Rigor, commitment, and generative AI: a conceptual framework for student engagement in EAP writing

Synthesis: Al Nafjan argues that whether generative AI strengthens or hollows out an EAP writing course depends less on the tool than on two features of the classroom: how demanding the writing tasks are, and how committed the teacher is to students' learning. Crossing those dimensions yields four configurations — Neglectful, Indulgent, Punitive, and Rigor with Support — linked by explicit propositions to four engagement profiles. On this account GenAI does not make teacher involvement less necessary. It makes it more necessary, because fluent prose stops being evidence that thinking happened.

Conceptual paper: the author synthesizes prior theory and empirical work into a framework and a set of testable propositions. No data were collected or analyzed, so the propositions are expectations awaiting empirical test. Every figure below is a count of the framework's own components or an attributed claim about another study, not a finding of this one.

What the framework argues

The paper starts from a split in the literature rather than from a gap in evidence. Work on GenAI in writing instruction tends to argue either for its affordances or against its risks, and both sides keep their eyes on the technology or on student conduct. Al Nafjan's complaint, following Jiang and Hyland (2026), is that this leaves the instructional conditions under which particular kinds of AI use emerge largely unexamined. The problem is sharp in EAP because GenAI can produce acceptable academic prose while reducing the visibility of the cognitive, rhetorical, and epistemic work that prose is supposed to represent — a decoupling of performance from learning.

Her answer is the Rigor–Commitment Alignment Framework (RCAF), a two-dimensional model of the pedagogical environment. Academic rigor is the extent to which tasks require sustained cognitive effort and disciplinary reasoning. It is operationalized through three subdimensions. Cognitive demand is the degree to which tasks require higher-order processes beyond recall (Webb, 2002). Epistemic authenticity is the degree to which tasks approximate disciplinary practice rather than decontextualized school activity (Lea and Street, 2006). Performance accountability is the requirement that students show their own reasoning through visible drafting and revision (Bearman et al., 2024).

Teacher commitment is defined narrowly and deliberately. Al Nafjan sets aside the organizational and occupational senses of the term and means commitment to students' learning, enacted through instructional responsiveness, relational support, and critical scaffolding. She treats it as a task- and context-specific pedagogical presence rather than a stable classification of a teacher. A teacher may be deeply committed to the profession and still enact support unevenly across tasks. That distinction matters for the AI argument: a model can supply assistance, but it cannot supply situated attention, professional judgment, or sustained responsibility for a student's development.

The framework's central claim is interaction rather than addition. Rigor sets the demands students meet; commitment shapes how they meet them. Proposition 1 therefore predicts that rigor will be more strongly associated with deep-agentic engagement when commitment is high, and more strongly associated with strategic-defensive use or disengagement when commitment is low.

The architecture comes from Bandura's (1986) triadic reciprocal causation. Rigor and commitment jointly make up the environmental vertex; goal orientation, Self-Efficacy, proficiency, and GenAI literacy make up personal factors; drafting, delegating, verifying, and avoiding make up behavior. The relationships run in both directions, and Proposition 4 formalizes the loop: students' engagement and AI-use practices give teachers signals that may prompt them to raise process accountability, tighten surveillance, or withdraw support.

What the four configurations predict

The intersection of the two dimensions gives four positions, borrowed in structure from Baumrind's (1991) demandingness-by-responsiveness account of parenting and from Aelterman et al.'s (2019) circumplex model of motivating teaching. In the Neglectful configuration both dimensions are weak, and the paper's illustration is a literature-review paragraph assigned with no models, no interim checkpoints, and feedback limited to a grade — conditions in which a student can paste the prompt into a chatbot, lightly edit the output, and lose nothing.

Indulgent environments pair weak rigor with strong commitment. Warmth is present, but nothing obliges the student to evaluate, revise, or justify what the tool produced, so support without demand settles into surface engagement. Punitive environments invert this: high standards with thin support, where AI is encountered mainly through detection and penalty. Al Nafjan's reading is that this pushes students toward concealment rather than responsible use, teaching them to evade scrutiny instead of to write.

