The Active Learning–GenAI Synergy Framework: Ethical Integration of Generative AI in EFL/ESL Writing in Resource-Constrained Contexts

1.1 Significance of the study

This research makes key contributions to AI-enhanced active learning in EFL/ESL contexts. Theoretically, it develops an Active Learning–GenAI Synergy Framework that positions AI tools as cognitive partners, contributing to theories of human–AI collaboration (Luckin, 2017; Holmes et al., 2019). Pedagogically, it demonstrates how ethically integrated AI can shift writing instruction from teacher-centered to dynamic active learning, offering evidence for student-centered approaches in resource-limited contexts (Prince, 2004; Freeman et al., 2014). Practically, it provides scalable solutions for large EFL/ESL writing classes in Pakistani higher education, informing curriculum development, instructor training, and AI integration policies.

2.1 Active learning theory: Foundations and principles

The active learning process is inherently a challenge to the traditional transmission model, which inculcates a student as an active knowledge builder, as opposed to a receiver of the knowledge. The working definition of student-centered learning by Bonwell and Eison (1991) refers to any instructional approach that involves students in the process of learning, stressing that students should do more than merely being asked to listen; they need to read, write, talk, or be asked to solve problems. The idea of this definition focused on interaction, participation, and thinking, not on the consumption of content.

Prince (2004) extended this framework by emphasizing the metacognitive dimension, noting that active learning requires students to engage in “meaningful learning activities and think about what they are doing” (p. 223). This metacognitive emphasis becomes particularly relevant in AI-mediated contexts where tools can facilitate reflective practice through feedback and revision cycles.

Freeman et al.’s (2014) comprehensive meta-analysis of 225 studies provided compelling empirical support, revealing that active learning increased student performance by 6% and reduced failure rates by 36% compared to traditional lecturing. Their conclusion that active learning represents “the preferred, evidence-based choice for instructors” (p. 8410) has been replicated across disciplines, establishing active learning’s superiority over instructor-centered methods and providing the theoretical foundation for AI-enhanced pedagogical approaches.

2.2 Theoretical frameworks of active learning

Active learning theory draws from three foundational frameworks that inform AI integration in educational contexts. Social constructivism emphasizes Vygotsky’s (1978) Zone of Proximal Development (ZPD), where AI tools function as “more knowledgeable others” providing scaffolding through immediate feedback and collaborative dialogue. Johnson and Johnson (2009) extend this through collaborative learning theory, demonstrating that learning is enhanced when students work toward shared goals—processes readily facilitated through AI-mediated peer collaboration.

The main theory of AIs-enhanced active learning cycles is experiential learning theory. The four-step model offered by Kolb (1984), concrete experience, reflective observation, abstract conceptualization, and active experimentation, corresponds directly to collaborations with AI-writing: students take concrete steps in writing with AI assistance, contemplate the results produced by AIs, theoretically abstract writing concepts based on the suggestions provided by the AI, and implement experimental approaches to revision.

2.3 Active learning in writing instruction

Collaborative and back and forth writing pedagogy inherently lends itself to active learning. In their meta-analysis, Graham and Perin (2007) also found that active learning strategies that show significant improvement in writing are collaborative writing (effect size = 0.75), strategy instruction (effect size = 0.82), and peer assistance (effect size = 0.89). These results indicate the usefulness of AI integration, which can provide more teamwork experiences and direction.

Flower and Hayes (1981) redefined composition as a cycle of planning, translating, and reviewing-with metacognitive activity and reflection on revision. AI-based solutions can facilitate every step with idea generation, vocabulary proposals, and automatic feedback. Bereiter and Scardamalia (1987) differentiate between the knowledge-transforming process of writing, which presupposes an active treatment of content and development of arguments, and the knowledge-telling paradigm in which AI can become an aid to critical reflection and perspective analysis.

2.4 AI applications in writing instruction: Evolution and effectiveness

As writing applications have developed beyond rudimentary grammar checkers, this has also meant a shift toward active learning pedagogies and a greater emphasis on student agency and collaboration (Nazari et al., 2021; Imran & Almusharraf, 2024).

