The Framework Resources Student Guide GenAI-101 Module Defend Tool Downloads Assessment Policy Student-Facing Policy Educator Guidance SAGE in Practice CQUniversity — Independent GAM Mapping Evidence Community About

Your students are using AI.
You cannot tell who did the thinking.

SAGE is a validated pedagogical framework that guides students through structured AI interaction — and verifies they own the result.

Steps 1–5
Develop capability
Steps 1–5 guide students through AI-integrated tasks: generate, evaluate against standards, refine with evidence, stress-test via AI Critic, and reflect.
Step 6 — Defend
Verify attainment
Step 6 — Defend — requires each student to demonstrate, under supervised conditions, that they own the reasoning behind their submission.
One continuous workflow: AI-supported learning and proportionate assurance within the same staged process.
Version 3.0 · 3 June 2026 Version 3 DOI pending publication Elkhodr & Gide · Central Queensland University

1. What Is SAGE

The Structured AI-Guided Education (SAGE) framework is a validated pedagogy for developing students' capacity to orchestrate generative AI outputs rather than passively accept them. It treats GenAI as an assistive tool whose outputs require explicit, evidence-based decisions to accept, modify, or reject; it is neither an authority to mimic nor a tool to ban.

SAGE was developed through three years of sustained empirical research involving more than 1,500 students across 30+ studies, spanning cybersecurity management, data analytics, Professional Communication, and systems analysis and design, delivered across campuses in Australia, China, and the United Kingdom. The framework has been referenced by researchers at institutions including the University of Warwick and the National University of Singapore.

Version 3.0 of this guide defines the canonical SAGE Student Orchestration Levels and retains the sixth step — Defend — introduced in Version 2, which places assurance of individual competence within the cycle itself rather than relying on an additional supervised assessment. The complete evidence base, discipline-specific examples, implementation templates, and detailed rubrics are available in Version 1 of this guide.

2. Why SAGE

Assessment redesign in the age of generative AI should not begin only with the question, “How can we reduce the risk of students using AI?” It should begin with the learning outcome, the capability students must demonstrate, and the evidence required to show that learning has occurred.

Risk-only starting point

How can we stop or detect AI use?

This approach often adds restrictions, surveillance, or detection measures to an existing task without first asking whether the task still measures the intended learning.

Capability-first starting point

What should students be able to do?

Define the capability, decide whether it should be demonstrated with or without GenAI, and select evidence that validly demonstrates student learning.

The mindset shift
Risk is not ignored. It is considered after the intended capability, authentic role of GenAI, and required evidence have been established.

Four questions before designing the assessment

1
What capability should students develop?

Begin with the learning outcome rather than the existing assessment format.

2
What must students demonstrate independently?

Identify foundational knowledge, reasoning, or procedural competence that must remain personally internalised.

3
Where is GenAI an authentic professional tool?

Consider whether graduates will be expected to use, test, challenge, and take responsibility for AI-supported work.

4
What evidence would demonstrate the capability?

Select evidence that matches the learning: performance, explanation, application, artefacts, decisions, or supervised defence.

The answer may differ across learning outcomes

Programming Write and debug foundational code independently, or use GenAI to develop, test, validate, and defend a reliable software solution.
Networking Explain the TCP/IP layers and apply them to an unseen troubleshooting scenario.
Statistics Use SPSS to produce appropriate descriptive statistics and correctly interpret the resulting output.
Professional judgement Evaluate AI-generated recommendations against evidence, standards, and contextual constraints before making a defensible decision.

A written submission is not always the most valid evidence of these capabilities. Similarly, conducting an assessment during a scheduled class does not automatically make it secure. Evidence should be selected because it matches the capability, not merely because it is familiar or convenient.

A capability-first assessment-design sequence
Learning outcome Required capability Role of GenAI Evidence of learning Assessment design Risk controls

SAGE operationalises this approach. Steps 1–5 structure critical and responsible human–AI collaboration. Step 6, Defend, provides direct verification where the intended capability requires assurance of individual understanding, judgement, or performance.

