Students are asking for AI guidance, not just policy

Universities have responded to generative artificial intelligence primarily with policies, academic integrity warnings and assessment guidance. Some have embraced it, and others have banned its use entirely. For some students, such as those from linguistically diverse backgrounds, AI functions not as a shortcut but as an equaliser, enabling intellectual contribution without linguistic barriers. Blanket prohibitions risk removing scaffolding that levels the playing field.

While AI policies are necessary, they are not sufficient. Policies define boundaries, but they do not teach verification. We can uphold academic integrity by helping students demonstrate how they engaged with AI: what they accepted, modified or rejected, and how they made those decisions. This is an orchestration problem, and it requires an orchestration solution. Students are not calling for more rules – they are calling for a systematic method. In a study of 167 first-year ICT students, 51.5 per cent requested specific guidance on how to verify the accuracy of AI-generated content, and 33.5 per cent requested step-by-step examples demonstrating how to query AI tools effectively. 

We developed the Structured AI-Guided Education (SAGE) framework to help students understand how to verify AI-generated outputs for accuracy and relevance against authoritative sources. With this guidance, 73 per cent did verify AI outputs systematically. A further 81 per cent engaged in deep revision in response to identified inaccuracies and gaps, either rewriting AI-generated content entirely or using iterative follow-up questioning to refine it. Only 14 per cent adopted surface-level approaches such as minor formatting adjustments.

SAGE was developed and validated across six empirical studies involving more than 500 students at five Australian university campuses. The workflow comprises six steps.

In the first step, “generate”, students work from a standardised prompt designed by the educator, specifying the task context, constraints and discipline-specific parameters that the AI model requires but cannot determine from the prompt alone. This establishes a consistent baseline, ensuring equity across the cohort, and models effective prompt construction through practice.

• Spotlight guide: AI and assessment in higher education
• Designing assessments with generative AI in mind
• We should be thinking about assessments and AI in a completely different way

In the second, “evaluate”, students verify AI outputs against authoritative sources. SAGE embeds this verification requirement into the assessment design itself, so that students develop cross-referencing skills through doing. If students cannot verify a claim, they cannot use it. 

The next stage, “refine”, requires students to modify outputs with evidence-based justification, recording each decision as “accept”, “modify” or “reject” in a structured decision log. 

The next stage, “AI critic”, involves students submitting their refined work back to AI for critique, then evaluating the evaluator, determining which feedback reflects genuine insight and which reflects algorithmic limitation. 

The fifth stage is “reflect”. Here, students produce a brief metacognitive analysis documenting what they learned about AI limitations and their own decision-making processes. After doing this, most of our participants indicated they would modify their AI strategy in future assessments. This demonstrates the growth mindset this phase is designed to cultivate.

The final step is “defend”. After completing the open, AI-integrated task, students demonstrate their competency under brief, supervised conditions. This is not a full examination. In cybersecurity, it might involve a 10-minute timed risk-scoring exercise based on an unseen scenario. In programming, it could be a short live code walkthrough. In health sciences, it might take the form of a focused clinical reasoning viva voce. The format varies by discipline, but the principle remains the same: the student must demonstrate independent competence.

SAGE does not require wholesale assessment transformation. Educators can begin with three targeted changes: a one-paragraph permitted-use statement clarifying what AI may and may not do, a lightweight decision log for two or three key outputs and rubric criteria that assess the quality of justification rather than the volume of AI interaction. 

Getting started with SAGE

Seven steps to integrate the framework into any unit, regardless of discipline

1. Identify your critical consideration

Choose the competency that must persist across all phases. This anchors every task. Examples include: accessibility in systems analysis, patient safety in nursing and regulatory compliance in cybersecurity. 

2. Design the human baseline

This is what students must produce without AI – bullet points, sketches, interview notes, manual calculations. This is the comparison anchor. Without it, there is no way to assess what value was added.

3. Create a standardised prompt

Include context AI cannot know, such as user demographics, regulatory constraints, stakeholder needs. Standardised prompts ensure fairness across the cohort and model effective prompt construction.

4. Build the decision log template

The decision log is the primary assessment artefact. Students must accept, modify, or reject each AI output with 50-75 word justifications citing specific standards or professional reasoning. 

5. Set a mandatory rejection quota

Require compulsory rejection of AI suggestions with contextual justification. This single design choice prevents passive acceptance and forces students to exercise domain judgement.

6. Weight the rubric: 70 per cent process, 30 per cent product

Assess justification quality, gap detection, contextual reasoning and metacognitive reflection. The product still matters, but the documented process carries most of the assessment weight.

7. Design the “defend” component

Require students to explain and justify their work in a supervised live session. Our research showed only 12 per cent of unsupervised submissions produced an auditable evidence trail. The defence shifts the evidence of learning from the submitted document to the student’s demonstrated reasoning.

Generative AI is now part of the higher education environment, and it looks like it will be fully integrated within the next three to five years. The strategic question is no longer whether to permit it, but whether institutions will equip students with structured, auditable workflows that protect both validity and learning. 

Our students have already demonstrated, through their verification behaviours, their revision depth, and their appetite for guidance, that they are ready to be partners in this process rather than subjects requiring surveillance. Learning how to use AI ethically, critically and efficiently is not a concession to technological convenience. It is the defining educational challenge of this decade. SAGE offers a practical, evidence-based path to meet it.

Mahmoud Elkhodr is a senior lecturer in cybersecurity and an AI education researcher; Ergun Gide is a professor of ICT and a researcher in strategic AI-driven higher education, both at CQUniversity, Australia. 

If you would like advice and insight from academics and university staff delivered direct to your inbox each week, sign up for the Campus newsletter.