Reimagining teacher-AI co-design in learning task design: trends and perspectives

Synthesis: Teacher-AI co-design in learning task design. A systematic review by Zongran Wang, Liu Mingzhuo & A.Y.M. Atiquil Islam (2026) synthesizing 28 empirical studies (2015–2025) on how teachers collaborate with AI — including generative AI — to design learning tasks. GenAI is predominantly used to support lesson planning, prompt generation, and creative ideation, while traditional AI is used for learning analytics and Feedback design. The dominant collaboration mode is AI as assistant/content generator, with AI as a fuller co-designer or dialogic partner much rarer. Four pedagogical affordances recur — efficiency, responsiveness, Creativity, and equity — and the review argues these form a functional-affordance framework for guiding teacher design work with AI.

Overview

Teachers increasingly act as instructional designers working with AI rather than merely using it for delivery. This review maps the theoretical frameworks, AI features, teacher–AI collaboration modes, and design objectives across empirical studies of teacher–AI co-design of learning tasks, in both K-12 and higher education.

Method

• Search: EBSCOhost and Web of Science, 2015–2025; Boolean term combinations ("Teacher Design Practices" AND "AI", "Learning Task Design" AND "AI", "Teacher-AI" AND "design", "Teacher AI co-design") plus snowballing.
• Screening: 672 records → 28 empirical studies after eligibility checks (population = K-12 and higher-ed teachers co-designing tasks; interventions = task creation/application).
• Appraisal: Mixed Methods Appraisal Tool (MMAT); 19 studies rated high quality (5/5), with common limitations reported across designs.

Key findings

• GenAI vs. traditional AI roles differ. GenAI tools are predominantly used for lesson planning, prompt generation, and creative ideation; traditional AI tools are more commonly used for learning analytics and feedback design.
• AI is mostly an assistant, rarely a co-designer. Teacher–AI collaboration manifests in multiple forms, but AI most often serves as an assistant or content generator, and less frequently as a co-designer or dialogic partner.
• Four pedagogical affordances. AI features map onto efficiency (workload reduction), responsiveness (just-in-time instructional adaptation), creativity (idea generation and design space expansion), and equity (differentiated, culturally responsive pathways). Creativity and equity are associated with a broader range of GenAI features; efficiency and responsiveness rely on a narrower set of traditional functions.
• Uneven contexts. Teacher education (36.7%) and higher education dominate the literature; K-12 is underrepresented, attributed to curricular rigidity and limited teacher autonomy.
• Fragmented theory. Theoretical grounding is uneven; many studies use traditional linear instructional-design models that may not fit the distributed, iterative nature of AI co-design.

What this means for practice

• Instructors. Choose the AI tool by the design problem it addresses: efficiency and responsiveness come from a narrower set of traditional AI functions, whereas Creativity and equity are associated with a broader range of generative AI features.
• Instructors. Treat AI output as a draft to be reinterpreted against learner context. Collaboration across the 28 reviewed studies was most often assistant-based, and AI characteristically lacks emotional nuance, cultural awareness, and knowledge of students' prior knowledge.
• Faculty developers. Teach AI as a collaborator that can be questioned, refined, and shaped rather than a static tool, and address the recurring limitations teachers reported: prompt literacy and output quality, over-reliance and loss of pedagogical control, curriculum alignment, trust, infrastructure, explainability, and affective awareness.
• Designers. Move beyond output-generation uses toward dialogic co-design roles, which the review finds underexplored relative to AI-as-assistant and AI-as-content-generator.
• Researchers. Ground new studies in distributed, iterative theoretical models rather than traditional linear instructional-design models, and study real K-12 classrooms, where co-design research is underrepresented relative to teacher education and higher education.

Limitations

• The review synthesizes 28 empirical studies drawn from 672 screened records, and the evidence is unevenly distributed: teacher education accounts for 36.7% and higher education 30.0%, with K-12 underrepresented.
• Only 19 of the 28 studies met five of five MMAT criteria, with 6 rated moderate and 3 lower; the common limitations reported across the included designs were limited reflexivity and small sample sizes.
• The authors state no formal statistical assessment of reporting bias (such as funnel plot analysis) was conducted, and certainty of evidence was not appraised with GRADE, given the conceptual and design-oriented focus of the included studies.
• The search drew on EBSCOhost and Web of Science only (2015–2025) plus snowballing, and many included studies lacked explicit theoretical grounding, which limits what the synthesis can say about mechanisms.

Citation

Wang, Z., Liu, M., & Islam, A. Y. M. A. (2026). Reimagining teacher-AI co-design in learning task design: trends and perspectives. Humanities and Social Sciences Communications, 13, 757.