ESL Students and ChatGPT: Coaching Without Outsourcing Your Thinking

Generative artificial intelligence has stormed into academic writing with the force of a tidal wave, and ESL students stand at the very epicenter of this transformation. The allure is undeniable: instant grammar correction, vocabulary suggestions, and coherent paragraph structures delivered in milliseconds.

Yet beneath this convenience lurks a profound pedagogical crisis that threatens to undermine the very purpose of language education. When learners outsource their thinking to algorithms, they risk becoming passive consumers of machine-generated prose rather than active architects of their own academic voices.

The longitudinal study referenced in this analysis tracked ESL students over an extended period, examining whether sustained generative-AI use displaces or supports learner agency in academic writing. The findings challenge simplistic narratives about technology as either savior or saboteur.

What emerges is a nuanced portrait of dependency patterns, cognitive engagement, and the delicate balance between assistance and autonomy. This article dissects those findings, offering a pragmatic framework for leveraging ChatGPT as a coaching tool without surrendering intellectual ownership.

For educators, administrators, and language learners alike, the stakes could not be higher. Academic integrity policies are scrambling to catch up with technological reality, while students navigate an ethical gray zone where the boundaries between collaboration and plagiarism blur.

The solution lies not in prohibition but in structured engagement—establishing explicit rules that transform AI from a crutch into a catalyst for genuine linguistic development. This analysis provides that roadmap, grounded in empirical evidence and practical classroom wisdom.

TL;DR Sustained ChatGPT use among ESL students creates a double-edged sword: it accelerates surface-level writing fluency while potentially stunting deep cognitive engagement and independent authorship. The longitudinal study reveals that learner agency—the capacity to make intentional rhetorical choices—erodes when AI tools are used passively. However, structured coaching protocols that position AI as an interactive tutor, not an answer generator, preserve and even enhance agency. This article establishes concrete usage rules, distinguishes legitimate scaffolding from academic dishonesty, and provides actionable strategies for educators and learners to harness AI without outsourcing thinking.

The Agency Paradox: How AI Assistance Reshapes ESL Writing Cognition

The longitudinal data paints a troubling picture of cognitive offloading among ESL writers who rely heavily on generative tools. When students paste prompts and receive polished paragraphs, they bypass critical decision-making processes essential to language acquisition. Vocabulary selection, syntactic variation, and rhetorical structuring—all core competencies—remain undeveloped when algorithms perform these operations invisibly. The study's participants who used ChatGPT extensively showed measurable declines in independent revision skills over time.

Yet the research also identifies a counterintuitive finding: moderate, strategic AI use correlates with improved writing confidence and willingness to engage with complex topics. The paradox lies in how the tool is deployed.

Students who treat ChatGPT as a brainstorming partner, asking it to generate counterarguments or alternative phrasings, demonstrate enhanced cognitive engagement compared to those who request complete sentences. The difference hinges on whether the learner remains the primary decision-maker throughout the writing process.

Learner agency, defined as the capacity to make intentional choices about content, structure, and expression, emerges as the critical variable separating productive from corrosive AI integration. When students delegate entire paragraphs, agency evaporates. When they interrogate AI outputs, challenge its suggestions, and selectively incorporate feedback, agency strengthens.

The study's longitudinal design reveals that these patterns compound over time, creating either virtuous cycles of growing independence or vicious cycles of deepening dependency.

Deconstructing the Dependency Spiral

The dependency spiral begins innocuously with grammar checking and vocabulary lookup, tools that genuinely support language development. However, the transition from consultation to substitution happens gradually, often unnoticed by both student and instructor. Learners begin requesting sentence completions, then paragraph structures, then entire essay drafts.

Each step reduces the cognitive load required for writing, making the AI crutch increasingly indispensable. The study documents this trajectory across multiple academic semesters.

