Adopt, adapt, or reject? Analysing
student-AI interaction in university curriculum-aligned writing tasks
Yeong-Ju Lee[1]*,
Rauno Parrila1, & Danielle Colenbrander1
|
*Correspondence: |
|
1Australian
Centre for the Advancement of Literacy, Australian Catholic University,
Australia |
Abstract
Artificial
intelligence (AI) is reshaping academic writing instruction. However, there is
limited evidence on how students actually interact
with AI and how pedagogy can guide ethical and reflective engagement. Piloted
in a first-year undergraduate English language and literacy unit at an
Australian university, this paper presents the design and trial of
curriculum-aligned, task-based writing materials comprising an AI-assisted
guide with structured prompts to support essay preparation with AI. Analysis
draws on AI-interaction logs from a case study, complemented by a survey of 207
students. Thematic coding and interactional move analysis of AI-interaction
logs observed four recurrent moves – clarification, uptake, adaptation, and
rejection – in which the learner negotiated AI feedback. Survey data indicated
that while students viewed AI as useful for efficiency and confidence, concerns
about plagiarism and over-reliance persisted. This study offers early insights
into how structured scaffolds can support ethical, reflective engagement with
AI and inform future research.
Keywords:
Artificial intelligence, academic writing, task-based language teaching,
scaffolding, student-AI interaction, AI-interaction logs
1. Teaching and Learning Context
Undergraduate students often struggle with
academic writing, especially in linguistically diverse contexts like Australia,
where around 30% of students are international (Australian Government, Department
of Education, 2025), and 25% of households use languages other than English
(LOTE) (Australian Bureau of Statistics, 2021). In this context of higher
education, artificial intelligence (AI) chatbots such as ChatGPT, Gemini, and
Claude are rapidly transforming how students approach academic writing (Barrot,
2023; Jeon & Lee, 2023; Wang et al., 2024; Zhang et al., 2024). However,
current implementations of AI in language and literacy learning often remain
unstructured and disconnected from pedagogy (Lee et al., 2026; Wu, 2024),
raising a need for structured, curriculum-aligned approaches to AI-supported
writing.
This paper presents an innovative practice piloted
in a first-year undergraduate English language and literacy unit at an
Australian university to support students’ AI-assisted writing practices. In
the unit, students were allowed to use AI to prepare for a major essay
assignment, and guidance on ethical AI use was provided in tutorials, including
a template for acknowledging whether or not they used AI. Drawing on Task-Based
Language Teaching (TBLT) and scaffolding theory (Ellis et al., 2019), we developed
a curriculum-aligned four-week writing task sequence to scaffold essay
preparation in two conditions: AI-assisted writing and self-reflection writing.
This paper focuses on the AI-assisted writing condition to illustrate how
structured scaffolds can shape learner-AI interaction and inform task design.
The analysis draws on detailed AI-interaction logs from one student who
voluntarily completed the full AI-assisted writing sequence and on survey data
from a wider cohort (n = 207) to provide contextual understanding of broader
student practices on AI use.
2. Rationale for Innovation
Recent developments in generative AI and large
language models (LLMs) have transformed language and literacy education, with
chatbots such as ChatGPT, Gemini, and Claude now capable of providing
contextually relevant, personalised feedback and sustaining goal-oriented
dialogue (Cotton et al., 2024; Sánchez-Torrón et al.,
2025; Wang et al., 2024; Warschauer & Xu, 2024; Zhang et al., 2024). In the
context of AI-supported language learning and teaching, studies demonstrate
that AI can facilitate autonomous revision, improve writing coherence, and
enhance student confidence, emphasising the importance of mediation through
teacher guidance and reflective interaction (Annamalai et al., 2023; Chen et
al., 2023; Jeon & Lee, 2023; Michel et al., 2025; Teraif,
2025; Wu, 2024). However, concerns persist regarding superficial pedagogical
alignment, limited teacher preparation, and ethical risks such as plagiarism
and authorship ambiguity (Ahn et al., 2024; Alm, 2024; Lee et al., 2026).