3. The SAGE Cycle

At the core of SAGE is a six-step cycle that moves students from passive AI consumption to critical AI orchestration, culminating in supervised demonstration of the competency developed through the preceding five steps. The cycle is designed to produce graduates who can deploy generative AI as a professional tool within the constraints of their discipline. They will be taught skills to evaluate AI outputs against industry standards, regulatory requirements, and domain-specific evidence rather than accepting them uncritically. This is the competency that employers now expect. It is not the ability to prompt an AI tool, but rather the ability to judge, correct, and take professional responsibility for what AI produces.

Step 1
Generate
Use GenAI to produce an initial output using structured or autonomous prompts.
→
Step 2
Evaluate
Compare the output against domain standards, regulations, and authoritative sources.
→
Step 3
Refine
Modify the output with evidence-based reasoning. Document what was changed and why.
↓
Step 4
AI Critic
Audit the critic
Ask GenAI to act as a critic, examiner, reviewer, auditor, or devil's advocate, then evaluate its criticism and justify what you accept, modify, or reject.
→
Step 5
Reflect
Conduct metacognitive analysis of the AI's strengths, limitations, and failure patterns.
→
Step 6
Defend
Demonstrate competency under supervised conditions. The student proves they own the reasoning, not just the document.

The cycle is sequenced to move students progressively through Bloom's revised taxonomy. Steps 1–3 operate at the Apply and Analyse levels: students generate output, compare it against domain standards, and modify it with evidence. Step 4, AI Critic, inverts the AI relationship: the student assigns AI a critical role and must audit the critic by evaluating the evaluator, operating at Bloom's Evaluate level. Step 5 requires metacognitive synthesis. This is the highest cognitive operation in the taxonomy, in which students articulate patterns of AI strength and failure across the task. Step 6 closes the cycle by requiring the student to demonstrate, under supervised conditions, that the competency developed through the preceding steps has been individually internalised rather than merely documented.

Design Principle
Steps 1–5 develop the competency through open, AI-integrated tasks. Step 6 verifies it under supervised conditions where the process cannot be simulated. The first five steps serve both learning and formative assessment purposes. Step 6 carries the institutional assurance function.

4. Two-Stage Progression

SAGE operates through two stages that move students from guided practice to independent application with supervised assurance. This progression mirrors the cognitive trajectory defined by Bloom's revised taxonomy: Stage 1 scaffolds the lower-order operations of Apply and Analyse under instructor guidance, while Stage 2 requires students to operate independently at the Evaluate and Create levels before demonstrating competency under supervised conditions. The two-stage design ensures that open and assurance tasks are both addressed within a single integrated teaching sequence rather than treated as separate assessment instruments.

Stage 1 — Formative
Tutorial-Based Scaffolded Learning

Students practise the SAGE cycle under instructor guidance with structured prompts, worked examples, and immediate feedback. Scaffolding is heavy. The goal is to build the evaluation, refinement, and reflection skills that students will apply independently in Stage 2.

  • Generate with standardised prompts provided by instructor
  • Evaluate against supplied checklists, standards, peer-reviewed research and readings
  • Refine with peer review and tutor feedback
  • AI Critic under guided conditions
  • Reflect in short metacognitive entries (150–250 words)
Stage 2 — Summative
Assessment-Driven Independent Application

Students apply the SAGE cycle autonomously in summative tasks with reduced scaffolding. Process logging remains embedded in the open task for formative purposes. Assurance of individual attainment is placed in the Defend step under supervised conditions.