Writing fluency, paradoxically, appears to improve during this dependency period because AI-generated text eliminates the friction of language production. Students produce longer, grammatically cleaner assignments with less effort. Yet this apparent progress masks stagnation in underlying competence.

When the study's participants were asked to write without AI assistance, their unassisted output revealed significant gaps between their apparent and actual proficiency levels. The scaffolding had become a substitute for skill rather than a support for its development.

Metacognitive awareness—the ability to monitor and regulate one's own thinking—also deteriorates under conditions of heavy AI reliance. Students who outsourced writing tasks reported difficulty articulating why they made specific rhetorical choices, a hallmark of diminished agency. They could describe what the AI produced but struggled to explain their own authorial intentions.

This disconnect between output and understanding represents the most insidious form of cognitive erosion documented in the study.

Intervention timing proves crucial. Students who received structured guidance about AI use early in their academic careers developed healthier patterns than those who discovered the tools independently. The study suggests that explicit instruction about AI's limitations, combined with assignments designed to require original thinking, can interrupt the dependency spiral before it becomes entrenched. Educational institutions must therefore treat AI literacy as a core competency rather than an optional add-on.

Scaffolding vs. Substitution: The Critical Distinction

Educational scaffolding theory provides a useful lens for distinguishing productive from harmful AI integration. Legitimate scaffolding involves temporary support that gradually withdraws as competence develops. A language coach who provides sentence frames, then sentence starters, then mere prompts, exemplifies this principle. ChatGPT can function similarly when deployed deliberately. The key question becomes whether the support structure diminishes over time, forcing the learner to internalize skills.

Substitution, by contrast, involves permanent replacement of learner effort with external assistance. When AI generates complete texts that students submit with minimal modification, the tool has shifted from scaffold to substitute.

The study's data shows that substitution patterns correlate strongly with declining writing self-efficacy—students begin to doubt their own capabilities precisely because they never exercise them. This self-fulfilling prophecy entrenches dependency further.

Effective scaffolding requires intentional design. Assignments must be structured so that AI assistance addresses specific, identified weaknesses rather than circumventing entire tasks. For instance, a student struggling with thesis statements might use ChatGPT to generate three possible thesis formulations, then evaluate and refine each one.

This approach maintains the learner's decision-making authority while leveraging AI's generative capacity. The study found that such targeted use produced measurable gains in unassisted writing performance.

Assessment practices must also adapt to distinguish scaffolded from substituted work. Process-oriented assessments—drafts, reflections, annotated bibliographies—provide windows into student thinking that final products alone cannot offer. When instructors evaluate the journey rather than merely the destination, they create incentives for genuine engagement with AI tools. The study recommends portfolio-based assessment as a mechanism for tracking agency development over time.

Prompt Engineering as Cognitive Exercise

Reframing prompt writing as a cognitive exercise transforms ChatGPT interaction from passive consumption into active learning. Crafting effective prompts requires students to articulate their communicative intentions, specify audience considerations, and identify rhetorical constraints—all higher-order thinking skills.

The study's participants who received prompt-engineering instruction demonstrated improved ability to analyze writing tasks independently, suggesting transferable metacognitive benefits.

Iterative refinement represents another cognitive opportunity. Rather than accepting the first AI response, students should be trained to critique, revise, and resubmit prompts based on output quality. This iterative loop mirrors the revision process essential to academic writing.

Each cycle requires the learner to evaluate whether the AI's output meets their communicative goals, identify gaps, and formulate more precise instructions. This practice strengthens the evaluative judgment that underlies authorial agency.

Comparative analysis of multiple AI responses adds another dimension to cognitive engagement. When students generate several alternative versions and analyze their relative merits, they exercise critical thinking about language choices. Why does one phrasing sound more formal? How does word order affect emphasis?

These questions, prompted by AI output comparison, deepen linguistic awareness in ways that passive acceptance never could. The study documents significant gains in metalinguistic knowledge among students trained in this approach.