Collectively, the findings of these studies suggest that AI’s effectiveness
depends on how its use is mediated, scaffolded, and aligned with pedagogical
goals.
Despite increasing interest and evidence of AI use
in language education, three key gaps remain. First, while most studies focus
on perception data (Alm, 2024; Pack & Maloney, 2023), there is limited
evidence of how students actually engage with AI feedback during writing, for
example, through turn-taking, selective uptake, or rejection. Second, few
interventions embed AI within curriculum-aligned, assessment-linked tasks.
Third, concrete models for fostering ethical and reflective AI use in academic
writing remain limited.
Addressing these gaps, the innovation presented in
this paper embeds AI use within scaffolded, curriculum-aligned writing tasks
informed by TBLT and scaffolding theory (Ellis et al., 2019). TBLT provides a
design-oriented framework for developing authentic, goal-driven activities in
which tasks serve as units of both instruction and assessment, integrating
linguistic, cognitive, and reflective processes. Scaffolding theory complements
this by conceptualising guided support that can be gradually adjusted to foster
learner independence. Together, these theories offer a principled foundation
for designing AI-assisted writing tasks that promote guided participation,
critical decision-making, and autonomy.
3. Description of the Innovation
The innovation involved the development of a
curriculum-aligned, four-week sequence to support students’ preparation for an
essay assignment, evaluating texts for suitability in early literacy
instruction. Drawing on TBLT and scaffolding principles, the sequence comprised
interrelated writing activities modelling guided participation and gradual
independence, providing a structured progression across four academic writing
sub-skills: vocabulary, clarity, coherence, and argumentation. The tasks were
integrated into students’ essay drafting process, with structured AI prompts
provided for interaction with AI. The prompts were designed to guide critical
evaluation of potential revisions and articulation of the reasoning behind
possible changes. They modelled cognitive processes for analysing and refining
writing through AI feedback, embedding opportunities to adopt, adapt, or reject
AI-generated suggestions. Scaffolding was gradually reduced to promote autonomy
and ethical decision-making. Materials comprised:
l An AI-interaction writing
guide: A weekly sequence of writing tasks and AI prompts designed to help
students generate synonyms, paraphrase, check coherence, and build arguments.
Each task required evaluation of AI suggestions (e.g., adopting, adapting, or
rejecting them) to encourage active reasoning and decision-making (see Appendix
1).
l An AI usage guide: A
supplementary resource outlining principles for responsible AI engagement,
including prompt design, evaluation of outputs, and reflection strategies to
avoid plagiarism or over-reliance.
This innovative sequence was piloted as part of a
curriculum implementation within the unit. One participant completed four
weekly tasks using the AI-interaction guide with ChatGPT Plus, sharing
AI-interaction logs capturing prompts, responses, and revision decisions. To
contextualise the pilot within broader student practices, a cohort-wide survey
(n = 207) was also conducted to examine purposes of AI use, perceived benefits
and challenges, and attitudes toward responsible engagement.
4. Reflection
This pilot implementation generated insights at
two levels: micro-level evidence from AI-interaction logs and macro-level
cohort perspectives. A mixed-methods approach captured both process-level
interaction data and wider cohort trends. Qualitative analysis of
AI-interaction logs employed thematic discourse coding to identify recurring
functions of AI use (vocabulary, clarity, coherence, argumentation) and
interactional move analysis to classify exchanges as clarification, uptake,
adaptation, or rejection (see Appendix 2). Descriptive quantitative analysis of
survey responses extended these insights to the wider cohort, identifying
common purposes of AI use, perceived benefits, ethical concerns, and calls for
clearer guidance (see Appendix 3). Insights from the logs and survey were
integrated to reveal both individual usage patterns and broader student needs.