2.4.1 Early systems and feedback tools

Early experiments were on automated feedback and grammatical markup. Purcell et al. (2013) discovered that Google Docs supported enhanced writing with the help of scaffolding and collaborative writing features. Nonetheless, tools demanded organized instructional schemes that supported critical participation, which was the core of active learning theory (Nobles & Paganucci, 2015).

2.4.2 Advanced evaluation systems

Sophisticated automated writing evaluation (AWE) technology was an important step forward. Fitria (2021) has documented that AI-based applications such as Grammarly fostered metacognitive awareness and reflective practice through facilitating critical self-assessment. Chang et al. (2021) reported statistically significant gains when AI tools gave instantaneous, domain-specific feedback in addition to increasing the independence of the learners-commonly referred to as active learning.

2.4.3 Contemporary generative AI integration

Recent generative AI research provides new active-learning opportunities. Su et al. (2023) showed how argumentative writing is done through collaboration of students with the AI partner, ChatGPT, as a cognitive partner to support collaborative learning and knowledge construction. Song and Song (2023) identified increases in coherence, order, and drive when AI was incorporated into structured frameworks, with a call to employment in moderation to encourage agency over reliance. Huang et al. (2023) contrasted AI-aided and conventional settings. Under the AI setting, where students received personalized prompts and guidance to enhance revision tasks, academic performance was more effective, and engagement was higher than in the traditional setting.

2.5 Ethical AI integration and active learning

The combination of AI technology and active learning pedagogy proposes significant ethical issues that need to be discussed to ensure that the educational outcomes of the integration of technology are productive, not destructive. Active learning theory is based on the principles of student choice, critical thinking, and genuine engagement, which may be negatively affected by the misuse of AI tools.

Regarding ethical use of AI in higher education, policy guidelines were created by UNESCO (2021) and the Higher Education Commission of Pakistan (2023), which were outlined not only as essentials to preserve academic integrity but also to take advantage of the technological affordances. These rules reflect the active learning theory, encouraging responsibility, critical thinking, and reflection of the students engaged in learning conditions provided by an AI system.

In the case of the ethical application of AI in higher education, the Higher Education Commission of Pakistan developed institutional policies (2023) that were described as necessary to ensure the absence of academic dishonesty and to utilize technological capabilities. These instructions are consistent with active learning theory and foster responsibility, critical thinking, and reflection among students in the learning conditions an AI system presents.

The framework is in line with the international standards: educational AI is classified as a high-risk area that needs transparency, human control, and risk management (EU AI Act, 2024); the recommendations on generative AI in education and research (2023) by UNESCO focus on critical digital literacy, academic integrity, and fair access; the Safe AI in Education Manifesto includes the focus on preserving human agency and avoiding dependency (UNESCO, 2023). These convergent principles were used in designing protocols.

In the case of the ethical application of AI in higher education, the Higher Education Commission of Pakistan developed institutional policies (2023) that were described as necessary to ensure the absence of academic dishonesty and to utilize technological capabilities. These instructions are consistent with active learning theory and foster responsibility, critical thinking, and reflection among students in the learning conditions an AI system presents.

2.6 Research gaps and theoretical synthesis

Although research on AI in writing instruction is growing, there are noticeable gaps in how generative AI tools can be ethically integrated into active learning approaches, especially in EFL/ESL settings where resources are limited. The current state of research on AI effectiveness focuses on pedagogical curriculum, supporting student agency, critical thinking, and group learning insufficiently. Most studies have been carried out in high-resource learning environments and do not provide many insights into how AI tools can tackle systemic issues like excessive classroom sizes, poor quality teacher feedback, and the availability of resources to provide personalized learning. The present study fills these gaps by exploring how ethically scaffolded generative AI tools can be used to promote active learning in EFL/ESL writing instruction.

3.1 Research design and theoretical alignment

The pragmatic mixed-methods action research design is employed by combining quantitative performance measures and a qualitative phenomenological evaluation through which the full comprehension of AI-mediated active learning experiences was understood. The action research framework’s emphasis on cyclical reflection, pedagogical innovation, and collaborative inquiry aligns with both active learning theory and ethical AI integration practices (Stringer, 2014; McNiff & Whitehead, 2011).

