016/j.tics.2016.07.002 

 

van Dis, E. A. M., Bollen, J., Zuidema, W., van Rooij, R., & Bockting, C. L. (2023). ChatGPT: Five priorities for research. Nature, 614(7947), 224–226. https://doi.org/10.1038/d41586-023-00288-7 

Yamada, K. (2003). What prevents ESL/EFL writers from avoiding plagiarism?: Analyses of 10 North-American college websites, System, 31(2). 247-258. https://doi.org/10.1016/S0346-251X(03)00023-X.