Together, these data illuminate how structured, curriculum-aligned scaffolding
can shape reflective and ethical engagement with AI in academic writing.
4.1
Micro-level
insights: Development through scaffolded interaction
Thematic and interactional move analyses of the
AI-interaction logs identified four recurrent moves – clarification (requests
for further information), uptake (adoption of AI suggestions), adaptation
(modified integration of AI suggestions), and rejection (explicit non-adoption
of AI suggestions) – that structured engagement with AI feedback. These moves
aligned with the staged task design across the four weeks.
In the vocabulary and clarity tasks (Weeks 1-2),
AI functioned primarily as a lexical and stylistic resource. The student
frequently requested synonyms or simpler rephrasings
and often incorporated AI’s responses after clarification. Exchanges were
dominated by clarification and uptake, yet even at this early stage she
compared alternatives before selecting contextually appropriate terms.
Structured prompts therefore supported attention to disciplinary tone rather
than mechanical copying.
In the coherence task (Week 3), interaction
shifted toward evaluative engagement. Suggested transitions or revisions were
frequently modified or rejected when they appeared generic. Adaptation became
the dominant move, with AI outputs serving as prompts for reflection rather
than solutions. The student extended dialogue by questioning structural
suggestions and testing alternatives before revising. Feedback thus became a
site of negotiation rather than acceptance.
In the argumentation task (Week 4), engagement
developed into dialogic reasoning. The student increasingly sought
justification, questioned the sufficiency of evidence, and considered
alternative viewpoints before reinforcing her position. Interaction moved
beyond surface-level revision toward metacognitive reflection on rhetorical
purpose and instructional sequencing. Multi-turn exchanges supported reasoning
about coherence, persuasiveness, and pedagogical fit.
A developmental trajectory emerged across the four
weeks: from frequent uptake of AI suggestions for lexical refinement to
selective adaptation at the discourse level, and finally to dialogic reasoning
in argumentation. This progression demonstrates how scaffolded prompts can
foster increasingly reflective and self-regulated engagement. Rather than
positioning AI as an answer-generating tool, the structured sequence required
evaluation, comparison, and justification, operationalising ethical engagement through
task design.
4.2
Macro-level
insights: Cohort perspectives
Survey findings (see Appendix 3) contextualise
these case-level insights within broader cohort practices. AI use was
widespread (72%), with ChatGPT as the dominant platform (87%), followed by
Grammarly (30%), QuillBot (12%), and Copilot (11%),
with a small number of students also reporting use of Gemini or Perplexity. The
most common purposes were brainstorming ideas (58%), grammar or spelling
support (54%), and structuring written work (48%). Smaller groups reported
using AI for paraphrasing (24%) and referencing (30%), indicating that AI use
extended beyond idea generation into textual reformulation and citation-related
practices. Efficiency (62%), overcoming writer’s block (49%), and increased
confidence (36%) were frequently cited benefits. Students therefore viewed AI
as a tool for sustaining writing momentum.
At the same time, concerns were substantial:
over-reliance (61%), plagiarism risks (47%), and ‘robotic’ or generic outputs
(42%) were frequently reported. Several responses reflected ongoing uncertainty
about authenticity and authorship, with students expressing hesitation about
how much AI use remained acceptable. More than half of respondents (55%)
expressed a need for clearer guidance regarding acceptable and responsible use,
including explicit expectations and practical demonstrations of prompt design.
Taken together, the case and survey data converge
most clearly at the lexical and early drafting levels. In the case, frequent
requests for synonyms and paraphrasing were observed; similarly, survey
respondents most often reported using AI for grammar support and idea
generation. This convergence suggests that AI is already embedded in
early-stage writing processes, but that movement toward discourse-level
reasoning may depend on explicit modelling and scaffolded prompts.