However, prompt engineering must not become another form of outsourcing. Students who simply learn to generate better prompts still rely on AI for the actual writing. The cognitive benefit emerges only when prompt crafting is paired with substantive engagement with the output—evaluation, selection, modification, and integration.

The study emphasizes that the learner must remain the final author, making conscious choices about what to accept, reject, or transform from AI suggestions.

Institutional Policies and Academic Integrity Frameworks

Universities and language programs face the urgent challenge of developing coherent policies that acknowledge AI's presence while protecting academic integrity. The study's findings suggest that blanket prohibitions are both unenforceable and pedagogically counterproductive.

Students will use available tools regardless of policy; the question is whether institutions provide guidance for responsible use or leave learners to navigate ethical gray zones alone. Proactive policy development emerges as a critical institutional responsibility.

Disclosure requirements represent one promising policy direction. When students must explicitly acknowledge AI assistance and describe its nature, they engage in metacognitive reflection about their writing process. This requirement also creates accountability structures that discourage wholesale delegation.

The study found that disclosure mandates, when combined with educational programming about appropriate use, reduced problematic AI reliance without eliminating beneficial applications.

Assignment design must evolve to emphasize process over product. In-class writing assessments, handwritten drafts, and oral defenses of written work provide authentic measures of student capability that AI cannot easily replicate. These assessment methods also communicate institutional values about the importance of independent thinking.

The study recommends that programs integrate such assessments strategically, particularly at key developmental milestones in students' academic trajectories.

Faculty development constitutes another essential pillar of institutional response. Instructors cannot guide students toward productive AI use if they lack understanding of the tools themselves. Professional development programs that familiarize faculty with ChatGPT's capabilities and limitations, while providing pedagogical strategies for integration, enable more effective classroom guidance. The study's institutional case studies demonstrate that faculty buy-in significantly influences student adoption patterns.

Practical Coaching Protocols: Transforming ChatGPT from Crutch to Catalyst

Establishing explicit usage rules transforms ChatGPT from an ambiguous threat into a structured learning resource. The study's most successful interventions shared common features: clear guidelines about what constitutes acceptable assistance, transparent communication about expectations, and progressive release of responsibility over time.

These protocols acknowledge AI's utility while establishing boundaries that protect learner agency. The following framework synthesizes evidence-based practices from the longitudinal research.

Rule-based engagement begins with categorizing writing tasks by cognitive demand. Mechanical tasks—grammar checking, punctuation correction, basic vocabulary lookup—warrant unrestricted AI use because they address surface-level errors without compromising authorial thinking. Strategic tasks—thesis development, argument structuring, evidence selection—require structured AI engagement that maintains learner decision-making. Creative tasks—original ideas, personal voice, novel connections—should remain entirely AI-free to preserve authentic authorship.

The progressive release model mirrors established scaffolding theory. Early in a course, students may receive substantial AI support as they learn task requirements and genre conventions. As competence develops, support withdraws systematically. Final assignments require unassisted writing to demonstrate mastery.

This trajectory communicates that AI serves as a temporary bridge to independence, not a permanent replacement for skill. The study's data confirms that students internalize this expectation when it is explicitly articulated.

Feedback loops constitute the final component of effective coaching protocols. Students should document their AI interactions, noting what they requested, why they made those requests, and how they evaluated responses. This documentation creates accountability and provides instructors with insight into student thinking processes.

Regular conferences where students explain their AI usage patterns enable early intervention when problematic reliance emerges. The study found that such reflective practices significantly reduced dependency trajectories.

Designing AI-Resilient Writing Assignments

Assignment design represents the frontline defense against problematic AI reliance. Traditional prompts that ask students to produce generic essays on broad topics invite AI substitution because the cognitive demands are predictable and formulaic. The study recommends redesigning assignments to require personal engagement, contextual knowledge, and original synthesis—elements that resist algorithmic reproduction. Authentic tasks grounded in students' experiences and interests naturally protect learner agency.