  • Generate with self-designed prompts; human baseline required first
  • Evaluate against standards, stakeholder input, and research
  • Refine with documented accept/modify/reject decisions
  • AI Critic with domain-specific persona assignment
  • Reflect in evidence-based justification
  • Defend under supervised conditions (format varies by discipline)
Constructivist Learning Bloom's Taxonomy Situated Learning Zone of Proximal Development

5. What Students Do in Each Step

Step 1 Generate

Students use a generative AI tool to produce an initial output for the task at hand. In Stage 1, standardised prompts are provided to ensure comparable outputs across the cohort. In Stage 2, students design their own prompts but must first produce a human baseline (outline, notes, or preliminary analysis) before engaging the AI tool. This baseline demonstrates independent understanding of the problem and prevents complete delegation.

Step 2 Evaluate

Students compare the AI output against authoritative sources relevant to their discipline: industry standards, regulatory frameworks, peer-reviewed research, clinical guidelines, or case-study constraints. They identify what is present, what is missing, and what is incorrect. In Stage 1, instructors provide checklists and evaluation templates. In Stage 2, students identify evaluation criteria independently.

Step 3 Refine

Students modify the AI output on the basis of their evaluation. Each modification is documented as an accept, modify, or reject decision with a brief evidence-based justification citing specific standards, sources, or contextual constraints. The refinement produces a professional-standard artefact that the student can defend as their own intellectual product.

Step 4 AI Critic
Audit the critic

Students ask GenAI to act as a critic, examiner, reviewer, auditor, or devil's advocate. They must then evaluate the AI's criticism rather than automatically accepting it, deciding which recommendations to accept, modify, or reject and providing reasons. This step develops metacognitive sophistication and strengthens the student's authority over the AI.

Step 5 Reflect

Students produce a metacognitive analysis documenting the AI's strengths and weaknesses as observed during the task, the specific domain errors identified, the corrective reasoning applied, and the broader implications for AI reliability in their discipline. Functional reflections — those that critique domain content directly — are distinguished from procedural reflections that merely describe workflow benefits.

Step 6 Defend

Students demonstrate, under supervised conditions, that they can reproduce or explain the reasoning documented in their open assessment without reliance on the artefacts that produced it. The format is determined by disciplinary context (see Section 6). The defining criterion is that the student shows they own the competency — they can identify risks, explain trade-offs, justify decisions, and respond to challenge questions — not merely that they submitted a document containing these elements. This step carries the institutional assurance function: it is the evidence that the intended learning outcomes have been individually achieved.

SAGE Student Orchestration Levels

The SAGE Student Orchestration Levels describe the observable quality of a student's collaboration with generative AI. They are different from an AI guidance or permission level. A guidance level states what use of GenAI is permitted in an assessment; an orchestration level describes the capability that the learning design seeks to develop and that assessment criteria may evaluate.

The levels represent a developmental progression. Passive Acceptor is a baseline condition that SAGE is designed to remediate, rather than a desired learning outcome.

Passive Acceptor

Baseline condition — not a target

Accepts AI output with minimal critical evaluation, disciplinary verification, or justification.

Observable indicator
The process record contains accept decisions only, or explanations that merely repeat the AI output rather than testing it against an authoritative source, disciplinary standard, rubric, or other evaluation anchor.
Developmental purpose
SAGE is designed to move students beyond passive acceptance through evaluation, refinement, documented decision-making, and reflection.

Selective Adapter

Foundational target

Filters AI output using disciplinary knowledge and evidence, accepting, modifying, or rejecting selected elements.

Observable indicator
The student records meaningful accept, modify, and reject decisions, with important modifications and rejections justified against a specific source, professional standard, rubric criterion, dataset, policy, or disciplinary principle.
Developmental purpose
The student begins to exercise visible human judgement rather than treating AI output as authoritative.

Balanced Integrator

Developing target

Systematically combines human and AI contributions through explicit trade-off reasoning, contextual judgement, and iterative refinement.

Observable indicator
The student explains the trade-offs at points where the human and AI positions diverged, including why one contribution was preferred, combined, revised, or rejected in the specific assessment context.
Developmental purpose
The student treats GenAI as a collaborative tool while retaining responsibility for disciplinary accuracy, contextual suitability, and final decisions.

