ChatGPT and the ESL Writer: Preserving Voice and Agency in the Age of AI

The intersection of generative artificial intelligence and academic writing has produced one of the most consequential pedagogical debates of this decade. For multilingual learners, particularly international students navigating English as a Second Language (ESL) environments, the stakes extend far beyond grammar correction or stylistic polish.

The emergence of tools like ChatGPT presents a profound question: can these technologies enhance linguistic development without eroding the writer's authentic voice and intellectual agency?

A longitudinal study published in System (Elsevier) offers one of the most detailed examinations of this tension to date. Researchers followed two advanced ESL students over an entire academic year, documenting how their engagement with ChatGPT shaped both their writing proficiency and their sense of ownership over their work.

The findings reveal a nuanced landscape where AI serves simultaneously as a scaffold, a crutch, a tutor, and a potential threat to authorial identity.

This analysis unpacks the study's methodology, its key discoveries, and the practical strategies that emerged for maintaining learner agency in an AI-saturated academic environment. For educators, curriculum designers, and international students themselves, the implications are immediate and actionable.

TL;DR A year-long longitudinal study of two advanced ESL students reveals that ChatGPT can significantly support language development when deployed strategically, but unchecked reliance erodes writer agency and authentic voice. The research identifies a critical distinction between using AI as a linguistic scaffold versus a content substitute. Students who maintained metacognitive awareness and used ChatGPT for targeted language queries preserved their authorial identity while improving fluency. The study proposes a framework for "agentic AI use" that balances efficiency with developmental integrity, offering concrete strategies for international students facing institutional and peer pressure to adopt AI tools.

The Research Design: A Year in the Life of Two ESL Writers

The longitudinal methodology represents a deliberate departure from snapshot studies that capture only momentary AI interactions. By shadowing two advanced ESL students across four academic quarters, the researchers captured the evolution of habits, attitudes, and writing behaviors as they developed in response to repeated AI engagement.

Both participants were enrolled in rigorous academic programs requiring substantial written output. Their proficiency levels placed them at advanced stages of ESL acquisition, yet both reported persistent challenges with idiomatic expression, academic register, and syntactic complexity. These specific pain points became the primary sites of ChatGPT interaction.

Participant Profiles and Contextual Pressures

The first participant, a graduate student in social sciences, faced intense publication pressure from their department. The second, an undergraduate in engineering, struggled with the transition from formulaic writing to the nuanced argumentation expected in humanities electives. Both experienced explicit encouragement from peers and some instructors to use AI tools.

Institutional policies at their university remained ambiguous throughout the study period. This ambiguity created a gray zone where students experimented with varying degrees of transparency, sometimes disclosing AI use and other times concealing it. The researchers documented these decisions as data points revealing the ethical calculus students perform under pressure.

Data Collection: Beyond Self-Reported Usage

The study triangulated multiple data sources to build a comprehensive picture. Weekly writing journals captured students' reflective processes, while screen recordings documented actual ChatGPT interactions in real time. Semi-structured interviews every six weeks probed the reasoning behind specific AI queries and the emotional responses to AI-generated suggestions.

Draft-to-final comparisons allowed researchers to trace precisely which AI suggestions students accepted, rejected, or modified. This granular tracking distinguished between students who used ChatGPT as a thesaurus replacement and those who outsourced entire paragraphs. The distinction proved pivotal in understanding agency outcomes.

Core Findings: The Divergence Between Scaffolding and Substitution

The most striking pattern to emerge from the data was the bifurcation of AI usage into two distinct modes. The first mode, termed "linguistic scaffolding," involved using ChatGPT to clarify vocabulary, test sentence structures, and explore alternative phrasings while retaining full authorial control over content and argumentation.

The second mode, "content substitution," involved delegating substantive writing tasks to the AI. Students in this mode would provide a topic and request complete paragraphs, then lightly edit the output. The researchers observed that these two modes produced dramatically different outcomes in both writing development and perceived agency.

Scaffolding Behaviors: The Productive Path

Students who maintained agency treated ChatGPT as an interactive dictionary and style consultant. They would ask for three versions of a sentence, then select and modify the one that best matched their intended meaning. This process forced them to articulate their communicative goals explicitly, strengthening metacognitive skills.