Process documentation requirements create visibility into student writing journeys. When assignments mandate submission of outlines, drafts, revision notes, and reflective memos, instructors gain insight into whether AI assistance supported or replaced student thinking. These artifacts also serve pedagogical purposes, helping students recognize their own development over time.

The study's participants who maintained writing portfolios demonstrated stronger metacognitive awareness than those who submitted only final products.

Oral components provide another layer of authenticity verification. When students must present and defend their written work verbally, they cannot hide behind AI-generated prose. The study found that oral defenses revealed significant discrepancies between written quality and verbal articulation among heavy AI users, exposing gaps that written assessment alone missed. Integrating such components into assessment structures creates powerful incentives for genuine engagement with writing tasks.

Collaborative writing assignments offer additional protection against AI substitution while developing valuable skills. When students write together, they must articulate their thinking to peers, negotiate meaning, and justify choices—all processes that externalize cognitive engagement.

The study documented that collaborative writing groups developed healthier AI usage patterns than individual writers, partly because peer accountability created social pressure against delegation.

Teaching AI Literacy as a Core Language Skill

AI literacy must be integrated into language curricula as a fundamental competency rather than treated as an external concern. Students need explicit instruction about how generative AI works, what it can and cannot do, and how to evaluate its outputs critically.

This knowledge demystifies the technology and positions students as informed users rather than passive consumers. The study's educational interventions demonstrated that AI literacy training significantly improved students' ability to use tools strategically.

Critical evaluation skills form the cornerstone of AI literacy. Students must learn to assess AI-generated text for accuracy, appropriateness, and stylistic quality. This requires developing criteria for evaluation and practicing application of those criteria across diverse writing contexts.

The study found that students who received evaluation training became more discerning users, rejecting inappropriate AI suggestions more frequently and engaging more deeply with acceptable ones.

Understanding AI limitations prevents unrealistic expectations and inappropriate reliance. Students should learn that language models can produce fluent but factually incorrect text, that they lack genuine understanding of context, and that their stylistic preferences may not align with academic conventions.

This knowledge enables students to approach AI outputs with appropriate skepticism and to verify information independently. The study documented that realistic expectations correlated with healthier usage patterns.

Ethical reasoning about AI use requires structured discussion and case-based learning. Students benefit from examining scenarios that raise questions about authorship, originality, and academic integrity, then articulating and defending their positions. These discussions develop moral reasoning capacities that guide behavior when policies are ambiguous.

The study found that students who had engaged in such ethical deliberations made more thoughtful choices about AI use than those who had not.

Monitoring Progress and Intervening Early

Early identification of problematic AI reliance enables timely intervention before dependency becomes entrenched. The study identified several warning indicators: sudden improvements in writing quality inconsistent with demonstrated ability, uniform stylistic patterns across assignments, and inability to discuss writing processes in detail.

Instructors who monitor these indicators can initiate conversations with students about their AI usage and provide alternative support structures.

Diagnostic assessments that measure unassisted writing ability provide baseline data for tracking development. Administered at course beginning, midpoint, and conclusion, these assessments reveal whether students are actually developing skills or merely becoming more adept at using AI.

The study's longitudinal design demonstrated that such assessments effectively distinguished genuine improvement from apparent progress masked by AI assistance.

Individualized support plans address the diverse needs of students who struggle with AI boundaries. Some students over-rely on AI because of language anxiety; others because of time pressure; still others because of uncertainty about expectations. Understanding the underlying motivation enables targeted intervention.

The study found that addressing root causes—whether through confidence building, time management support, or expectation clarification—proved more effective than generic warnings about academic integrity.

Celebrating unassisted achievement reinforces the value of independent work. When students produce quality writing without AI assistance, that accomplishment deserves recognition. The study recommends that programs highlight such achievements through publication opportunities, awards, or public acknowledgment. Positive reinforcement for independent authorship creates aspirational models that counterbalance the allure of AI shortcuts.