The design entails pre-test/post-test experimental and control group comparisons supplemented with semi-structured interviews, where methodological triangulation explored the relationship between the AI-enhanced active learning performance (quantitative measure) and the quality of student engagement coupled with ethical awareness development (qualitative measure). Action research aspect facilitated ongoing pedagogical upgrades over the course of intervention (15 weeks) and made real-time alterations to AI integration plans through student feedback- in keeping with the iterative, reflective qualities of active learning and ethical AI usage.

3.2 Theoretical framework integration: Active learning theory and Kolb’s experiential learning cycle

The methodological design is grounded in Kolb’s Experiential Learning Theory, which posits that “learning is a continuous, cyclical process involving four key stages: Concrete Experience, Reflective Observation, Abstract Conceptualization, and Active Experimentation” (Kolb, 1984). This theory states that “learning is the process whereby knowledge is created through the transformation of experience. Knowledge results from the combination of grasping and transforming experience”.

The AI-enhanced writing tasks were systematically designed to facilitate each stage of Kolb’s experiential learning cycle:

Active learning principles integration

The methodology incorporated core active learning principles identified by Prince (2004) and Freeman et al. (2014):

Technology Acceptance Model (TAM) Integration

Furthermore, Davis’s (1989) Technology Acceptance Model provided the theoretical framework (See Appendix 5) for understanding student acceptance and utilization of AI tools within active learning contexts. The study investigated four key TAM constructs within the context of cognitive partnerships:

3.3 Participants and sampling strategy

The study employed convenience sampling to recruit 148 undergraduate students from Bahauddin Zakariya University, Multan, Pakistan. Participants were naturally occurring intact classes from two academic programs: the experimental group (Digital Marketing students) participated in AI-mediated active learning experiences, while the control group (BBA students) received conventional textbook-based writing instruction, ensuring ecological validity while maintaining practical feasibility within the resource-constrained context. The demographics of participants are given in Table 1.

The quasi-experimental design enabled receiving an opportunity to compare AI-mediated active learning to a conventional pedagogy based on teacher-centered education without any harm to ethics in the context of education equity. However, all results are of an exploratory nature due to the core confounded comparison because experimental and control groups are intact classes from different degree programmes, threatening internal validity. Pre-intervention assessment confirmed group comparability. Mann-Whitney U tests showed no significant differences in pre-test writing scores (Experimental: M rank = 72.45, SD = 8.32; Control: M_rank = 76.55, SD = 8.67; U = 2,561, p = .532). Both programs required equivalent English prerequisites (IELTS 5.5), with no differences in prior English grades (t = 0.58, p = .564). Demographics were comparable: gender distribution (χ² = 1.89, p = .169) and age (t = 0.73, p = .468). Both groups received instruction from qualified faculty (MA/MPhil in English, 6-8 years of experience) using standardized assessment rubrics, ensuring instructional consistency despite intact class design.

3.4 Instructional intervention design: AI-mediated active learning framework

The design of the instructional intervention in the experimental group involved an Active Learning-GenAI Synergy Framework that presented the AI tools as learning partners in collaborative opportunities. The three important roles that AI played under this intervention transformed conventional writing instruction in the following ways:

Weekly active learning structure

Each week followed a structured active learning cycle that integrated AI collaboration with experiential learning principles elaborated in Table 2.

The instructional intervention spanned 15 weeks (See Appendix 2), from January 2025 to June 2025. The experimental group engaged with various AI tools, including ChatGPT, Claude AI, Meta AI, and Canva, while the control group continued with textbook-driven lessons focused on grammar drills, sample formats, and teacher explanations.

Each week in the experimental group was dedicated to a specific writing genre or task, with AI tools introduced as assistive scaffolds. These tools were not intended to replace human effort, but to guide students through revision, coherence building, vocabulary selection, and formatting tasks listed in Table 3.

