ARTICLES

 

Artificial intelligence meets academic writing: Examining ChatGPT's affordances and limitations in South Africa

 

 

Mampoi Irene Mabena

English Department, Faculty of Humanities, University of Johannesburg, Johannesburg, South Africa. mampoim@uj.ac.za. https://orcid.org/0000-0003-1651-0833

 

 

---

ABSTRACT

Many studies have explored the affordances of ChatGPT for English language learning. However, these have overlooked its use in academic literacy courses in higher education institutions. Therefore, in this paper, I aim to explore the affordances and limitations of ChatGPT when used to explain features of academic writing, compose academic expository essays, and give feedback on draft essays in South Africa. The data was derived from conducting the usability testing activity, during which I interacted with the ChatGPT tool to respond to questions guiding this paper. The findings were interpreted from the perspective of critical pedagogy. The findings reveal that ChatGPT can do more than what it is unable to do; it can generate content that meets the requirements of an academic literacy course and produce a well-written expository essay that adheres to many of the conventions of academic writing. Findings further reveal that ChatGPT can provide valuable personalised feedback on draft sections of academic essays. These findings suggest that, depending on how it is used, ChatGPT can act as a knowledge depositor, reinforcing the banking model of education or can support a more dialogic, empowering form of learning. I argue that the Eurocentric education currently in practice is likely to foster students' overreliance on ChatGPT, thereby reinforcing the banking model of education rather than the dialogic problem-posing approach.

Keywords: ChatGPT, generative AI, expository essay, academic literacy, critical pedagogy

---

 

 

Introduction

Generative AI has emerged at a time when South African education is still struggling to transform from having its roots in colonial and apartheid histories. One part of this struggle involves the continued adherence to a monolingual approach to teaching and learning, despite teaching mostly Black students who lack proficiency in the language of instruction. English has maintained its hegemony as the medium of instruction, communication, and the primary language of research, while African languages, spoken by most of the student population, remain on the periphery, and their promotion is merely a constitutional imperative, rather than an actionable practice (Rakgogo, 2024). South African scholars have proposed the use of translanguaging as an effective counter to the English-only approach to teaching and learning (Mabena, 2023; Makalela, 2019; Mbirimi-Hungwe, 2021; Motaung, 2024; Probyn, 2019; Yafela, 2021). However, despite these efforts, South African education remains monolingual.

Language is one way the Western episteme still dominates in academic literacy courses (Lillis & Turner, 2001), which ignore largely African students' unique ways of being, knowing, and doing that stem from their lived experiences (Angu, 2018). Ndlovu-Gatsheni (2018) has argued that Africans always read, wrote, and argued in their communities before their modes of communication were suppressed by colonisation. Assessment practices compound these struggles since assessment is primarily generic and adheres to perceived universal norms rooted in the Western conceptualisation of literacy, despite the recognition that literacy practices are discipline- and society-specific. Assessment approaches often disregard students' social and cultural diversity (Angu et al., 2019). Students are required to write essays in a specific academic language, adhering to all the conventions of so-called academic writing, an approach which disregards students' lived experiences, and they are generally assessed in ways that adhere to "Western norms and standards" (Crossouard & Oprandi, 2022, p. 182).

If we bear these higher education contexts in mind, the emergence of generative AI, such as ChatGPT, can be seen to pose particular challenges in South Africa. The uncritical use of ChatGPT in contexts that already privilege English, Western academic conventions, and reproduced assessment practices may exacerbate existing inequalities. Moreover, when assessment practices continue to disregard students' social and cultural diversity, students may come to view ChatGPT as a shortcut for obtaining the correct academic answer rather than as a dialogic partner in knowledge construction. My focus in this paper is on assessing the affordances of ChatGPT and determining if it can assist students in bridging the epistemological gap in higher education or if it will exacerbate existing cultural norms by providing another global North-oriented source of knowledge.

Several studies have investigated the use of ChatGPT in higher education. Much of the current debate centres on ChatGPT's capabilities and limitations in promoting teaching and learning, as well as its potential to substitute for a teacher (see Baskara & Mukarto, 2023; Bin-Hady et al., 2023; Jeon et al., 2025; Kohnke et al., 2023). For instance, Kohnke et al. (2023) provided insight into how ChatGPT supports language teaching and learning, while scholars such as Baskara and Mukarto (2023) and Bin-Hady et al. (2023) have explored mainly the opportunities and drawbacks of using ChatGPT to support teaching and learning. According to these authors, ChatGPT supports teaching and learning by developing learners' writing and conversational skills and engaging them in interactive tasks. It also provides feedback on students' work and suggests relevant learning activities to further assist students, allows personalised and independent learning experiences, identifies contextual meanings of words and dictionary definitions, corrects grammatical and word choice errors, while also adjusting complex language by providing a simplified version according to students' needs. However, the main concern is this tool's potential to promote dishonesty in academia. Jeon et al (2025) further explored GenAI systems from a translanguaging point of view and found that despite their proficiency in many languages, GenAI systems operate from the perspective of statistical probability, favouring languages already prevalent while under-representing those in the minority.