4.3
Key
lessons
The case and survey data indicate that AI’s
pedagogical value depends not on access alone, but on structured,
curriculum-aligned mediation. In this pilot context, AI use concentrated most
strongly on lexical precision and sentence-level clarity unless guided by
staged prompts requiring comparison, justification, and decision-making. When
structured prompts required the writer to adopt, adapt, or reject AI
suggestions and explain reasoning, engagement extended toward discourse-level
reasoning and argument development. The innovation demonstrates that ethical AI
literacy can be operationalised through design rather than through restriction.
By embedding structured decision points into assignment-linked tasks, abstract
concerns about plagiarism and over-reliance were translated into visible
moments of evaluative reasoning. This aligns with calls for reflective and
mediated AI integration in writing instruction (Lee et al., 2026; Pack &
Maloney, 2023), while extending these discussions by illustrating how scaffolding
can be embedded within authentic assessment preparation.
The developmental trajectory observed across the
four weeks further illustrates how scaffolded sequencing can support gradual
movement from surface-level revision toward dialogic and metacognitive
engagement. While previous classroom-based studies have reported improvements
in coherence, autonomy, and motivation (Ahn et al., 2024; Wu, 2024; Michel et
al., 2025), this pilot provides process-level insight into how such development
may unfold through structured interaction with AI.
The survey findings suggest that many students
remain uncertain about acceptable and ethical AI use, echoing broader concerns
about autonomy and academic integrity (Alm, 2024; Cotton et al., 2024). They
add evidence to the need to move beyond policy statements toward embedded
instructional modelling. The innovation demonstrates that structured prompts and
annotated task-based demonstrations may reduce ambiguity and foster more
deliberate, self-regulated engagement.
In summary, this pilot indicates that AI’s
perceived benefits – efficiency, confidence, and drafting support – can be
channelled toward reflective and educationally purposeful practice when
mediated through intentional task design. These insights offer a foundation for
refining curriculum-aligned materials and for developing more principled
frameworks for AI integration in academic writing.
4.4
Impact
This study has both pedagogical and theoretical
impacts. Pedagogically, it offers a design-oriented model for integrating AI
into academic writing instruction by demonstrating how AI interaction can be
embedded within curriculum-aligned, assessment-linked task sequences. By
requiring the learner to adopt, adapt, or reject AI suggestions and articulate her
reasoning, the intervention operationalised evaluative engagement within
writing practice. Ethical AI use is thus guided through structured
decision-making embedded in task design. The study also provides a concrete
example of how scaffolding can be distributed across a staged progression of
writing sub-skills. Rather than relying solely on general guidance about
responsible AI use, the AI-assisted writing guide aligns AI interaction with
specific academic objectives and makes reflective engagement a visible
component of revision. This study thus offers a framework for educators seeking
to integrate AI into writing instruction while maintaining disciplinary
standards and fostering metacognitive awareness.
Theoretically, the study advances a process-level
account of AI-mediated writing development grounded in observable learner
behaviour. The identification of recurrent interactional moves (clarification,
uptake, adaptation, and rejection) provides a structured lens for analysing how
learners negotiate AI feedback within task-based writing processes. This interactional
perspective complements emerging research examining learner engagement with
generative AI in language learning contexts (Michel et al., 2025; Zhang et al.,
2024). This reframes AI literacy as regulated participation in mediated
dialogue rather than mere tool use. The findings also extend task-based
language teaching and scaffolding theory into learner-AI interaction contexts.
By encoding scaffolding within prompt design and task sequencing, the study
illustrates how instructional mediation can be structured through task design.
5. Limitations
This pilot study has several limitations. Using a
single learner constrains the generalisability of the findings. Nonetheless, it
demonstrates that analysing AI-interaction logs can yield valuable pedagogical
insights for academic writing instruction. Although triangulation between the
case and survey data provided useful contextual insights, larger-scale
interventions are needed to determine whether these patterns hold across
diverse learners and contexts. Future research should involve larger cohorts to
allow systematic comparison of outcomes and measure the influences of AI
interactions on writing performance through controlled designs to robustly
evaluate effectiveness.