Weekly Sequence: (1) Genre introduction (15 min); (2) Prompt demonstration (10 min); (3) AI-assisted writing (60 min); (4) Strategy sharing (20 min); (5) Ethical reflection (20 min); (6) Submission with AI disclosure.

3.5 Ethical integration protocol

The study implemented a comprehensive Ethical Integration Protocol designed to ensure that AI collaboration promoted rather than undermined active learning principles:

This paper followed an elaborate Ethical Integration Protocol in accordance with the EU AI Act (European Commission, 2024), UNESCO principles (UNESCO, 2023), and Safe AI principles:

3.6 Data collection methods and instruments

Quantitative data collection

Ethical Awareness: Coded reflection quality (1 = descriptive to 4 = metacognitive), AI disclosure accuracy, interview analysis. Progression: M = 1.8 (Week 1) to M = 3.4 (Week 15). No missing data. See complete rubric (Appendices-3).

Weekly Writing Performance Tracking: Six intermediate writing assessments provided a timeline on how students improved over the course of the intervention as well as the periods when students were making significant improvements towards the goal.

Qualitative data collection

Semi-Structured Interviews (See Appendix 4) were conducted through purposive sampling with eight participants, showing differing performance levels (3 high, 3 medium, 2 low achievers), gender-balanced (4F, 4M). Individual 30-45 min interviews explored cognitive partnership, ethical reflection, metacognitive development, active learning, and technology acceptance. Audio-recorded, transcribed verbatim, accuracy-checked. See full protocol in the Appendices-4.

3.7 Ethical considerations and safeguards

The study maintained rigorous ethical standards aligned with both research ethics and AI integration ethics by seeking approval from the Bahauddin Zakariya University’s Research Ethical Committee (Ref No. 28/UREC/2024 dated 16-10-2024) in compliance with the Declaration of Helsinki. Written Informed Consent was obtained (See Appendix 1) and all participants received comprehensive information about the study purposes, procedures, and their rights of voluntary participation, confidentiality, and participant anonymity. Educational Equity was ensured for the Control group that received access to AI integration training following study completion. For AI Ethics Integration, clear guidelines are distinguished between AI assistance and academic misconduct. Privacy Protection was ensured as no personal data was input into AI systems. Students disclosed AI use in all academic submissions as Transparency Requirements. Ethical reflection activities promoted responsible AI collaboration for critical engagement.

3.8 Data analysis framework

Analysis was quantitative based on Descriptive Statistics (Mean scores, standard deviations, and performance distributions of all writing assessments) and Inferential Testing (Within-group pre/post comparisons of non-parametric data) using Wilcoxon Signed-Rank Tests and Mann-Whitney U Tests (Between-group comparisons of experimental and control conditions), and Gender Analysis (Separate analyses of the varying effects of all demographic groups). Magnitudes that were given as rank-biserial correlation (r = Z/S N): small 0.30 or less, medium 0.30-0.49, large 0.50 or more. The Type I error was controlled by using Bonferroni correction on intermediate tests (0 = .008).

Six phases used by Braun and Clarke (2006): familiarization, coding (65 initial codes), theme generation, review, definition (five final themes), reporting. Two independent researchers coded 25 percent (κ = 0.82); the differences were resolved by discussion.

4.1 Active learning environment effectiveness: Quantitative outcomes

The quantitative analysis reveals significant evidence supporting the effectiveness of AI-mediated active learning environments compared to conventional teacher-centered instruction. These findings demonstrate how cognitive partnerships with AI tools facilitated measurable improvements in writing performance while promoting the student engagement and metacognitive development characteristic of active learning experiences.

4.1.1 Overall writing performance: Active learning vs. traditional instruction

The comparison between AI-mediated active learning (experimental group) and traditional instruction (control group) provides compelling evidence for the superior effectiveness of cognitive partnership approaches in EFL writing development given in Table 6.

The AI group showed great improvement with a large effect size (Z = -6.325, p < .001, r = 0.68), whereas the traditional instruction showed small improvement (Z = -2.128, p = .033, r = 0.23), indicating that cognitive partnerships led to better writing development.