In South Africa, a plethora of research has examined the role of ChatGPT in higher education and students' learning experiences (Chauke et al., 2024; Huallpa et al., 2023; Mlambo, 2024; Singh, 2023; Sokhulo et al., 2025; Tarisayi, 2024). The research motives of these scholars seem similar-to explore the impact of ChatGPT on teaching and learning, and on academic integrity. However, the target research participants differed. For instance, Mlambo (2024) and Tarisayi (2024) examined the responses of academics and lecturers, while others assessed undergraduate and postgraduate students (Chauke et al., 2024; Huallpha, 2023). Generally, the findings indicate that while ChatGPT is a convenient and effective learning resource that promotes students' comprehension of content, its use also raises concerns about plagiarism and academic integrity, as well as having the potential to erode critical thinking among its users.

As shown above, studies on the use of ChatGPT in academic literacy courses at higher education institutions are underrepresented, despite their critical foundational role, which should have attracted research attention. Therefore, this study contributes to ongoing debates on ChatGPT's abilities and impact by specifically examining the affordances and limitations of ChatGPT (a version of GPT-4o) in its presentation of academic literacy content and in the composition of expository essays acceptable in South African higher education environments. This will determine how well this tool promotes Freire's (1970) banking model of education or his alternative, problem-posing educational approach. The paper is guided by the following questions:

1. What are the capabilities and limitations of ChatGPT-4o in explaining specific elements required in expository essays?

2. What are the capabilities and limitations of ChatGPT-4o in creating a full academic expository essay with all the required features included?

3. What are the capabilities and limitations of ChatGPT-4o in providing insightful feedback on essay drafts?

 

Literature review

Chat Generative Pre-trained Transformer (ChatGPT) was invented by Microsoft-backed company OpenAI in November 2022 (Ray, 2023). Among many AI-powered chatbots (e.g., Lummi, Polite Post, Sci Summary, Audio Box, Meta AI), ChatGPT is among the most advanced (Kohnke et al., 2023). It can engage effectively in conversation with its users, identify the contextual meanings of words, explain and correct language mistakes, create various kinds of texts such as stories, emails, and recipes for various dishes, generate quizzes and comprehension passages and questions, interpret texts, provide dictionary based definitions, and simplify complex discourse to make it more reader friendly for beginners (Baskara & Mukarto, 2023). Nonetheless, the main concern identified by many academics is the ethics surrounding the use of ChatGPT in education; there are concerns that it promotes dishonesty and cheating among students; this has a negative impact on assessments and, consequently, the credibility of qualifications (Cassidy, 2023). Furthermore, the originality and accuracy of some of ChatGPT's responses are also questionable since it does not cite sources to validate its answers. According to Baskara and Mukarto (2023), ChatGPT occasionally provides inaccurate responses. It offers users the option to agree or disagree with its suggestions by displaying thumbs up or down, indicating some level of uncertainty on its part. At other times, it provides incorrect answers with authority and confidence, often misleading its users. The other weakness of this chatbot is that it is culturally biased; the texts in the database emanate mainly from the English corpus, implying that culturally based questions that are outside the scope of English hegemony may not receive a satisfactory response (Bin-Hady et al., 2023). Generative AI continues to improve and advance, making it essential for us to create an archive of its various stages of development and competency. In this article, I aim to contribute to the record of what generative AI can do at this stage, specifically in the context of an academic literacy course. It is also important to note that while OpenAI's ChatGPT is foregrounded here, other LLMs have become popular and the abilities of one versus another are contested vociferously.

 

Freire's critical pedagogy

I interpret the academic literacy learning assisted by technology from the perspective of the critical pedagogy framework formulated by Freire (1970). The tenets of this theoretical framework align with the characteristics of generative AI. For instance, Freire conceptualised education as a liberating tool rather than one that dominates others. He felt that it is a tool that should develop learners' critical thinking skills, enabling them to interrogate and challenge injustices, oppression, and unfair systems in their communities, and contribute to building a fair and humane society. He was very critical of a banking concept of education, which he described as a learning situation in which teachers become depositors of knowledge, while students become passive recipients of that knowledge (Alm & Watanabe, 2023). Generative AI has the potential to develop students' critical thinking and contribute to building a humane society, but it can also merely replicate banking education depending on how it is used. Freire further stipulated that learning should be grounded in the lived realities of learners, their languages, and cultural practices; in South Africa, ChatGPT does not recognise this. However, it does allow students to have agency (and some privacy) without feeling ashamed of their level of ability to learn academic skills. Freire proposed a dialogic principle, a problem-posing pedagogy in which the teacher and the student become co-creators of knowledge through engagement in a dialogue in a process of knowledge transfer. The problem-posing pedagogy model is said to foster mutual respect since students' prior knowledge is validated and their contributions are valued, thereby dismissing the assumption that teachers have all the answers. The model is therefore said to enable students to become agents of change as they apply their learning to the real world (Alm & Watanabe, 2023). ChatGPT can support a more dialogic, empowering form of learning by providing personalised feedback that students can analyse critically to improve their work. However, if not carefully interrogated, it can reinforce the banking model and turn students into passive recipients of knowledge. This framework offers a critical point of view through which to explore the affordances of ChatGPT and its potential to either reinforce the banking model or empower students as agents of change and co-creators of knowledge. Alm and Watanabe (2023, p. 22) have confirmed that Freire's "banking versus empowering, dialogic education" is a pertinent framework for analysing AI tutors like ChatGPT.