Building Sustainable Habits for Lifelong Learning

The ultimate goal of AI coaching protocols extends beyond individual assignments to the development of sustainable lifelong learning habits. Students who internalize healthy AI usage patterns will carry those habits into graduate study, professional careers, and personal writing endeavors.

The study's longitudinal design revealed that habits established during ESL instruction persisted years later, shaping participants' relationships with emerging technologies.

Transferable metacognitive strategies—planning, monitoring, evaluating—serve learners across all contexts. When students learn to apply these strategies to AI interactions, they develop generalizable self-regulation skills that enhance all learning. The study found that participants who practiced metacognitive approaches with AI demonstrated improved performance in unrelated academic tasks, suggesting broad cognitive benefits beyond writing specifically.

Adaptive expertise enables learners to navigate technological change confidently. The specific tools available will continue evolving, but the underlying principles of responsible use remain constant. Students who understand how to evaluate new tools, integrate them thoughtfully, and maintain their own agency will thrive regardless of what technologies emerge. The study emphasizes that teaching principles rather than specific tool skills provides durable value.

Community building around responsible AI use creates supportive environments for sustained practice. When students share strategies, discuss challenges, and hold each other accountable, they reinforce healthy norms. The study documented that peer learning communities significantly enhanced the effectiveness of institutional AI policies, suggesting that social reinforcement amplifies formal guidance.

Evidence-Based Recommendations for Educators and Learners

The longitudinal study's findings translate into concrete recommendations for educational practice. For educators, the priority lies in designing assignments and assessments that protect learner agency while acknowledging AI's presence. For learners, the challenge involves developing self-regulatory strategies that leverage AI's benefits without surrendering intellectual ownership. The following recommendations synthesize the study's evidence into actionable guidance for both constituencies.

Institutional leadership must articulate clear values about the role of AI in education. These values should emphasize human agency, critical thinking, and authentic authorship while acknowledging the legitimate uses of generative tools. When institutional messaging is consistent and values-driven, it provides a framework within which instructors and students can make principled decisions.

The study found that institutions with explicit AI values statements experienced fewer integrity violations and more productive AI integration.

Curriculum integration should occur across courses rather than in isolated workshops. AI literacy, like writing proficiency, develops through repeated practice across diverse contexts. The study recommends embedding AI discussions and activities throughout the curriculum, ensuring that students encounter consistent messages and progressively sophisticated guidance. This approach prevents the fragmentation that occurs when AI education is siloed in single sessions.

Assessment reform must accompany pedagogical innovation. Traditional assessment methods that emphasize final products create incentives for AI substitution. The study recommends shifting toward process-oriented assessment that values drafts, reflection, and revision. This shift requires institutional commitment and faculty development, but the payoff in terms of authentic learning outcomes justifies the investment.

Strategic Frameworks for Classroom Implementation

Classroom implementation begins with transparent communication about AI policies and expectations. Students need to understand not only what is prohibited but also what is encouraged and why. The study found that students responded more positively to policies that explained pedagogical rationale than to simple prohibitions.

When students understand that AI restrictions protect their own development, they become willing partners rather than resistant rule-breakers.

Modeling appropriate AI use demonstrates rather than merely describes desired behaviors. When instructors demonstrate how they might use ChatGPT to brainstorm, critique, or revise their own writing, students observe productive patterns in action.

The study found that instructor modeling significantly influenced student adoption of healthy AI practices, suggesting that example carries more weight than exhortation.

Structured practice opportunities allow students to develop AI skills in low-stakes environments. Before students are expected to navigate AI use independently in high-stakes assignments, they should practice with feedback and support. The study recommends progressive practice activities that build from simple to complex AI interactions, each accompanied by reflection and discussion.