4.1.2 Progressive skill development through active learning cycles

The analysis of writing performance across six intermediate assessments ( Table 7) reveals how cognitive partnerships with AI tools facilitated progressive skill development through repeated experiential learning cycles, consistent with Kolb’s (1984) theory of continuous learning improvement.

The results show that there was an evident trend of progression in which the benefits of AI-mediated active learning became more prominent with time. There were marked differences at Test 5 (p < .001), so it appears that cognitive partnerships take time before they yield any meaningful benefit. This tendency contributes to the theory of experiential learning provided by Kolb that underlines the idea that learning becomes stronger with the number of cycles of experience, reflection, conceptualization, and experimentation.

Bonferroni correction (= .008) only confirms Tests 5-6 as significant with large effects (r = 0.72, r = 0.58); Tests 2-4 had small non-significant effects (r = 0.15-0.24), showing advantages occurred after 6-8 weeks without multiple testing inflation.

4.1.3 Gender-differentiated active learning engagement

Analysis of gender differences within the AI-mediated active learning environment ( Table 8) reveals differential engagement patterns, providing insights into how cognitive partnerships develop across demographic groups.

Female students in the AI-mediated active learning environment demonstrated 19% higher engagement levels and a more significant improvement (p < .001) compared to male students (p = .010). Qualitative data suggests this difference may reflect greater responsiveness to the reflective observation and metacognitive awareness components of the experiential learning cycle, indicating that female students may engage more deeply with the ethical reflection prompts and collaborative peer activities that characterize active learning environments.

Notably, in the traditional instruction environment, the pattern reversed, with male students showing significant improvement while female students’ gains were not statistically significant ( Table 9). This contrast suggests that AI-mediated active learning environments may be particularly beneficial for female EFL learners, potentially because these environments emphasize collaborative reflection, peer interaction, and metacognitive awareness—learning approaches that align with documented female preferences for collaborative and reflective pedagogies.

4.2 Cognitive partnership development: Qualitative insights

Semi-structured interviews with eight volunteer students from the AI-mediated active learning group provide rich insights into how cognitive partnerships developed and functioned within the experiential learning framework. These qualitative findings illuminate the mechanisms through which AI tools served as cognitive partners, metacognitive mirrors, and equity tools.

4.2.1 AI as cognitive partner: Collaborative knowledge construction

Students consistently described their relationships with AI tools in terms of collaborative partnership rather than tool usage, indicating successful development of the cognitive partnership model. Interview participants emphasized how AI tools supported active knowledge construction through interactive dialogue and collaborative problem-solving.

Student Voice - Collaborative Idea Generation: “AI tools have assisted me a lot in writing. ChatGPT provided me with relevant arguments that helped me to improve my sentence structure, but I had to think critically about which ideas fit my own perspective.”

This response demonstrates how students engaged in the Abstract Conceptualization stage of Kolb’s cycle, using AI input as a foundation for developing their own understanding rather than passively accepting generated content.

Student Voice - Interactive Revision Process: “Claude AI was better for long paragraphs, and it gave me coherent ideas. But I learned to combine its suggestions with my own thoughts to create something that was truly mine.”

This reflection illustrates Active Experimentation, where students tested different approaches to integrating AI suggestions within their own writing development process, demonstrating the experimental and iterative nature of effective active learning.

4.2.2 AI as metacognitive mirror: Reflective awareness development

The weekly ethical reflection prompts successfully promoted metacognitive awareness and reflective observation—key components of both active learning and cognitive partnership development. Students reported enhanced ability to evaluate their own learning processes and AI collaboration strategies.

Student Voice - Metacognitive Development: “The weekly reflection prompts made me think about how I was using AI. I started to notice when I was depending too much on AI suggestions versus when I was using them to enhance my own ideas.”

This response demonstrates the development of metacognitive awareness that enabled students to regulate their own learning and maintain agency within the cognitive partnership. Such self-awareness is essential for effective active learning and prevents the passive dependence that could undermine educational goals.

