The Scaffolded AI literacy (SAIL) framework: Results of a Delphi study for equitable AI literacy framework design in education

Synthesis: MacCallum, Parsons, and Mohaghegh (2026) report on a three-round Delphi study that created the Scaffolded AI Literacy (SAIL) framework — a broadly applicable, age-agnostic framework for developing equitable AI literacy across all stages of education. Unlike most existing frameworks, which aggregate older literature or focus on non-generalizable contexts, SAIL provides a scaffolded competency pathway across levels and explicitly addresses second- and third-level digital divides. Grounded in Aotearoa New Zealand education while incorporating international and Indigenous perspectives, the framework organizes AI literacy into three domains (AI Concepts; Application of AI and Technical Skills; AI Digital Citizenship) and six categories, progressing through four levels (Understand and Explore AI → Apply and Integrate AI → Evaluate and Create AI → AI++, beyond AI literacy).

Key Findings

• The SAIL framework is organized into three domains and six categories: (1) Concepts of AI (The Impacts of AI; What AI Is and How It Works), (2) Application of AI and Technical Skills (Cognitive Skills; Applied Skills), and (3) AI Digital Citizenship (Social, Cultural, and Ethical Issues; Risks and Mitigations). Each category is detailed across four levels.
• The framework defines four scaffolded levels (three plus one): Level 1 - Understand and Explore AI, Level 2 - Apply and Integrate AI, Level 3 - Evaluate and Create AI, and AI++ - Transform and Develop AI (Beyond AI Literacy) — the last marking the move beyond literacy into expert practice.
• SAIL is age-agnostic: Level 1 concepts are considered essential for all ages, while higher levels are introduced at appropriate ages and stages, with educators adapting delivery to context.
• The framework explicitly addresses the three-level digital divide (van Deursen & van Dijk): access/infrastructure, skills and competencies, and outcomes/benefits — arguing that equity requires more than closing the device gap; it requires building skills to use AI effectively and critically so benefits are distributed fairly.
• It incorporates Indigenous perspectives (Māori data sovereignty, cultural misrecognition, equitable participation) as a core competency strand ("Indigenous Perspectives on AI"), reflecting Aotearoa New Zealand's international leadership in bicultural education and Indigenous data sovereignty.
• The study reframes AI literacy to include critical awareness of bias (data-related, algorithmic, user-interaction bias), equity, AI Governance, and cultural diversity — distinct from general digital literacies.

Study Design & Method

The authors conducted a three-round Delphi study (an expert-panel consensus method) between May 2023 and May 2024. An expert panel comprising educators (tertiary, high school, primary, and support roles) and industry professionals (data scientists, AI leads, founders) was recruited from Aotearoa New Zealand and internationally (Australia, Canada, South Africa, the UK, USA), with deliberate attention to Indigenous voices. Round 1 (17 fully completed responses) asked experts to identify the knowledge, skills, and understandings needed at each level of an initial four-level maturity model (Informed → Empowered → Engaged → Active participant); responses were thematically coded. Round 2 asked experts to rank competencies by importance and flag misplaced, irrelevant, or missing items, leading to rationalization into three domains and six categories. Round 3 validated the framework structure and refined competency wording. Following the Delphi, the draft was opened to wider review through communities of practice and an online feedback form, yielding further refinements (e.g., renaming levels, separating "AI++ — beyond AI literacy," renaming the "Issues, Challenges, and Opportunities" domain to "AI Digital Citizenship").

What this means for practice

• Instructors. Place learners by capability rather than by year group: SAIL is age-agnostic, with Level 1 (Understand and Explore AI) treated as essential for all ages and higher levels introduced at appropriate stages, so delivery follows what a learner can do, not the cohort they arrived in.
• Instructors. Teach equity as progression through the second and third digital divides — skills and outcomes — not as access to devices, and cover the bias the framework names (data-related, algorithmic, and user-interaction).
• Designers. Thread Indigenous perspectives through the curriculum as a competency strand in its own right (Māori data sovereignty, cultural misrecognition, equitable participation) rather than adding them as a supplementary module.
• Administrators. Map existing provision against SAIL's three domains, six categories, and four levels before commissioning new AI literacy programs, so a single pathway runs from early years through higher education and teacher training instead of separate ad hoc offerings.

Limitations

The Delphi study had a modest fully-completed response rate in Round 1 (17 respondents) and was deliberately grounded in Aotearoa New Zealand, potentially limiting direct generalizability despite international participation. Under ethical approval, data could not be publicly shared to protect panel anonymity. The framework's broad, age-agnostic design means educators must adapt delivery to their specific contexts, and the authors note the field is rapidly evolving — future work should examine how emerging frameworks align with, extend, or diverge from SAIL and integrate insights across approaches. The introductory framework review was not a systematic literature review but contextual framing.

Citation

MacCallum, K., Parsons, D., & Mohaghegh, M. (2026). The Scaffolded AI literacy (SAIL) framework: Results of a Delphi study for equitable AI literacy framework design in education. Computers and Education: Artificial Intelligence.