Critical Synthesiser

Advanced target

Proactively identifies gaps, assumptions, and bias and develops a traceable synthesis, reframing, solution, or contextual adaptation that extends beyond both the original human position and the AI output.

Observable indicator
The final work contains a substantive development that was present in neither the student's initial attempt nor the original AI output, together with evidence explaining how and why that development was produced.
Developmental purpose
The student uses GenAI as one input within an advanced process of critical inquiry, synthesis, innovation, and disciplinary judgement.

Steps 1–5 provide the process through which stronger orchestration capability can develop. Generate creates material for examination; Evaluate introduces authoritative anchors; Refine makes human contribution visible; AI Critic tests the developing work; and Reflect requires students to account for their decisions. Step 6, Defend, verifies whether the student has personally internalised the reasoning represented in the final work.

6. Defend — Discipline Implementation Examples

The Defend step is format-agnostic. The principle remains constant: the student demonstrates competency under conditions where the process cannot be simulated. The following table illustrates how Defend may be implemented across representative disciplines.

Discipline Defend Format What the Student Demonstrates
Cybersecurity Timed risk-scoring exercise or incident response viva Classifies and prioritises threats for an unseen scenario using the same frameworks applied in the open task. Explains why specific controls were selected over alternatives. Responds to challenge questions on regulatory alignment.
Programming Supervised code walkthrough or live debugging session Explains design decisions in submitted code. Debugs a seeded error in a related module under observation. Demonstrates understanding of logic, structure, and trade-offs rather than surface familiarity with the output.
Health Sciences Structured clinical reasoning viva or case interpretation Applies clinical decision-making to a new patient scenario. Justifies care plan modifications against clinical guidelines. Identifies where AI-generated recommendations require human override based on patient-specific factors.
Business Oral defence of strategic recommendation with examiner challenge Presents and defends a strategic position. Responds to examiner objections with evidence-based reasoning. Demonstrates understanding of stakeholder constraints, market assumptions, and ethical implications.
Education Teaching demonstration with reflective justification Delivers a short teaching segment based on a lesson plan developed with AI assistance. Explains pedagogical choices, adaptation for specific learner needs, and why certain AI-suggested approaches were modified or rejected.
Engineering Design defence with constraint interrogation Defends design decisions under examiner questioning. Explains trade-offs between competing requirements. Demonstrates that safety, sustainability, and regulatory considerations were understood, not merely included in the document.
Law Moot argument or case analysis viva Argues a legal position with reference to relevant authorities. Responds to judicial challenge on the strength of the analysis. Distinguishes AI-generated legal summaries from independently reasoned legal arguments.
Implementation Note
The Defend step does not require a full examination. A 10–15 minute structured oral per student, a timed in-class exercise, or a supervised practical demonstration is sufficient provided the task requires the student to reproduce reasoning rather than recall content. The objective is verification of competency, not endurance testing.

7. Why Defend

The inclusion of a supervised assurance step is not a theoretical preference. It is an empirically grounded response to a documented structural limitation of unsupervised assessment under GenAI-rich conditions.

In a structured audit of 25 group submissions across two cybersecurity management cohorts, assessments designed using the full SAGE protocol, incorporating base prompts, structured decision tables, mandatory AI interaction logs, and reflective commentary, were examined for process fidelity using a five-check protocol. Full traceability between documented AI outputs and human evaluation claims was not achieved in most student submissions. Only 3 of 25 submissions (12%) produced evidence chains that were substantially auditable. The remaining submissions exhibited logical inconsistencies between reported AI positions and appended outputs, compliance-pattern text in evaluation cells, and structural indicators consistent with audit trail simulation (Elkhodr & Gide, 2026, STEM Education).

These findings are consistent with sector-wide guidance. TEQSA's 2025 resource Enacting Assessment Reform in a Time of Artificial Intelligence identifies structural assessment redesign, not detection, as the sustainable response. SAGE with Defend responds by embedding AI-supported learning and supervised assurance within one continuous six-step pedagogical process.