Student Voice - Ethical Integration Success: “I avoided copying full paragraphs from AI. I learned to rewrite and reflect. The ethical prompts aided me in appreciating the distinction between AI help and AI substitution of my way of thinking.”

This reflection demonstrates how the Ethical Integration Protocol was effective in facilitating responsible teamwork on AI that did not compromise academic integrity, while utilizing technological affordances, as one of the major goals of the active learning framework.

4.2.3 AI as equity tool: Addressing resource constraints

Students highlighted the ways AI tools solved age-old resource constraints in bigger EFL/ESL classes by providing individual learning experiences that would be otherwise unattainable in resource-poor settings.

Student Voice - Personalized Learning Access: “Experiences with Canva to write proposals helped me to figure out how to structure information. Personal attention is something we do not usually get in large classes, but AI tools provided me individual feedback anytime it was necessary.”

This is their answer, which shows the AI tools were equity tools that provided access to individualized instruction and real-time feedback on a massive scale and overcame the systemic barriers that generally limit active learning despite favorable evidence on the value of active learning.

Student Voice - Increased Confidence and Agency: “This has added confidence in writing emails and reports. I feel more confident about working independently, yet still there is support when I need.”

This reflection demonstrates how cognitive partnerships facilitated self-efficacy and learner autonomy as the central by-products of active learning practices that help a student become a lifelong learner and ultimately a professional.

4.3 Technology acceptance within active learning contexts

Qualitative Analysis of interviews through the Technology Acceptance Model (TAM) framework reveals high levels of student acceptance of AI tools within active learning environments, with acceptance levels directly correlated to the degree of active engagement and cognitive partnership development.

4.3.1 Perceived usefulness in active learning contexts

The majority of participants expressed positive views about AI integration within active learning environments, with most of the experimental group students rating AI tools as “very effective” or “extremely effective” for developing language skills through active collaboration rather than passive consumption.

Students’ perceived usefulness was specifically tied to AI tools’ capacity to facilitate active learning processes:

4.3.2 Behavioral intention for continued cognitive partnership

Student Voice - Sustained Engagement: “These tools are useful, but I do not think we need to rely on them fully. They ought to be wisely exploited. I will use them again in my next courses because I think better not less with them.”

This reaction is evidence of high levels of manipulation of cognitive partnership concepts, meaning that students gained the analytical awareness required to commit to persistent ethical AI collaboration under active learning conditions.

The impressive behavioral intention rates (86% expressed intentions to continue using AI), along with the indicators of the critical awareness, allows supposing that the active learning model has been effective in terms of supporting responsible adoption of technology instead of mindless dependence.

4.4 Active learning process indicators

The qualitative analysis demonstrated particular signs of successful development of the effective processes of active learning in the cognitive partnership with AI-based tools:

4.4.1 Experiential learning cycle evidence

Direct Experience: AI drafting, revision, and multi-modal composition were consistently dealt with by students in a hands-on way. The discussions during the week demonstrated an increasingly informed analysis of AI cooperation strategies

Abstract Conceptualization: Increased writing principles and ethical considerations were evidenced by the students as they explained such principles

Active Experimentation: This is where students were observed testing out various strategies and perfecting their cognitive partnership strategies

4.4.2 Collaborative learning enhancement

Students indicated more peer interactions and collaborative problem-solving, which was supported by common AI tools and the reflection activities. The cognitive partnership model promoted rather than inhibited human-to-human collaboration, as students shared strategies, discussed ethical considerations, and engaged in collective reflection on AI integration experiences.

4.4.3 Higher-order thinking development

Observation of weekly discussions and analysis of interview transcripts revealed evidence of critical evaluation, synthesis, and creative application—higher-order thinking skills that characterize effective active learning environments. Students moved beyond surface-level AI interaction to engage in sophisticated analysis of AI outputs, evaluation of ethical considerations, and creative integration of AI suggestions within original compositions.