6. Future Directions
This small-scale pilot study demonstrates how AI
can be integrated into academic writing through structured and reflective
design. It offers practical guidance for educators and exploratory directions
for researchers seeking to build pedagogically grounded approaches to
AI-assisted language and literacy learning.
For educators, the study highlights AI literacy
development through task design. Structured prompts that require students to
justify whether they adopt, adapt, or reject AI suggestions help transform
assistance into reflection. By embedding these scaffolds within authentic,
curriculum-linked tasks, teachers can cultivate students’ critical awareness
and ethical reasoning. These findings also point to the need for professional
learning that enables educators to model responsible prompting and engagement with
AI tools.
For researchers, the innovation offers a
replicable methodological model for examining student-AI interaction. The use
of AI-interaction logs and thematic coding demonstrates how process-level
analysis can capture the dynamics of regulation, evaluation, and adaptation.
Future research should extend this approach to explore the longitudinal
development of AI literacy and test the scalability of different scaffold types
across disciplines.
Effective integration of AI in language education
requires both pedagogical intentionality and empirical grounding. When guided
by structured design and reflective practice, AI can serve as a catalyst for
ethical judgment, learner agency, and deeper engagement with academic
discourse.
References
Ahn,
J., Lee, J., & Son, M. (2024). ChatGPT in ELT: Disruptor? Or well-trained
teaching assistant? ELT Journal, 78(3), 345–355. https://doi.org/10.1093/elt/ccae017
Alm,
A. (2024). Exploring autonomy in the AI wilderness: Learner challenges and
choices. Education Sciences, 14(12),
Article 1369. https://doi.org/10.3390/educsci14121369
Annamalai,
N., Eltahir, M. E., Zyoud, S. H., Soundrarajan,
D., Zakarneh, B., & Al Salhi, N. R. (2023).
Exploring English language learning via chatbot: A case study from a
self‑determination theory perspective. Computers & Education: Artificial
Intelligence, 5, Article 100148. https://doi.org/10.1016/j.caeai.2023.100148
Australian
Bureau of Statistics. (2021). Australia 2021 Census
all persons QuickStats. https://www.abs.gov.au/census/find-census-data/quickstats/2021/AUS
Australian
Government, Department of Education. (2025). International
student monthly summary and
data tables. https://www.education.gov.au/international-education-data-and-research/international-student-monthly-summary-and-data-tables
Barrot,
J. S. (2023). Using ChatGPT for second language writing: Pitfalls and
potentials. Assessing Writing, 57, Article 100745. https://doi.org/10.1016/j.asw.2023.100745
Chen,
Y., Jensen, S., Albert, L. J., Gupta, S., & Lee, T. (2023). Artificial intelligence
(AI) student assistants in the classroom: Designing chatbots to support student
success. Information Systems Frontiers, 25, 161–182. https://doi.org/10.1007/s10796-022-10291-4
Cotton,
D. R. E., Cotton, P. A., & Shipway, J. R. (2024). Chatting and cheating:
Ensuring academic integrity in the era of ChatGPT. Innovations in Education
and Teaching International, 61(2), 228–239. https://doi.org/10.1080/14703297.2023.2190148
Ellis,
R., Skehan, P., Li, S., Shintani, N., & Lambert, C. (2019). Task-based
language teaching: Theory and practice. Cambridge University Press.
Jeon,
J., & Lee, S. (2023). Large language models in education: A focus on the
complementary relationship between human teachers and ChatGPT. Education and
Information Technologies, 28, 15873–15892. https://doi.org/10.1007/s10639-023-11834-1
Lee,
S., Choe, H., Zou, D., & Jeon, J. (2026). Generative AI (GenAI) in the
language classroom: A systematic review. Interactive Learning Environments,
34(1), 335–359. https://doi.org/10.1080/10494820.2025.2498537
Michel,
M., Bazhutkina, I., Abel, N., & Strobl, C.