References

Core SAGE Framework Publications

  1. M. Elkhodr, E. Gide, R. Wu, and O. Darwish, "ICT students' perceptions towards ChatGPT: An experimental reflective lab analysis," STEM Education, vol. 3, no. 2, pp. 70–88, 2023. [Online]. Available: https://doi.org/10.3934/steme.2023006
  2. R. Sandu, E. Gide, and M. Elkhodr, "The role and impact of ChatGPT in educational practices: Insights from an Australian higher education case study," Discover Education, vol. 3, no. 1, art. 71, 2024. [Online]. Available: https://doi.org/10.1007/s44217-024-00126-6
  3. M. Elkhodr and E. Gide, "The SAGE framework for developing critical thinking and responsible generative AI use in cybersecurity education," Discover Education, vol. 4, art. 225, Nov. 2025. doi: 10.1007/s44217-025-00935-3
  4. M. Elkhodr and E. Gide, "AI leads, humans lead, or collaborate? Empirical findings and the SAGE roadmap for embedding GenAI in systems analysis and design education," STEM Education, vol. 6, no. 2, pp. 194–229, 2026. doi: 10.3934/steme.2026009
  5. M. Elkhodr and E. Gide, "From Permission to Pedagogy: The Structured AI-Guided Education Assessment Policy (SAGE-AP) for Generative AI in Higher Education," Education Sciences, vol. 16, no. 6, art. 986, 2026. doi: 10.3390/educsci16060986
  6. M. Elkhodr and E. Gide, "Embedding assurance within learning: Empirical evidence from the SAGE framework for repositioning take-home assessment in AI-integrated higher education," STEM Education, vol. 6, no. 4, pp. 584–605, 2026.
  7. M. Elkhodr and E. Gide, "AI as Critic: Validating SAGE Pedagogy for Human Authority and Responsible GenAI Use in Systems Analysis and Design Education," EdArXiv Preprints, 2025. [Online]. Available: https://osf.io/preprints/edarxiv/8j3xf
  8. M. Elkhodr, A. Azra, and E. Gide, "How First-Year Students Actually Use ChatGPT in Permitted Assessments: Empirical Typologies, Verification Gaps, and the Policy-Practice Divide," submitted to Discover Education, 2026. Preprint: https://doi.org/10.21203/rs.3.rs-8628653/v1