Rigor with Support combines high demand with high support and carries the greatest predicted probability of deep-agentic engagement. GenAI there is used to generate alternatives, test clarity, probe counterarguments, and check disciplinary appropriateness, with the teacher's role expanding rather than shrinking because judging when output is useful, insufficient, or misleading is itself evaluative judgment.

The four engagement profiles — deep-agentic ("Driver"), strategic ("Optimizer"), surface ("Glider"), and disengaged or withdrawn — are built from the same four features: motivational purpose, cognitive investment, form of agency, and orientation toward GenAI. Al Nafjan is explicit that these are prototypes, not learner types, and that the propositions describe shifts in relative probability rather than one-to-one correspondences.

Why GenAI amplifies the differences

The framework's load-bearing proposition is the third one. Al Nafjan argues that differences among configurations will be larger when GenAI is readily available, when assessment rewards the final product, and when composing processes stay invisible. Earlier writing technologies touched wording, grammar, and fluency. GenAI reaches planning, reasoning, source integration, and revision, so a student can hand over the parts of composing that used to leave traces.

She cites empirical work to support the surrounding claims rather than the framework itself. Farrokhnia et al. (2026) reportedly found that AI-generated feedback can be judged technically superior while producing no better revision outcomes, which points to learner uptake, the concern of feedback literacy research, as the mediating variable. Lin and Crosthwaite (2024) reportedly found GPT-assisted corrective feedback insensitive to proficiency and task context in ways teacher feedback is not. Zhou et al. (2026) reportedly showed that student engagement predicts later changes in teacher practice, and that controlling practices predict later disengagement — the empirical warrant for the reciprocal loop.

The framework also draws a line between strategic and surface engagement that is easy to blur. Both reduce effort, but strategic use is deliberate and responsive to assessment criteria, while surface use is passive compliance with uncritical reliance on output. The distinction matters for academic integrity policy, because only one of the two looks like a violation from the outside.

Al Nafjan concedes that crossing two pedagogical dimensions is not new. The claimed contribution is the mechanism: a demand-support model built for ordinary instruction does not explain why identical tool access supports critical evaluation in one classroom and cognitive substitution in another. GenAI enters as a contextual affordance that changes the pathways between effort and performance, not as an independent axis or a cause.

What this means for practice

• Instructors. Diagnose before adopting. The framework asks which dimension is weak in a given task, so a warm course with undemanding writing needs higher cognitive demand rather than more encouragement, and a demanding course with thin support needs conferences and scaffolding rather than lower standards.
• Writing program designers. Make process visible. Because the amplification argument turns on assessment design, requiring drafts, revision memos, and justification of source choices does more to preserve the link between text and thinking than any rule about tool use.
• Teacher educators. Use the configurations as case-analysis material. The labels are offered as practitioner shorthand for professional discussion, not as a way to sort colleagues into types.
• Administrators. Write policy that is pedagogically contingent. A blanket prohibition in a Punitive environment may intensify covert use, while the same rule in a Rigor with Support environment may forbid a productive practice; institutional guidance should specify the conditions under which AI use is likely to support learning.

Limitations

• The framework is conceptual and unvalidated. Al Nafjan states that it is propositional, that its diagnostic value is heuristic, and that using it as an assessment or professional development instrument requires operational and empirical testing.
• Both dimensions are internally complex. Rigor bundles three subdimensions and commitment three more, and the paper does not establish that they vary independently or that the subdimensions carry equal weight.
• The engagement profiles are prototypes rather than mutually exclusive or exhaustive categories, and their empirical distinctiveness remains untested.
• The framework is built for English-medium higher education EAP writing, and it foregrounds teacher-influenced conditions. Institutional assessment regimes, disciplinary conventions, AI access and policy, socioeconomic differences, and students' prior proficiency may constrain or moderate what teachers can enact, and secondary, community-based ESL, and low-access EFL settings stay outside its stated scope.
• Rigor with Support is not a claim that more demand is always better. Excessively demanding tasks may remain counterproductive despite support, and the same level of rigor can be experienced differently by different students.
• GenAI tools are changing quickly, so the specific AI practices attached to each profile may shift even if the underlying logic holds.

Citation

Al Nafjan, E. (2026). Rigor, commitment, and generative AI: a conceptual framework for student engagement in EAP writing. Frontiers in Education, 11, 1944529.