These students also used ChatGPT to test hypotheses about language. They would write a sentence, ask the AI to identify awkward phrasing, and then interrogate the AI's reasoning. This Socratic approach transformed the tool from an answer machine into a thinking partner, reinforcing rather than replacing their linguistic decision-making.

Substitution Behaviors: The Erosion of Voice

The substitution pattern revealed a troubling trajectory. Students who began by asking for paragraph suggestions gradually increased their delegation, eventually requesting full sections with minimal input. Their drafts became more fluent but increasingly generic, with a detectable loss of personal idiom and argumentative specificity.

Interviews with these students exposed a paradox: they recognized the writing was not truly theirs, yet they felt unable to stop. The efficiency gains created a dependency loop, where the effort of writing independently felt increasingly arduous compared to the instant gratification of AI-generated prose.

The Agency Framework: What the Study Reveals About Writer Identity

Agency, as conceptualized in this research, extends beyond mere authorship to encompass the writer's sense of control over their rhetorical choices, their confidence in their linguistic judgments, and their ability to articulate why they write the way they do. The study operationalized agency through a composite measure of these factors.

Both participants began the year with comparable agency scores. By the study's conclusion, however, their trajectories had diverged dramatically. The scaffolding-oriented student reported increased agency, describing a growing sense of mastery over academic English. The substitution-oriented student described feeling like a "translator of AI output" rather than a writer.

The Role of Metacognitive Monitoring

The critical differentiator between the two trajectories was metacognitive engagement. Students who maintained agency consistently asked themselves why they were using ChatGPT, what they expected to learn, and whether the interaction was serving their developmental goals. This reflective practice transformed AI use from a reflexive habit into a deliberate pedagogical choice.

The researchers identified specific metacognitive prompts that correlated with positive outcomes. These included asking "What am I trying to say?" before querying the AI, and "Why did the AI suggest this change?" after receiving output. These questions forced students to articulate their communicative intentions, making the AI a tool for clarifying rather than obscuring their voice.

Emotional Dimensions of AI-Assisted Writing

The study also captured the emotional labor involved in navigating AI use. Students reported anxiety about being perceived as cheating, frustration when AI suggestions conflicted with their intuitions, and guilt when they relied on substitution. These emotions shaped their willingness to disclose AI use and their satisfaction with their writing.

Interestingly, the scaffolding-oriented student reported that successful AI interactions produced a sense of intellectual excitement, akin to solving a puzzle. The substitution-oriented student described a hollow satisfaction, where the final product looked good but felt disconnected from personal effort. This emotional contrast reinforced the behavioral divergence.

Practical Strategies for Maintaining Agency

The study's most actionable contribution lies in its articulation of concrete strategies that preserve writer agency while leveraging AI's benefits. These strategies emerged organically from the scaffolding-oriented participant's behavior and were subsequently validated through the researchers' analysis of what made those behaviors effective.

For ESL students facing pressure to use AI, these strategies offer a roadmap for engagement that prioritizes learning over efficiency. They are not prescriptive rules but rather principles that can be adapted to individual writing contexts and proficiency levels.

Strategy One: Interrogative Query Design

Instead of asking ChatGPT to "write a paragraph about X," students should pose targeted questions that require the AI to explain its reasoning. Queries like "Why is this sentence awkward?" or "What are three ways to express this idea, and what are the trade-offs?" transform the AI into a tutor rather than a ghostwriter.

This approach forces students to engage with the underlying linguistic principles rather than merely accepting surface-level corrections. The researchers found that students who adopted interrogative queries demonstrated measurable improvements in their ability to self-edit, a skill that transferred to their AI-free writing.

Strategy Two: The Revision Dialogue

Rather than accepting AI suggestions wholesale, students should engage in a multi-turn dialogue where they push back, request alternatives, and ask for explanations. This iterative process mirrors the revision conversations students might have with human writing tutors, building the same cognitive muscles.

The study documented that the scaffolding-oriented student routinely asked follow-up questions like "That version is more formal, but does it lose the emphasis I wanted?" This negotiation process ensured that the final text reflected the student's intentions, with the AI serving as a linguistic consultant rather than an author.

Institutional Implications: Rethinking ESL Pedagogy

The study's findings carry significant implications for how universities and language programs approach AI in the classroom. Current institutional responses range from outright bans to uncritical embrace, with most institutions occupying an uncomfortable middle ground of vague policies and inconsistent enforcement.