 

Methodology

This study was motivated by the high volume of students' essay submissions that the learning management system flagged as AI-generated. These were well-written essays which deserved good marks had they been written by the students themselves. Moreover, during individual consultations with students, questions were asked about some misalignment between what they were taught in class and what they received from the AI tools they used, indicating an active and appropriate use of these tools. I became keen to assess the capabilities of ChatGPT, one of the AI tools popular among students (see Chauke et al., 2024).

I employed a qualitative research design (see Creswell, 2014). This choice was based on the qualitative nature of the research questions since they aimed to closely analyse and interpret ChatGPT's abilities to explain the features of academic expository essays, generate an academic essay, and provide feedback on essay drafts. The data collection instrument was the ChatGPT chatbot itself. Based on the research questions, ChatGPT was best suited to provide all the answers for this study, so I employed purposive sampling to select a suitable AI system for it. Among the several AI chatbots, ChatGPT stands out as the most relevant since it is regarded as one of the most advanced AI-powered chatbots capable of addressing all fields and areas of knowledge.

The data collection tool I adopted was a usability testing activity. This testing method is designed to assess the functionality of digital products, such as websites, applications, and chatbots, as real users attempt to complete a task on them (Moran, 2019). It is designed to identify the challenges and opportunities associated with a particular tool. In this study, I conducted a usability testing activity to explore ChatGPT's affordances and limitations in explaining the features of expository academic essays and in creating such essays while adhering to the conventions governing them. I submitted questions related to the study's objectives online so ChatGPT could respond. The questions were rephrased if the answers provided were not precise or relevant. The questions asked were basic information-seeking ones, typical of those that students themselves would ask based on the AI-generated essays they had produced for assessment and the questions they had asked about module content information they received from the AI. While serving as the lecturer for the Academic Literacy module, I used the module's typical content to assess the accuracy of ChatGPT's responses. The data collection procedure reflects the authentic engagement of a first-year student attempting to complete a task. I evaluated the full expository essay generated by ChatGPT using the rubric I created in my capacity as module coordinator and specialist in academic literacy content. Regarding ethical considerations, it is essential to note that this study is non-intrusive since it involved no direct contact with human participants through interviews or observations. Instead, the data was derived from the AI tool ChatGPT. This chatbot is publicly available to everyone who wishes to use it.

 

The findings

The online test with ChatGPT was conducted in January 2025. The test revealed several aspects of its ability to explain specific elements required in expository essays and to create a comprehensive academic essay with all the necessary features. The findings are categorised into the following themes: (1) Ability to distinguish characteristics of academic and non-academic writing; (2) Ability to identify elements of the introduction, body paragraph, and conclusion of an expository essay; (3) Ability to generate a full essay, and (4) Ability to give insightful feedback on essay drafts.

Theme 1: Ability to distinguish characteristics of academic and non-academic writing

The findings reveal that ChatGPT's information on what is required for academic writing was accurate, and it successfully differentiated between academic and non-academic writing. Excerpt 1 illustrates my conversation with ChatGPT and its performance in completing the given task.

Excerpt 1

Researcher: Please be a lecturer for the first-year university academic literacy module. Draw a table where you show the characteristics of academic and non-academic writing.

ChatGPT: Certainly! Here is a table outlining the key characteristics of academic and non-academic writing.¹

 

 

As shown in excerpt one, ChatGPT successfully completed the requested task. It listed accurately the characteristics of academic and non-academic writing, which align with the academic writing features discussed by Beekman et al. (2023) and by Chin et al. (2012). I inquired further about the use of the passive voice, personal pronouns, and emotive and cautious language, which should have been included among the characteristics mentioned, and received an accurate response, as shown in Table 2.

 

 

As shown in Table 2, ChatGPT produces probabilistic text that reflects the content covered in the academic literacy course. This finding aligns with points made in earlier work by Kohnke et al. (2023) and Baskara and Mukarto (2023), who have shown that ChatGPT supports language teaching and learning by providing explanatory notes on various topics. Essentially, ChatGPT delivered a mini lecture (a list of facts) on command, which mirrors one-way knowledge transmission rather than encouraging critical thinking (see Freire, 1970).

Theme 2: Ability to identify elements of the introduction

Regarding the introductory paragraph for the expository essay, I requested ChatGPT to provide a specific structure or list the elements students must follow when writing their introduction. Excerpt 2 provides the ChatGPT response.

Excerpt 2

ChatGPT:

Sure! In an expository essay, the introduction serves to engage the reader and set the scene for the detailed descriptions that will follow. Here are the key elements typically found in the introduction of an expository essay:

Hook: The opening sentence or two is designed to grab the reader's attention. This could be an intriguing statement, a question, a quote, or a vivid description that hints at the main subject of the essay.