Extended Research Portfolio: GenAI in Education

  1. M. Elkhodr and E. Gide, Eds., Generative Artificial Intelligence Empowered Learning: A New Frontier in Educational Technology, 1st ed. New York, NY: Taylor and Francis, 2025, 248 pp. doi: 10.1201/9781003422433. eBook ISBN: 9781003422433
  2. R. J. Eddine, E. Gide, and A. Al-Sabbagh, "Generative AI in higher education: A cross-sector analysis of ChatGPT's impact on STEM, social sciences, and healthcare," STEM Education, vol. 5, no. 5, pp. 757–801, 2025. doi: 10.3934/steme.2025035
  3. K. Wangsa, S. Karim, E. Gide, and M. Elkhodr, "A systematic review and comprehensive analysis of pioneering AI chatbot models from education to healthcare: ChatGPT, Bard, Llama, Ernie and Grok," Future Internet, vol. 16, no. 7, art. 219, 2024. doi: 10.3390/fi16070219
  4. K. Wangsa, R. Sandu, S. Karim, M. Elkhodr, and E. Gide, "A systematic review and analysis on the potentials and challenges of GenAI chatbots in higher education," in Proc. 21st Int. Conf. Information Technology Based Higher Education and Training (ITHET), Paris, France, Nov. 2024, pp. 1–7. doi: 10.1109/ITHET61869.2024.10837608
  5. A. Al Tawara, J. El-Den, E. Gide, and Y. Sebastian, "A systematic review and comprehensive analysis of AI-enabled re-skilling and upskilling in education: Transformative strategies for the future," in Proc. 21st Int. Conf. Information Technology Based Higher Education and Training (ITHET), Paris, France, Nov. 2024, pp. 1–10. doi: 10.1109/ITHET61869.2024.10837638
  6. A. Al Tawara, E. Gide, and J. El-Den, "Systematic review and comprehensive analysis of integrating human-centered AI in higher education: Enhancing teaching, learning, and ethics," in Proc. 12th Int. Conf. Future Internet of Things and Cloud (FiCloud), Aug. 2025, pp. 305–312. doi: 10.1109/FiCloud61071.2025.00051
  7. H. Ranasinghe, E. Gide, and M. Elkhodr, "The significance of GenAI empowered ERP systems course teaching in quality education," in Proc. 21st Int. Conf. Information Technology Based Higher Education and Training (ITHET), Paris, France, Nov. 2024, pp. 1–7. doi: 10.1109/ITHET61869.2024.10837679
  8. G. Chaudhry, E. Gide, E. Yadegaridehkordi, and R. Tumpa, "Generative AI-powered teaching and learning in engineering and project management higher education: A systematic review," in Joint International Conference on AI, Big Data and Blockchain, Cham, Switzerland: Springer Nature, Aug. 2025, pp. 99–113.
  9. E. Gusman, E. Gide, M. Elkhodr, and G. Chaudhry, "The benefits and challenges of using artificial intelligence in teaching English as a foreign language in higher education," in Proc. 21st Int. Conf. Information Technology Based Higher Education and Training (ITHET), Paris, France, Nov. 2024, pp. 1–7. doi: 10.1109/ITHET61869.2024.10837597
  10. E. Gusman, E. Gide, G. Chaudhry, and M. Elkhodr, "A comprehensive review to identify the challenges and opportunities of using digital technology in English teaching in higher education," in International Society for Technology, Education, and Science, 2023.
  11. R. Sandu, E. Gide, S. Karim, and P. Singh, "A framework for GenAI-empowered curriculum and learning resources: A case study from an Australian higher education," in Proc. 21st Int. Conf. Information Technology Based Higher Education and Training (ITHET), Paris, France, Nov. 2024, pp. 1–8. doi: 10.1109/ITHET61869.2024.10837623
  12. N. Sandu and E. Gide, "Adoption of AI-Chatbots to enhance student learning experience in higher education in India," in Proc. 18th Int. Conf. Information Technology Based Higher Education and Training (ITHET), Sept. 2019, pp. 1–5. doi: 10.1109/ITHET46829.2019.8937382
  13. M. Elkhodr, K. Wangsa, E. Gide, and S. Karim, "A systematic review and multifaceted analysis of the integration of artificial intelligence and blockchain: Shaping the future of Australian higher education," Future Internet, vol. 16, no. 10, art. 378, 2024. doi: 10.3390/fi16100378
  14. M. Elkhodr, E. Gide, and N. Pandey, "Enhancing mental health support for international students: A digital framework for holistic well-being in higher education," STEM Education, vol. 4, no. 4, pp. 466–488, 2024. doi: 10.3934/steme.2024025
  15. N. Abbasi, E. Gide, P. Kalutara, and P. Lawrence, "Barriers and opportunities in online delivery of architecture and building design studios: Australian educators' perspectives," 2024.

Foundational Frameworks and Standards Referenced

  1. L. W. Anderson and D. R. Krathwohl, A Taxonomy for Learning, Teaching, and Assessing: A Revision of Bloom's Taxonomy of Educational Objectives. New York, NY: Longman, 2001.
  2. The University of Sydney, "Academic integrity," 2025. Available: sydney.edu.au
  3. International Organization for Standardization, "ISO/IEC 27001:2022 Information security management systems — Requirements," Oct. 2022. Available: iso.org/standard/27001