New Study Reveals How Prompt Skills and Smart Assessment Make Students Transparent With AI

Generative artificial intelligence has swept through university campuses faster than almost any educational technology before it, and with that speed has come an uncomfortable question: when a student submits work shaped by a chatbot, who actually knows what? A new study published in Frontiers of Digital Education tackles that question head-on, arguing that the answer lies not in surveillance software or blanket bans, but in a property the researchers call epistemic transparency — the willingness and ability of students to openly show how AI figured into their learning and to stand behind the knowledge they produce. The research, conducted by a team of Indian scholars led by Abhishek N. of Hassan University, provides some of the first quantitative evidence that specific, teachable skills can make students more honest and open about their AI use.

The study surveyed 649 undergraduate and postgraduate students across Indian higher education institutions, using a cross-sectional questionnaire design that captured how often and how thoughtfully students engaged with generative AI tools. Rather than treating AI use as a single behavior, the researchers decomposed it into distinct competencies. Prompt literacy — the craft of formulating effective instructions for AI systems — was measured as one construct. Critical AI literacy — the capacity to question, verify, and ethically evaluate what an AI produces — was measured as another. Assessment design, meaning the degree to which coursework is structured in ways that make genuine understanding visible, formed the third pillar of the model. The outcome variable, epistemic transparency, captured whether students disclose AI involvement, justify their sources of knowledge, and take responsibility for the claims they make in academic work.

To analyze the relationships among these constructs, the team employed partial least squares structural equation modeling, a statistical technique well suited to testing networks of direct, indirect, and interacting effects among latent variables that cannot be observed directly. The modeling rigor was considerable: the authors reported reliability checks using Cronbach’s alpha, discriminant validity assessment following the HTMT criterion established by Henseler and colleagues, and common method bias screening through Kock’s full collinearity approach. Missing data were handled with modern imputation strategies, and the measurement model passed the Fornell-Larcker benchmarks. In plain terms, the statistical machinery was strong enough that the findings deserve attention rather than dismissal.

The headline result is a chain of influence that runs from the humble prompt to the philosophy of knowledge. Prompt literacy, the study found, has a strong positive effect on critical AI literacy. Students who learn to write precise, contextualized, well-structured prompts do not merely get better outputs; they develop a more questioning stance toward the technology itself. The act of engineering a prompt forces a learner to articulate what they actually want to know, and that articulation appears to sharpen their skepticism about what comes back. This finding aligns with a growing body of literature, including work by Federiakin and colleagues framing prompt engineering as a genuine twenty-first-century skill, and research by Knoth and colleagues showing that AI literacy shapes prompting strategies in a reciprocal loop.

Intriguingly, prompt literacy’s direct effect on epistemic transparency was statistically significant but modest. Knowing how to talk to a chatbot does not, by itself, make a student candid about using one. The pathway runs instead through critical AI literacy, which partially mediated the relationship between prompting skill and transparent behavior. In other words, students become epistemically open when they can critically evaluate AI outputs — spotting hallucinations, recognizing training-data biases, and understanding that a fluent paragraph is not the same thing as a verified fact. The study’s authors suggest this is why prompt training alone, a popular quick fix in university workshops, is unlikely to change disclosure behavior on its own. It builds the tool skills but not the epistemic conscience.

The strongest practical lever, according to the data, was assessment design. When assignments are structured so that process matters — drafts, reflections, oral defenses, documented iterations, and evidence of reasoning — students report significantly higher epistemic transparency. Assessment design showed the largest direct contribution of any variable in the model, edging out both literacy constructs. This echoes a substantial international literature on evaluative judgment, including work by Bearman, Tai, Dawson, Boud, and Ajjawi arguing that assessment in the AI era should cultivate students’ capacity to judge quality rather than merely produce outputs, and studies by Bretag and colleagues linking poorly secured assessment designs to contract cheating. The new study extends that logic into the generative AI age: assessments that reward visible thinking make AI use legible, and legibility is the precondition for transparency.

One of the study’s more sobering findings is what did not work. The interaction between assessment design and critical AI literacy was not statistically significant, indicating that these two factors operate independently rather than synergistically. Transparent AI-supported learning, the data suggest, is not produced by combining clever assignments with critical students in some multiplicative fashion. Each factor contributes on its own track. For educators, this carries a practical implication: institutions cannot assume that strong assessment reform will automatically cultivate critical AI literacy, or that critical literacy curricula will compensate for opaque assessment practices. Both need deliberate, parallel investment.

The conceptual contribution of the paper may prove as influential as its statistical results. By defining epistemic transparency as a distinct, measurable learning outcome — grounded in the epistemology of social knowledge developed by Goldman and the ethics of knowing articulated by Fricker — the study gives universities something concrete to target. Transparency here is not a vague aspiration but an observable behavior: disclosing AI assistance, tracing claims to defensible sources, and exhibiting what the authors, drawing on Flavell’s metacognition tradition, describe as reflective monitoring of one’s own knowledge production. Related recent work, such as Lloyd’s proposal for community standards of epistemic responsibility in human-AI collaboration and Nieminen and Ketonen’s account of epistemic agency in assessment, suggests the field is converging on this framing from multiple directions.

For a sector gripped by anxiety over AI-fueled cheating, hallucinated citations, and unreliable detection tools — problems documented in studies ranging from Elsayed’s analysis of misinformation risks to Deep and colleagues’ evaluation of AI detectors — the study offers a refreshingly constructive agenda. Instead of policing students, it proposes equipping them: teach prompt literacy as a gateway skill, build critical AI literacy as the engine of honest evaluation, and redesign assessments so that the process of thinking is visible and valued. The authors are careful to note the limits of their design; a cross-sectional survey captures associations at one point in time and cannot establish causation, and the sample, drawn entirely from Indian institutions, may not generalize everywhere. Still, with 649 respondents and a rigorously validated model, the evidence is hard to ignore.

As generative AI becomes as unremarkable in seminar rooms as the calculator once was, the question facing higher education is no longer whether students will use these systems but whether their use will be honest, critical, and educationally meaningful. This study suggests the answer can be engineered — not through code, but through curriculum. If universities take its findings seriously, the future of AI in education may look less like an arms race between detectors and deceivers, and more like a classroom where the prompt, the critique, and the assignment all work together to make knowing itself transparent.

Subject of Research: The roles of prompt literacy, critical AI literacy, and assessment design in fostering epistemic transparency among university students using generative AI.

Article Title: Developing Epistemic Transparency in AI-Supported Higher Education: Roles of Prompt Literacy, Critical AI Literacy, and Assessment Design

Article References: N., A., Nayak, K. M., Divyashree, M. S., & Jain, U. (2026). Developing Epistemic Transparency in AI-Supported Higher Education: Roles of Prompt Literacy, Critical AI Literacy, and Assessment Design. Frontiers of Digital Education, 3(4), Article 28. https://doi.org/10.1007/s44366-026-0102-2

Image Credits: AI Generated

DOI: 10.1007/s44366-026-0102-2

Keywords: generative artificial intelligence, higher education, prompt literacy, critical AI literacy, epistemic transparency, assessment design, PLS-SEM, academic integrity, AI literacy, educational research, Developing, Epistemic