Context/Background Information: A few sentences that provide some background or context for the subject being described. This helps the reader understand the setting, the significance, or the main idea of the description.

Thesis Statement: A clear and concise statement that presents the main point or the dominant impression you aim to convey in your essay. This thesis statement guides the overall direction of your descriptions and provides a focus for the essay.

Overview of Main Points: Briefly mention the key points or aspects that will be described in more detail in the body of the essay. This gives the reader a preview of what to expect and helps them follow the structure of the essay.

As shown in its response in excerpt 2, ChatGPT not only lists the elements that make up the structure of an introduction but also mentions the purpose served by the introduction and it provides an explanation for each element. It even provides an example of an introduction, breaking it down into the key elements to enhance students' comprehension. However, I probed further when I realised that ChatGPT did not mention any sources in its explanation.

Excerpt 3

Researcher: Excellent! The structure is clearly defined. However, since this is an academic essay, where does citation feature in this introduction?

ChatGPT:

In an academic essay, citations are crucial for providing evidence and supporting claims, even in expository essays. While the introduction of an expository essay may not be as heavily cited as the body, it can still include citations, especially when establishing context or background information.

ChatGPT's response regarding citations is accurate. It is correct in saying that an introduction does not include many citations and it is also accurate in stating that citation is tied mainly to context and background information, which is an area that requires some factual or historical information about the subject matter. Additionally, ChatGPT explains that the other elements of an introduction do not necessarily require citations. It further provides an example of an introduction with cited background information and explains the importance of adhering to the specific citation style required by the assignment (e.g., Harvard, APA, MLA, Chicago) and of including a full reference list of the cited sources. ChatGPT appears not to confine itself to the parameters of the question; it usually provides more than is required and offers greater clarity on a subject. My findings further reveal the importance of asking specific questions that yield the correct response and the significance of probing when the response is inadequate. ChatGPT is an AI tool trained on Western knowledge, and since epistemologies in South African higher education remain Westernised, it navigates its stored data and provides accurate answers easily. Nonetheless, I did not observe ChatGPT asking for clarification, encouraging me to consider something, or personalising its explanations. Rather, it merely transfers facts, as in Freire's (1970) concept of banking education, implying that students who interrogate it this way are likely to receive one-way, replicated knowledge.

Theme 3: Elements of expository body paragraph

ChatGPT was asked further to list and explain the elements that must be included in the body paragraph of an expository essay. This is illustrated in the conversation that follows.

Excerpt 4

Researcher: Thanks, but please list and explain the elements of a body paragraph for an academic expository essay

ChatGPT:

Certainly! Here are the key elements of a body paragraph for an academic expository essay:

1. Topic Sentence

2. Explanation and Context

3. Detailed Descriptions

4. Sensory Details

5. Examples and Evidence

6. Analysis and Interpretation

7. Concluding Sentence

The chatbot described each mentioned element to ensure a thorough understanding. The listed elements are similar to those that first-year university students are generally taught in their academic literacy lectures (Beekman et al., 2023). However, in its list, it includes detailed descriptions and sensory details that are typical of a narrative essay and not an expository one. This implies that ChatGPT does not use the distinction between expository and narrative essays. According to Chin et al. (2012) and Nenotek et al. (2022), the body paragraph is comprised of a topic sentence, an explanation of the topic sentence, examples and evidence, an interpretation of the evidence, and a concluding sentence. This means that relying solely on ChatGPT's description may mislead novice students so lecturers should clarify these differences and ensure that learners comprehend the content fully.

Theme 4: Elements of a concluding paragraph

The last task on the essay structure that ChatGPT was requested to perform was to provide an overview of the structure of a concluding paragraph for an expository essay. The purpose was to determine if it was conversant with the key elements that must be included in this part of the essay.

Excerpt 5

Researcher: How do we now conclude? What elements are included in a conclusion paragraph for an academic expository essay?

ChatGPT:

The conclusion of an academic expository essay is crucial for summarizing the main points, reinforcing the thesis, and providing a final perspective on the topic. Here are the key elements to include in a conclusion paragraph for an academic expository essay:

1. Restatement of the Thesis

2. Summary of Main Points

3. Synthesis of Information

4. Final Insight or Thought

5. Implications or Call to Action (if relevant)

As with the body paragraph, ChatGPT explained each element and provided examples to illustrate it. However, it appears to have two extra elements listed in three (synthesis of information) and four (Final insight or thought), which could easily be merged with elements 2 (summary of the main points) and 5 (Implications or call for action), respectively. Students are taught primarily that a concluding paragraph has three main elements: a restatement of the thesis; a summary of the main points; and a final thought or verdict (Scholtz, 2016). The synthesis of information is usually part of the summary of the main points, while the final thought involves implications or a call to action. Nonetheless, ChatGPT treats them as separate elements, which may cause some confusion for a first-year student. As the findings affirm, ChatGPT does not engage in dialogic learning by involving the questioner in this knowledge exercise. Rather, it merely generates information as instructed, implying that students who interact with it in a similar way will become passive recipients of knowledge while ChatGPT becomes a knowledge depositor (see, again, Freire, 1970).