5.1 AI-mediated active learning: Theoretical validation and performance enhancement

The substantial improvement in the experimental group (Z = -6.325, p < .001) indicates that AI tools facilitated the experiential learning cycle proposed by Kolb (1984) in EFL/ESL writing. Although the quasi-experimental design restricts causal inference, baseline equivalence, progressive patterns, interaction effects, and qualitative triangulation can be used to attribute differences to the pedagogical approach. The progressive improvement, which is observed since Test 5 (p = .001) is consistent with what Kolb proposed that learning will only intensify under repetition of experience, reflection, conceptualization, and experimentation. This builds the meta-analysis by Freeman et al. (2014) that has found the benefits of active learning in all disciplines, verifying its concepts in AI-based EFL settings. The 6-8 weeks development time implies that cognitive partnerships are to be maintained over time, which is consistent with recursive models of learning. The large effect sizes (r = 0.68 overall; r = 0.72 at Test 5) and Bonferroni-adjusted significance will remove any doubts about Type I error inflation and trivial effects. The less impressive results of the control group (Z = -2.128, p = 0.033) support the statement of Prince (2004) that passive instruction is not as effective. Importantly, this paper shows that AI can be used as a facilitator to active learning in the resource-scarce situations where a conventional implementation is restricted. It is supported by the interview evidence: The significance of Test 5-6 is consistent with the qualitative reports of partnership maturation and advanced AI evaluation strategies. Qualitative results describe the transformation of the attitude toward AI as an instrument to the attitude toward it as a partner that justifies the model of cognitive partnership by Luckin (2017). This is opposed to Chang et al. (2021) and Fitria (2021), in which AI was viewed as feedback. Ethical reflections and cycles seem to have made the application of AI a true partnership. Student comments such as I learned to synthesize its recommendations with my own thoughts make evidence of the stage of abstract conceptualization in Kolb. In contrast to the case reported by Yan (2023), who raised dependent issues, the ethical procedure in this case reduced risks at the expense of AI, maintaining its cognitive advantages.

Although the quantitative evidence proves beneficial performance, to learn processes that AI facilitated active learning, it is essential to consider the metacognitive processes disclosed in qualitative data.

5.2 Active learning processes: Evidence and theoretical alignment

Metacognitive awareness, reflected in comments like “I started to notice when I was depending too much on AI,” validates Bonwell and Eison’s (1991) assertion that active learning requires reflection. This distinguishes the study from prior AI writing research. Liu et al. (2021) also noted metacognitive improvements but lacked the systematic ethical framework used here. Structured prompts ensured AI use remained cognitively demanding, aligning with Prince’s (2004) definition of meaningful learning. AI tools enhanced rather than inhibited collaboration. Students reported increased peer interaction through shared AI engagement and reflection tasks, extending Johnson and Johnson’s (2009) collaborative learning theory into AI-mediated contexts. This contrasts with Anderson and Rainie’s (2023) concerns about AI reducing collaboration, as the partnership model positioned AI as a facilitator rather than a replacement.

5.3 Gender-differentiated engagement: Theoretical implications

Female students showed 19% higher engagement and greater improvement (p < .001) than males (p = .010) in AI-mediated active learning. In contrast, in the control group, males improved significantly (p < .001) while females’ gains were insignificant (p = .062). This reversal suggests AI-mediated environments may align better with learning preferences associated with reflective and collaborative engagement.

Qualitative data point to stronger responsiveness among females to reflection and metacognition in Kolb’s cycle. This extends Al Mahmud’s (2023) findings by showing that pedagogical framing mediates gendered outcomes. These results support Vygotsky’s (1978) view that scaffolding must match learner characteristics. While both genders benefited, the framework appears particularly effective for those favoring reflective collaboration. The significant gains for males (p = .010) suggest the approach benefits all learners while amplifying certain preferences.

5.4 Resource constraint mitigation: Theoretical and practical significance

The AI tools were used to overcome the issue of high enrollment in EFL classes, which supported the idea of AI being an equity tool mentioned by Bliksten (2016). Access to individualized instruction was democratized through reports of getting personalized feedback whenever they needed it through cognitive partnerships. This expands Popenici and Kerr (2017) and Renz et al. (2020) and offers solid empirical support that AI could be used to address the obstacles to active learning. The model overcame the conflict between large classes and individual feedback, which is a barrier to active pedagogy traditionally. The findings can substantiate the perspective of Godwin-Jones (2019), according to which technology can facilitate pedagogy but cannot substitute it, as it allows active learning to take place under limitations. According to critics, technology is not the solution to pedagogical problems, but this paper demonstrates that even in challenging situations, AI that is incorporated into grounded structures can support student-centered practices.