(2025). Collaborative writing based on generative AI models: Revision and
deliberation processes in German as a foreign language. Journal of Second
Language Writing, 67, Article 101185. https://doi.org/10.1016/j.jslw.2025.101185
Pack,
A., & Maloney, J. (2023). Using generative artificial intelligence for
language education research: Insights from using OpenAI’s ChatGPT. TESOL
Quarterly, 57(4), 1571–1582. https://doi.org/10.1002/tesq.3253
Sánchez-Torrón, M., Ipek, E., & Enríquez Raído, V. (2025).
Creating terminological correspondence recognition tests with GPT-4: A case
study in English-to-Turkish translations in the engineering domain. International
Journal of Artificial Intelligence in Education, 35, 2185–2201. https://doi.org/10.1007/s40593-025-00465-x
Teraif,
Z. (2025). Can AI be friends? An innovative approach to integrating generative
AI as a critical friend in Bahrain Polytechnic's English foundation programme. Innovation in Language Learning and Teaching.
Advance online publication. https://doi.org/10.1080/17501229.2025.2551124
Wang,
X., Pang, H., Wallace, M. P., Wang, Q., & Chen, W. (2024). Learners’ perceived AI presences in AI-supported language
learning: A study of AI as a humanized agent from community of inquiry. Computer
Assisted Language Learning, 37(4), 814–840. https://doi.org/10.1080/09588221.2022.2056203
Warschauer,
M., & Xu, Y. (2024). Artificial intelligence for language learning: Entering a new era.
Language Learning & Technology, 28(2), 1–4. https://doi.org/10.64152/10125/73569
Wu,
Y. (2024). Study on the impact of utilizing ChatGPT and other AI tools for
feedback in EAP writing classrooms on the discursive writing performance of
English major students. Transactions on Social Science, Education and
Humanities Research, 4, 143–150. https://doi.org/10.62051/4se95x52
Zhang,
R., Zou, D., & Cheng, G. (2024). A review of chatbot-assisted learning:
Pedagogical approaches, implementations, factors leading to effectiveness,
theories, and future directions. Interactive Learning Environments, 32(8), 4529–4557. https://doi.org/10.1080/10494820.2023.2202704
Author
bio
Dr Yeong-Ju Lee obtained her PhD from the Department of
Linguistics at Macquarie University. She teaches courses in Applied
Linguistics, Language Education, and Literacy. Her research interests include
digital language learning and teaching, focusing on digital literacy,
multimodality, social media, digital games, and AI. She is the Chief
Investigator of the Teaching Development Grant-funded project on AI and
literacy at Australian Catholic University, and the Data Horizon Research
Centre-funded project on AI and language learning at Macquarie University.
ORCID: 0000-0002-9251-6535
Email: yeong-ju.lee@acu.edu.au & yeong-ju.lee@mq.edu.au
Prof Rauno Parrila is a Professor and the
Centre Director of Australian Centre for the Advancement of Literacy. His
research focuses on psychological, linguistic, and social correlates of both
typical and atypical development of reading and academic achievement. He has a keen
interest in reading instruction and interventions, reading development and
difficulties in different orthographies, and compensation of learning
difficulties in both children and adults.
ORCID: 0000-0003-4250-8980
Email: rauno.parrila@acu.edu.au
Dr Danielle
Colenbrander is a Lecturer in the
Australian Centre for the Advancement of Literacy. She teaches undergraduate
and postgraduate initial teacher education courses in language and literacy.
Her research investigates word reading, spelling, and reading comprehension assessment
and instruction, with a particular focus on the roles of vocabulary and
morphology. With a keen interest in translating research into practice, she has
worked closely with teachers and teaching assistants across Australia and the
UK.
ORCID: 0000-0001-5577-7501
Email: danielle.colenbrander@acu.edu.au