Theme 5: Generating a full academic essay

ChatGPT was tested on its ability to create a full expository essay on one of the topics assigned to the students which includes those elements with which it seems very conversant. The exercise was designed to assess its accuracy in identifying the structural components crucial to a well-written academic essay and to determine the extent to which it can help students create one. The topic given was "Discuss the impact of social media on human communication." The essay was evaluated based on the rubric used to assess students' work on a similar assessment at a large urban university in Johannesburg. The general impression was that the essay was well written since it scored 62%. As shown in Table 3, the marks were allocated as follows.

 

 

The scoring indicates that most of the elements that ChatGPT listed earlier were included in its generation of a full essay. However, the entire essay lacked evidence, which made it appear less academic. Citations were missing despite ChatGPT's emphasis on the critical role played by citations and referencing in academic essays. Earlier, it stated, "In an academic essay, citations are crucial for providing evidence and supporting claims, even in expository essays." One of the body paragraphs from the essay is quoted in Excerpt 6.

Excerpt 6

Social media has significantly enhanced the ease and speed of communication, fostering global connectivity. The instantaneous nature of these platforms allows users to share updates, photos, and videos in real-time, bridging geographical gaps and creating a sense of closeness despite physical distance. This has been particularly beneficial in maintaining personal relationships, as individuals can stay in touch with family and friends regardless of location. Furthermore, social media has democratized information dissemination and consumption. Platforms like Twitter and LinkedIn have become crucial for professional networking, enabling individuals to share industry news, insights, and opportunities with a broad audience, thus expanding professional horizons.

This excerpt is a body paragraph that contains almost all the elements, the topic sentence, and its supporting details. However, as articulated in the mark allocation, the concluding sentence is missing, and there is no citation to authenticate the information provided. This raises doubts about the validity of the information provided and undermines the academic nature that this essay should reflect. ChatGPT was further interrogated and asked to include sources in the essay and a reference list at the end. It rewrote the essay and included the sources. Most of these sources, seven out of the ten, were books. These were relevant but outdated sources, given the topic being discussed and how students are usually encouraged to cite sources published in the past 10 years (Beekman et al., 2023), unless it is seminal work. The following are some of the cited and referenced sources.

Excerpt 7

Carr, N. (2010). The Shallows: What the Internet Is Doing to Our Brains. W.W. Norton & Company.

Crystal, D. (2008). Texting: The Gr8 Db8. Oxford University Press

Marwick, A. E., & Boyd, D. (2011). I tweet honestly, I tweet passionately: Twitter users, context collapse, and the imagined audience. New Media & Society, 13(1), 114-133.

Excerpt 7 showcases some of the sources included in the essay, which, upon verification, were found to be accurate and relevant to the topic. This indicates that ChatGPT requires further prompting until it provides the expected answer. Kohnke et al. (2023) asserted that the quality of the response depends on the quality of the questions, so it is important to rephrase them when one is not satisfied with the response. Basically, how it is prompted determines the results, which implies that unethical users can, at least potentially, score high marks if they prompt it correctly. However, its first response is not to provide any sources; I prompted it several times before it included any. In fact, the first time, it failed to include sources despite being asked. The second time, it cited a few sources without a reference list. It was the third prompting that yielded the best results, specifically six citations with corresponding references. Citations and references are a key weakness in the current ChatGPT-generated text. Nonetheless, it produced a passing academic essay, suggesting that students who prompt this tool to write their essays are likely to pass the assignment. However, their development of critical writing skills may be undermined since their voices are suppressed by the tool, which reinforces that banking model of learning.

Theme 6: ChatGPT's Feedback on work submitted for its review

I went on to provide ChatGPT with the elements students are taught to include in their introductory paragraph. I then provided two samples of an introduction, one of which was very good and the other not as well-structured. The purpose was to assess whether ChatGPT can provide valuable feedback on students' work, should students request this feedback. The feedback received was very positive. The tool broke the paragraphs into the elements with which it was provided and commented on whether each paragraph met the requirements. It also commented on the grammar, which was not mentioned as part of the assessment. It identified the language errors and showed how these could be corrected. See below for one of the comments it produced on general background knowledge.

Excerpt 8

ChatGPT:

According to Themane (2023:162), structural issues in South African schools, such as poor infrastructure and unequal resources significantly prevent effective teaching and learning, especially in primary schools.

Comment: You have provided a relevant background detail and correctly included a citation. The sentence is well-linked to the topic and helps contextualise the problem.

Minor issue: there's a small punctuation omission after resources (should be "resources, significantly prevent")

As shown above, ChatGPT quoted a section of the introduction which it regarded as general background knowledge and commented on its successful inclusion. Not only does it identify punctuation and grammatical errors, it also provides feedback on how to correct each error. After providing detailed feedback on each element of the introduction, it offered an overall comment. The following is a general comment on one of the introductions that it reviewed.