Peer feedback mechanisms extend learning beyond instructor-student interactions. When students review each other's AI usage patterns and provide constructive feedback, they develop evaluative skills while reinforcing community norms. The study documented that peer feedback enhanced students' ability to make judicious decisions about AI assistance, complementing rather than replacing instructor guidance.

Learner Self-Regulation Strategies

Students must develop personal guidelines for AI use that align with their learning goals. These guidelines should specify when AI assistance is appropriate, what types of assistance are acceptable, and how AI outputs will be evaluated and integrated.

The study found that students who articulated personal policies demonstrated more consistent, principled AI use than those who operated without explicit guidelines.

Self-monitoring practices enable students to track their own AI usage patterns. Simple documentation—recording when AI was used, what was requested, and how outputs were incorporated—creates awareness that supports intentional decision-making.

The study found that students who maintained such records became more deliberate about their AI use and better able to articulate their writing processes.

Deliberate practice of unassisted writing maintains and strengthens independent skills. Even as students learn to leverage AI effectively, they must continue exercising their own writing muscles. The study recommends that students regularly produce writing without any AI assistance, ensuring that their independent capabilities remain robust. This practice also provides baseline data for tracking genuine development.

Seeking feedback from instructors and peers about AI usage patterns provides external perspective that self-monitoring alone cannot offer. Students may not recognize problematic patterns in their own behavior; others can identify them.

The study found that students who actively sought feedback about their AI use developed healthier patterns than those who relied solely on self-assessment.

Future Directions and Emerging Research

The longitudinal study opens avenues for future research that can refine understanding of AI's role in language education. Longer-term studies tracking students beyond their ESL programs would reveal whether healthy AI habits persist into advanced academic and professional contexts. The current study's timeframe, while substantial, cannot capture the full trajectory of learner development.

Comparative research across educational contexts would illuminate how institutional, cultural, and linguistic factors shape AI integration patterns. The current study focused on specific populations; broader investigation would enhance generalizability. The study's authors explicitly call for cross-cultural research to understand how different educational traditions influence AI adoption and agency preservation.

Intervention effectiveness research would identify which coaching protocols produce the strongest outcomes. While the current study documents the value of structured guidance, it does not systematically compare different intervention approaches. Future research should isolate specific protocol components to determine which elements drive positive results.

Technological evolution will continue reshaping the landscape. As AI tools become more sophisticated, new challenges and opportunities will emerge. The study's framework—emphasizing learner agency, structured engagement, and progressive release—provides principles that can adapt to technological change. Ongoing research must track these developments and refine recommendations accordingly.

Conclusion: The Authorial Imperative

The longitudinal evidence is unambiguous: ChatGPT and similar tools can either enhance or erode ESL learners' academic writing development, depending entirely on how they are deployed. The determining factor is learner agency—the preservation of the student's role as the primary decision-maker and author.

When AI serves as a coach that provides feedback, alternatives, and scaffolding, it accelerates development. When it serves as a substitute that produces text, it stunts growth.

Educational institutions bear responsibility for creating conditions that support healthy AI integration. This requires policy development, curriculum redesign, faculty training, and assessment reform—substantial investments that many institutions have yet to make. The study's findings provide compelling evidence that such investments yield significant returns in terms of student learning outcomes and academic integrity.

Students, for their part, must embrace the uncomfortable work of thinking. The temptation to outsource cognitive labor to machines will only intensify as AI capabilities expand. Resisting that temptation requires deliberate practice, self-awareness, and commitment to one's own development.

The study's participants who maintained their agency despite AI availability demonstrated that this resistance is possible and rewarding.

The future of ESL education will inevitably involve generative AI. The question is whether that involvement will produce empowered writers or dependent consumers. This study's answer is clear: the outcome depends on choices made by institutions, educators, and learners.

By establishing rules that protect agency, providing structured coaching protocols, and maintaining commitment to authentic authorship, we can ensure that AI serves human development rather than replacing it.