Teaching with Generative Artificial Intelligence: Enhancing Critical Thinking and Ethical Awareness in Academic Writing

Synthesis: Benali Taouis & Díaz García (2026), Journal of University Teaching and Learning Practice 23(5). This study integrates generative AI tools — ChatGPT (GPT-4) and Writefull for Word — into the English for Professional and Academic Communication (EPAC) course at the Universidad Politécnica de Madrid (UPM), an academic writing course. Framed by critical digital pedagogy (CDP), self-regulated learning (SRL), and Bloom's revised taxonomy, the intervention guided 96 final-year Spanish undergraduate students (in 24 project groups) through a sequential drafting–revision–reflection workflow while developing research-proposal (RP) sections. Analysis of 24 reflective reports (inductive thematic analysis with grounded-theory coding) showed that the process of verifying, revising, and adapting AI-generated content strengthened students' critical thinking and assessment skills, deepened content awareness, and fostered ethical, responsible academic practice. Rather than banning AI, the study models guided, intentional integration in which AI functions as a cognitive and linguistic scaffold — not a substitute for original thinking.

Key Findings

• Verifying AI content strengthens critical thinking and assessment skills. Confronting ChatGPT's hallucinations forced students to validate sources and fact-check outputs, turning initial drafts into objects of critical scrutiny rather than finished text. Students reported that the "critical thinking component lies in the human interpretation" of AI output, positioning themselves as active editors and evaluators.

• Students were active agents, not passive consumers of AI. Across group reports, learners demonstrated metacognitive engagement and critical reflection: refining prompts, rephrasing vague outputs, rejecting unhelpful suggestions, and interrogating both ChatGPT's claims and Writefull's stylistic recommendations (e.g., discarding "impact"→"effect" substitutions that shifted meaning).

• Two complementary tool roles emerged from thematic analysis. ChatGPT functioned as a generative and reflective writing companion (idea generation, gap identification, literature-review support, contextualized prompting), while Writefull functioned as a linguistic coach for grammar, vocabulary, and academic tone. Together they address higher-order (idea development, reflection) and lower-order (accuracy, style) aspects of writing.

• SRL and Bloom's taxonomy structure the cognitive progression. Activities were deliberately aligned with Bloom's revised taxonomy — from remembering (recalling RP structure, grammar rules) through applying (prompt design, draft generation), analyzing (checking for hallucinations, bias), and evaluating (judging source credibility) to creating (synthesizing AI input with original ideas). Students were required to justify every acceptance or rejection of AI feedback, reinforcing autonomy and responsibility.

• AI reduces linguistic anxiety via dynamic Scaffolding. Writefull alleviated language-related insecurity in this L2 (English) context by offering immediate grammatical and stylistic corrections, boosting student confidence while preserving authorship — provided students maintained a critical lens on its occasionally inconsistent suggestions (e.g., British vs. American English).

• AI is a support, not a substitute, for critical writing. By limiting AI use to just two RP sections and mandating documentation (screenshots, justifications, analytical reports), the design prevented over-reliance and transferred learning from instruction to practice, supporting reflective, self-regulated, and ethically aware academic writers.

Study Design & Method

• Context & participants. Conducted within the EPAC course for computing-related degrees at UPM (Escuela Técnica Superior de Ingenieros Informáticos), 2024–2025 academic year. Ninety-six final-year Spanish undergraduates self-organized into 24 interdisciplinary groups of four across four degree programs (Computer Engineering; Mathematics and Computer Science; Data Science and AI; Computer Engineering & Business Administration double degree). Ethical approval obtained; voluntary, anonymous, minimal-risk participation with informed consent.

• Theoretical framework. The design integrated three complementary lenses: critical digital pedagogy (ethical, reflective, socially aware engagement with technology, rooted in Freire's critical consciousness) to guide critical evaluation of AI content; self-regulated learning (goal-setting, monitoring, reflective evaluation per Zimmerman and Pintrich) to build autonomy and metacognition; and Bloom's revised taxonomy to scaffold cognitive progression from lower-order to higher-order thinking.

• Teaching intervention. After a training session on responsible AI use (bias, hallucinations, citation accuracy, prompt engineering), students drafted the Introduction and Innovation sections of a research proposal using ChatGPT (GPT-4), writing precise, content-aware prompts and screenshotting the process. In the revision phase they refined drafts with Writefull for Word, documenting modifications and justifications. A reflection phase required justifying acceptance/rejection of edits and verifying ChatGPT-generated citations. Final versions were submitted via Moodle as part of formative assessment.

• Measures & data collection. Multiple sources: AI-generated drafts, student revisions, interaction screenshots, analytical reflection reports (24 two-page reports guided by an assessment rubric), and final RP submissions.

• Data analysis. Inductive thematic analysis (Braun & Clarke) informed by grounded-theory coding (Corbin & Strauss), with open coding → axial coding → two overarching themes. Intercoder reliability was assessed on a random 20% sample; ChatGPT was used solely as an analytic cross-check, with final coding decisions retained by researchers. Researcher reflexivity and transparent documentation supported trustworthiness.

What this means for practice

• Instructors. Bound AI use to a defined slice of the assignment and grade the verification around it: students used ChatGPT only for the Introduction and Innovation sections of their research proposal, with screenshots of every prompt and a written justification for each suggestion accepted or rejected.
• Instructors. Require students to trace and check every AI-supplied citation. The study's gains in critical thinking came from confronting ChatGPT's hallucinations directly, not from being warned about them.
• Instructors. Pair a generative tool with a language-focused one — ChatGPT for idea generation, gap identification, and literature review, Writefull for grammar, vocabulary, and academic tone — and ask students to articulate why they reject suggestions that change the meaning or register of their text.
• Designers. Sequence AI tasks against Bloom's revised taxonomy and self-regulated learning phases, moving from recalling structure and rules, through prompt design and draft generation, to analyzing bias and hallucinations and judging source credibility, and leave the final synthesis to the student. The authors locate the educational value in this pedagogical intentionality rather than in the tools themselves.
• Designers. Build AI literacy and prompt engineering into the course rather than treating them as assumed skills, and make documentation of AI use — screenshots, justifications, reflective reports — a graded, process-visible part of formative assessment, the design element that displaced accountability onto students.

Limitations

• The evidence is 24 group reflective reports of about two pages each, submitted by 96 final-year undergraduates who self-organized into groups of four in a single English for Professional and Academic Communication course at one Spanish university, so the accounts are group-level and course-bound rather than 96 independent perspectives.
• Intercoder reliability was checked on a randomly selected 20% of the reports, and there is no independent outcome measure: the claims about strengthened critical thinking and assessment skill rest on students' own analytical reports, screenshots, and revisions.
• The intervention deliberately touched only two research-proposal sections, Introduction and Innovation, so it documents a bounded slice of the writing process; the authors state that the study is situated in a computing-focused context at a single Spanish institution and describe transfer to other settings as instructor adaptation rather than demonstrated generalization.

Citation

Benali Taouis, H., & Díaz García, A. (2026). Teaching with generative artificial intelligence: Enhancing critical thinking and ethical awareness in academic writing . Journal of University Teaching and Learning Practice, 23(5).