ChatGPT said:

The introduction is thoughtful, well-informed, and logically structured. It covers all the essential elements and shows critical engagement with social inequality in education. The main weaknesses are language accuracy, punctuation, and a lack of citation in the background section.

The comments on each element of the introduction, the correction of language and punctuation errors, and the overall feedback provided by ChatGPT suggest that, instead of producing AI-generated essays and receiving zeros when this is detected, students can dialogically utilise ChatGPT to provide personalised feedback on the essay drafts they produce. Students can then learn from this feedback, verify it with a lecturer or a tutor, and finally produce an essay that meets the module's expectations. This shows that ChatGPT can support a more dialogic, empowering form of learning by engaging students in questioning and providing personalised feedback. This is an ideal form of learning that aligns with a dialogic principle: ChatGPT (acting as a tutor) and the student become co-creators of knowledge through dialogue, facilitating knowledge transfer.

 

Discussion

The findings of this paper reveal ChatGPT's ability to generate content that meets students' course requirements. It can characterise and differentiate academic writing from non-academic writing and provide examples of how each aspect can be used, which adds clarity for the user.

The chatbot has also generated a passing expository essay that meets many of the conventions of academic writing. Further, I found that ChatGPT can play the role of a tutor by providing valuable personalised feedback on students' draft essays. Nonetheless, ChatGPT appears to be better at explaining the requirements of a well-written academic essay than actually producing one; it had to be prompted several times to include citations and a reference list, both of which it claimed were crucial to academic writing. The prompting reveals ChatGPT's mixed ways of responding: sometimes it provides more than is required (when explaining the elements required for each section of an expository essay), and at other times it produces less than expected. Furthermore, the sources it ultimately produces are not recent despite the recency of the topic and the availability of recent works in the field. The other concern is that some of the elements ChatGPT listed as requirements for the body and conclusion paragraphs are not necessarily stand-alone elements and could be overshadowed by other components it has already mentioned, while others are confused with elements prominent in narrative essays, which is misleading. Despite these shortcomings, there is more that ChatGPT can do than what it cannot. It can still generate a passing essay and serve as a learning resource that students can use to co-create knowledge.

These findings suggest that, depending on how it is interrogated, ChatGPT can act, at least potentially, as a knowledge depositor thus reinforcing the banking model of education (Freire, 1970), or it can support a more dialogic, empowering form of learning by providing personalised feedback that students can analyse critically to improve their work. This means that students stand at a crossroads of amplification or disruption. ChatGPT can either deepen epistemic dependency or become a catalyst for rethinking academic literacy. Previous research has demonstrated similarly how ChatGPT is a convenient and effective learning resource when used dialogically in the co-creation of knowledge, but it has also raised concerns about the erosion of critical thinking in instances where ChatGPT is used unethically to generate knowledge on behalf of students (Kohnke et al., 2023; Sokhulo et al., 2025; Tarisayi, 2024). However, South African higher education students face inherent learning challenges that are likely to promote the unethical use of ChatGPT. As explained earlier, the monolingual practices persistent in higher education (the English-only approach to teaching), epistemologies and assessment practices that remain foreign to Black students' lived experiences act as a motivating factor to students' use of ChatGPT as an all-knowing tutor, thereby reinforcing the banking model of education in which students passively receive knowledge instead of co-creating it. Students might well want to mitigate their learning challenges by consulting a technology tool that is as Westernised as the education they receive to generate knowledge for them.

 

Conclusion

ChatGPT has been found to be an effective additional teaching and learning tool since it provides the necessary content for an academic literacy programme. However, these capabilities must be harnessed intentionally to leverage fully their potential for more contextualised and personalised learning. This implies that educators should foster social justice and an equitable society by integrating ChatGPT into Freire's problem-posing pedagogy which encourages critical thinking and the co-creation of knowledge. For instance, students could use ChatGPT to explain certain aspects of an essay topic and challenge its response by comparing it with their own interpretation. Furthermore, students could use the tool during the drafting of the different sections of their essays by asking ChatGPT for feedback on the clarity of their argument, coherence, tone, and grammar, and then evaluate critically the suggestions it provides to decide whether to accept or reject them. This way, they would be testing their own ideas before consulting a lecturer. Approaching ChatGPT in this way means using it from a Freirean dialogic perspective, as a thought-stimulating tool, rather than an authority or essay generator. Institutions of higher learning should also introduce an AI ethics course that provides practical examples and activities that promote social justice. They should also establish systems to detect the unethical use of AI effectively and impose severe penalties for those who violate these rules. Additionally, in-person invigilated exams should be re-enacted to mitigate a surge in students' AI-generated essays. ChatGPT's affordances should be leveraged to promote effective teaching and learning and should therefore be approached from a problem-solving perspective rather than as a banking model that deposits knowledge into passive consumers. The chatbot (ChatGPT) itself could be redesigned for educational purposes to use more questioning to encourage dialogic learning. Most importantly, this paper serves as a call to revisit the South African higher education curriculum. The monolingual approach to teaching and learning, as well as Eurocentric epistemologies and assessment practices, should be decolonised to address the lived realities of students and their ways of knowing and being. This would mitigate over-reliance on and the unethical use of AI chatbots such as ChatGPT. Currently, ChatGPT may not be particularly helpful with this alien aspect of our curriculum, since it is based on Western canonical knowledge. However, once the curriculum decolonisation has been achieved and ChatGPT is fed with this new knowledge, it can support a more dialogic and empowering form of learning and help students with learning that is grounded in their lived realities, languages, and cultural practices, as we envision our decolonised curriculum to be.