The research suggests that neither extreme serves students well. Bans ignore the reality that students will use AI regardless, driving usage underground where it cannot be guided. Uncritical embrace risks normalizing substitution behaviors that undermine the very language development that ESL programs exist to foster.

Curriculum Design for AI Literacy

Rather than treating AI as a threat to be policed or a tool to be celebrated, the study advocates for explicit instruction in AI literacy as a component of ESL curricula. Students should be taught not just how to use ChatGPT, but how to use it in ways that serve their linguistic development rather than circumvent it.

This instruction would include modules on prompt design, critical evaluation of AI output, and metacognitive reflection on AI use. The researchers argue that these skills are as essential to modern academic writing as citation conventions or thesis development, representing a new form of rhetorical competence.

Assessment Reform in the AI Era

The study also raises profound questions about assessment validity. If students are using AI in ways that obscure their true proficiency, then traditional writing assessments may no longer measure what they claim to measure. The researchers call for assessment designs that either incorporate AI transparently or create conditions where its use is unnecessary.

Process-based assessments, including portfolios of drafts and reflective essays on writing choices, offer one path forward. These formats make the writing process visible, allowing instructors to evaluate the student's developmental trajectory rather than merely the final product, which may be AI-mediated.

Beyond the Study: Broader Implications for Multilingual Writers

While the study focused on two participants, its findings resonate with the broader experience of multilingual writers navigating AI-mediated academic environments. The pressures faced by these students, including language anxiety, time constraints, and the imperative to produce native-like prose, are amplified in the AI era.

The researchers position their work within a growing body of literature examining AI's role in language education. They note that the fundamental question is not whether students should use AI, but how they should use it to support rather than supplant their developing linguistic competence.

The Digital Divide in AI Access

The study also surfaces equity concerns that extend beyond its immediate findings. Students from privileged backgrounds may have access to more sophisticated AI tools, better instruction in their use, and more time to engage in reflective practices. Less privileged students may face pressure to use AI as a shortcut, further widening proficiency gaps.

Institutions must therefore consider AI literacy as an equity issue, ensuring that all students receive the guidance they need to use these tools developmentally. Without such intervention, AI risks becoming another mechanism by which educational advantages compound.

Future Research Directions

The researchers acknowledge the limitations of their small sample and call for larger-scale studies that can establish generalizable patterns. They specifically identify the need for research on AI use across different ESL proficiency levels, disciplinary contexts, and cultural backgrounds.

Longitudinal studies that track students beyond a single academic year would also illuminate whether agency-preserving strategies produce lasting benefits. The current study provides a foundation, but the rapidly evolving capabilities of AI tools demand continuous investigation into their pedagogical implications.

Conclusion: Charting a Path Forward for ESL Writers and Educators

The longitudinal study offers a timely and nuanced contribution to the discourse on AI in education. Its central message is neither alarmist nor utopian: ChatGPT can be a powerful ally in language development, but only when deployed with intentionality and metacognitive awareness.

For ESL writers, the takeaway is empowering. The research demonstrates that agency is not a fixed trait but a practice that can be cultivated through deliberate engagement with AI tools. By adopting interrogative query strategies, engaging in revision dialogues, and maintaining reflective journals, students can harness AI's power without surrendering their voice.

For Educators: A Call for Deliberate Pedagogy

Instructors must move beyond simplistic positions on AI and embrace their role as guides in this new landscape. This means designing assignments that make writing processes visible, teaching AI literacy as a core competency, and creating classroom cultures where students feel safe disclosing their AI use.

The researchers emphasize that the goal is not to police AI use but to ensure that it serves pedagogical objectives. When students understand how to use AI as a scaffold for their development, they become more confident, more capable writers who can navigate the complexities of academic English with authenticity and agency.

Final Reflections on the Study's Contribution

This research represents a significant step forward in understanding the human dimensions of AI-assisted writing. By focusing on the lived experiences of ESL students over an extended period, it reveals patterns that cross-sectional studies cannot capture and offers strategies grounded in real-world practice.

As AI tools continue to evolve, the principles articulated in this study, interrogative engagement, metacognitive reflection, and selective acceptance, will likely remain relevant. The specific prompts and techniques may change, but the underlying imperative to maintain writer agency in the age of AI will only grow more urgent.