5.5 Technology acceptance and active learning integration

The Technology Acceptance Model (TAM) is confirmed by high-perceived usefulness (79.7%) and intention to continue use (86%). Notably, by engaging in pedagogical integration, but not AI as an isolated tool, acceptance was enhanced. Such responses by students as they make me think better than think less are critical evaluations and not passive adoptions and represent the issue that Davis (1989) discusses about uncritical acceptance. The protocol (See Appendix 7) served as a non-compliance governance framework. The alignment of EU AI Act (transparency through disclosure), (European Commission, 2024), UNESCO principles (critical literacy) (UNESCO, 2023) and Safe AI standards (dependency prevention) the international frameworks were converted into the classroom practice. Mitigation Risk register mitigation in the case of dishonesty, dependency, bias can be seen as a demonstration of mitigation in a scalable way. Ethics awareness (M = 3.4/4.0) is high, and 91 per cent disclosure confirms effectiveness.

All these ethical, pedagogical, and practical aspects are important to inform larger theoretical synthesis of AI-mediated active learning.

5.6 Theoretical synthesis: Active learning–GenAI synergy framework

Findings validate the Active Learning–GenAI Synergy Framework, showing AI can function as cognitive partners, metacognitive mirrors, and equity tools. This extends theories of AI in education by embedding technology within active learning, rather than as separate interventions. The achievement of the framework concerning the propensity to write and the acquisition of digital literacies endorses the vision of cognitive partnerships proposed by Holmes et al. (2019) as a tangible illustration of resource-limited situations. It was established by carrying out an implementation using ethical procedures and experiential learning processes. AI reduced cognitive load while enhancing higher-order thinking, challenging assumptions that assistance undermines cognition. This suggests AI-mediated learning can both reduce routine burdens and increase complex engagement. The finding extends Gayed et al.’s (2022) work on hybrid systems, showing cognitive partnerships simultaneously lighten basic demands while fostering deeper processes. The framework achieved this duality by structuring AI use around progressive experiential cycles.

5.7 Implications for EFL writing pedagogy

The shift from product- to process-focused instruction validates Flower and Hayes’s (1981) theory while extending it into AI-mediated contexts, supporting Graham and Perin’s (2007) process approach, highlighting AI-enabled individualized scaffolding in large classes, sustaining agency and critical thought. Improvements in coherence, structure, and revision directly address Pakistani EFL/ESL challenges noted by Mahmood et al. (2020). Multimodal integration expanded academic writing to include digital composition, offering practical strategies for contemporary literacy in constrained settings.

5.8 Unexpected findings and theoretical implications

High engagement combined with critical awareness challenges dependency concerns, showing students can maintain an authorial voice while using AI. This supports the concept of “critical dependence”, where learners leverage AI while retaining evaluation and agency. The observed 6–8 week maturation of cognitive partnerships highlights sustained engagement needs, countering assumptions of immediate technological benefit while validating Kolb’s recursive model. Progressive improvement suggests long-term integration is preferable to short-term adoption, offering guidance for sustainable AI pedagogy.

5.9 Limitations and future directions

Three major categories of limitation arise.

Methodological: The quasi-experimental design in intact classes does not allow causal inference, although baseline equivalence was established and data triangulation was employed to strengthen internal validity.

Contextual: Higher education context peculiarities of Pakistan, specific AI tools and 15 weeks of interest limits the generalizability.

Measurement: Ethical awareness, which is prone to social desirability, is not tested through long-term retention.

The future research needs to focus on the following aspects, including randomized controlled trials, cross-cultural validation, longitudinal tracking, and systematic comparison of AI tools and pedagogical frameworks.