I acknowledge the limitation of this study; the use of my module content and rubric to evaluate ChatGPT's performance may reflect some bias and reflexivity. However, the content and rubric align closely with general first-year academic literacy practices and with how essays produced in this module are typically evaluated. Nonetheless, future research could involve more independent raters to enhance objectivity.

 

References

Albiladi, W. S., & Alshareef, K. K. (2019). Blended learning in English teaching and learning: A review of the current literature. Journal of Language Teaching and Research, 10, 232 - 238. http://dx.doi.org/10.17507/jltr.1002.03        [ Links ]

Alm, A., & Watanabe, Y. (2023). Integrating ChatGPT in language education: A Freirean perspective. Iranian Journal of Language Teaching Research, 11(3),19-30.         [ Links ]

Angu, P. E. (2018). Disrupting Western epistemic hegemony in South African universities: Curriculum decolonisation, social justice, and agency in post-apartheid South Africa. International Journal of Learner Diversity and Identities, 25(1), 9-22. http://doi.org/10.18848/2327-0128/CGP/v25i01/922        [ Links ]

Angu, P. E., Boakye, N., & Eybers, O. (2019). Rethinking the teaching of academic literacy in the context of calls for curriculum decolonization in South Africa. International Journal of Pedagogy and Curriculum, 27(1), https://doi.org/10.18848/2327-7963/CGP/v27i01/1-16        [ Links ]

Avello-Saez, D., & Estrada-Palavecino, L. (2023). ChatGPT and its impact on competence training in occupational therapists: A reflection on academic integrity. Cadernos Brasileiros de Terapia Ocupacional, 31(e3534). https://doi.org/10.1590/2526-8910.ctoEN271035342        [ Links ]

Baskara, R., & Mukarto, M. (2023). Exploring the implications of ChatGPT for language learning in higher education. Indonesian Journal of English Language Teaching and Applied Linguistics, 7(2), 343-358. http://dx.doi.org/10.21093/ijeltal.v7i2.1387        [ Links ]

Beekman, L., Dube, C., Potgieter, H., & Underhill, J. (2023). Academic Literacy (4th ed.). Juta.

Bian, N., Han, X., Sun, L., Lin, H., Lu, Y., & He, B. (2023). ChatGPT is a knowledgeable but inexperienced solver: An investigation of commonsense problem in large language models. https://doi.org/10.48550/arXiv.2303.16421

Bin-Hady, W. R. A., Al-Kadi, A., Hazaea, A., & Ali, J. K. M. (2023). Exploring the dimensions of ChatGPT in English language learning: A global perspective. https://www.emerald.com/insight/0737-8831.htm

Cassidy, C. (2023, January 9). Australian universities to return to 'pen and paper' exams after students caught using AI to write essays. The Guardian. https://www.theguardian.com/australia-news/2023/jan/10/universities-to-return-to-pen-and-paper-exams-after-students-caught-using-ai-to-write-essays

Chauke, T. A., Mkhize, T. R., Methi, L., & Dlamini, N. (2024). Postgraduate students' perceptions on the benefits associated with artificial intelligence tools for academic success: The use of the ChatGPT AI tool. Journal of Curriculum Studies Research, 6(1), 44 - 59. https://doi.org/10.46303/jcsr.2024.4        [ Links ]

Chin, P., Reid, S., Wray, S., & Yamazaki, Y. (2012). Academic writing skills student's book 2. Cambridge University Press.

Chuah, K. M., & Kabilan, M. K. (2021). Teachers' views on the use of chatbots to support English language teaching in a mobile environment. International Journal of Emerging Technologies, 16 (20), 223-237. http://dx.doi.org/10.3991/ijet.v16i20.24917        [ Links ]

Clarence, S., & McKenna, S. (2017). Developing academic literacies through understanding the nature of disciplinary knowledge. London Review of Education, 15(1): 38-49. https://doi.org/10.18546/LRE.15.L04        [ Links ]

Creswell, J. W. (2014). Research design: Qualitative, quantitative and mixed methods approaches. SAGE.

Crossouard, B., & Oprandi, P. (2022). Decolonising formative assessment. In J. Huisman & M. Tight (Eds.), Theory and method in higher education research (pp.181-196). Emerald Publishing Limited. https://doi.org/10.1108/S2056-375220220000008010

Freire, P. (1970). Pedagogy of the Oppressed. Bloomsbury Publishing.

Godsell, S., Shabangu, B., & Primrose, G. (2024). Against colonial residues, towards decolonising assessment: A case study of a university history course, Cogent Education, 11(1): 1-16. https://doi.org/10.1080/2331186X.2024.2362552        [ Links ]

Huallpa, J. J., Arocutipa, J. P., Panduro, W. D., Huete, L. C., Limo, F. A., Herrera, E. E., Callacna, R. A., Flores, V. A., Romero, M. A., Quispe, I. M., & Hernandez, F. A. (2023). Exploring the ethical considerations of using ChatGPT in university education. Periodicals of Engineering and Natural Sciences, 11(4), 105-115.         [ Links ]

Jeon, J., Wei, L., Tai, K. W. H., & Lee, S. (2025). Generative AI and its dilemmas: Exploring AI from a translanguaging perspective. Applied Linguistics, 46, 709-717. https://doi.org/10.1093/applin/amaf049        [ Links ]

Kohnke, L., Moorhouse, B. L., & Zou, D. (2023). ChatGPT for language teaching and learning. RELC Journal, 54 (2), 537-550. http://dx.doi.org/10.1177/00336882231162868        [ Links ]

Lillis, T., & Turner, J. (2001). Student writing in higher education: Contemporary confusion, traditional concerns. Teaching in Higher Education, 6 (1): 57-68. http://doi.org/10.1080/13562510020029608        [ Links ]

Mabena, M. I. (2023). Learners' reception of translanguaging pedagogy as a strategy for reading English L2 texts. Journal for Language Teaching, 57(2). https://doi.org/10.4314/jlt.v57i2        [ Links ]

Makalela, L. (2019). Uncovering the universals of ubuntu translanguaging in classroom discourse. Classroom Discourse, 10 (3/4), 237-251. https://doi.org/10.1080/19463014.2019.1631198        [ Links ]

Mbirimi-Hungwe, V. (2021). An Insight into South African multilingual students' perceptions about using translanguaging during group discussion. Applied Linguistics, 42 (2), 252-268. https://doi.org/10.1093/applin/amaa012        [ Links ]

Mlambo, P. B. (2024). Exploring the adoption of ChatGPT in higher education: A case of lecturers in a University of Technology in KwaZulu-Natal, South Africa. Paper presented at EDULEARN24 Conference, Palma, Spain.

Moran, K. (2019). Usability (User) testing 101. https://www.nngroup.com/articles/usability-testing-101/

Motaung, L. B. (2024). Translanguaging pedagogical practice in a tutorial programme at a South African university. International Journal of Language Studies,18 (1), 81-104.         [ Links ]

Ndlovu-Gatsheni, S. J. (2018a). Epistemic freedom in Africa: Deprovincialization and decolonization. Routledge.

Ndlovu-Gatsheni, S. J. (2018b). The dynamics of epistemological decolonisation in the 21st century: Towards epistemic freedom. Strategic Review for Southern Africa, 40(1), 16-45.         [ Links ]

Nenotek, S. A., Tlonaen, Z. A., & Manubulu, H. A. (2022). Exploring university students' difficulties in writing English academic essay. Al-Ishlah: Jurnal Pendidikan, 14 (1), 909-920. http://dx.doi.org/10.35445/alishlah.v14i1.1352        [ Links ]

Probyn, M. (2019). Pedagogical translanguaging and the construction of science knowledge in a multilingual South African classroom: Challenging monoglossic/post-colonial orthodoxies. Classroom Discourse, 10 (3/4), 216-236. https://doi.org/10.1080/19463014.2019.1628792        [ Links ]

Rakgogo, T. J. (2024). A linguistic evaluation of the South African higher education sector: A reflection on 30 years of democracy (1994-2024). Transformation in Higher Education, 9(0), a342. https://doi.org/10.4102/the.v9i0.342        [ Links ]

Ray, P. P. (2023). ChatGPT: A comprehensive review on background, applications, key challenges, bias, ethics, limitations and future scope. https://doi.org/10.1016/j.iotcps.2023.04.003

Scholtz, D. (2016). Improving writing practices of students' academic literacy development. Journal for Language Teaching, 50 (2), 37-55. http://dx.doi.org/10.4314/jlt.v50i2.2        [ Links ]

Singh, M. (2023). Maintaining the integrity of the SA university: The impact of ChatGPT on plagiarism and scholarly writing. South African Journal of Higher Education, 37(5), 203-220. https://dx.doi.org/10.20853/37-5-5941        [ Links ]

Sokhulu, L. H., Zulu, M. W. & Lott-Naidoo, D. (2025). Artificial intelligence in higher education South Africa: The role of ChatGPT in students' learning. Journal of Education, 99, 143-165. https://doi.org/10.17159/2520-9868/i99a08        [ Links ]

Tarisayi, K. S. (2024) ChatGPT use in universities in South Africa through a socio-technical lens. Cogent Education, 11(1), 1-11. https://doi.org/10.1080/2331186X.2023.2295654        [ Links ]

Yafele, S. (2021). Translanguaging for academic reading at a South African university. South African Linguistics and Applied Language Studies, 39(4), 404-424. https://doi.org/10.2989/16073614.2021.1981767        [ Links ]

 

 

Received: 12 May 2025
Accepted: 25 February 2026

 

 

1 While The Journal of Education uses British spelling, Chat GPT uses American. These extracts have not been edited.