Harnessing artificial intelligence for preservice teachers' development: A scoping review of applications, benefits, and challenges

Synthesis: Ziying, Yongchun, and Qiaoping (2026) conduct a PRISMA-ScR scoping review of 55 empirical studies (2020–2025) examining how artificial intelligence is applied in the Professional Development of preservice teachers. Following Arksey and O'Malley's five-stage framework and reflexive thematic analysis, they searched Web of Science, ScienceDirect, and EBSCOhost, screening 1,688 records down to 55 studies involving 3,098 preservice teachers across 18 countries. They map a diverse landscape of AI applications, technologies, and subject distributions, finding that AI acts as both a cognitive partner and a practice simulator across six application scenarios. AI enhances Learning Design, subject-specific instruction, practical teaching skills, evaluation efficiency, reflective practice, Critical Thinking, technology integration (Technological Pedagogical Content Knowledge (TPACK)), and pedagogical innovation. However, the integration is a "dual-edged force": alongside the benefits sit four categories of challenges — technical limitations, the risk of Over-Reliance, ethical and social concerns, and implementation barriers — underscoring the need for ethical and pedagogical grounding in teacher preparation.

Key Findings

Systematic search and reliability. The review screened 1,688 initial records; after removing 74 duplicates and excluding 1,423 in title/abstract screening, 55 empirical studies were included via independent double-blind review plus forward/backward snowballing. Coding was conducted in MAXQDA 2024 with strong inter-rater reliability (Cohen's kappa 0.75 on 10 articles, rising to 0.88 on a further 5).

Scope and distribution. Studies spanned 18 countries (most prolific: China n = 12, South Korea n = 10, the US n = 5, Germany n = 5, Turkey n = 4) and a total of 3,098 preservice-teacher participants. Publication surged recently — 44% (n = 24) from 2024 and 38% (n = 21) from 2025 — and mixed-methods designs dominated (58%, n = 32), followed by qualitative (31%, n = 17) and quantitative (11%, n = 6).

Six application scenarios. AI supported (1) instructional design and lesson preparation, (2) subject-specific teaching, (3) teaching practice and Simulation training (AI-powered virtual students and virtual classrooms), (4) teaching evaluation and Feedback, (5) reflection and critical thinking (AI dashboards, reflective coaches), and (6) technology integration and teaching innovation.

Technology hierarchy. Generative AI was the predominant technology, appearing in 52% of studies, with ChatGPT the most-specified tool (66.7% of GenAI studies = 34.7% of the total sample). Dialogue/interaction systems (NLP, chatbots) underpinned 20.0%, education-specific systems (VR simulations, Intelligent Tutoring systems) 14.7%, machine learning 9.3%, and Automated Assessment 4.0% — with over half of studies combining multiple AI technologies into Multimodal AI solutions.

Disciplinary concentration. Among the 40 studies with a clear disciplinary context, language education was most represented (42.5%, n = 17) and Math Education second (27.5%, n = 11); science and physics each accounted for 7.5%, interdisciplinary/comprehensive 10%, and history and chemistry 2.5% each — leaving limited work in the arts and humanities.

AI as cognitive partner and practice simulator. AI both supports reasoning and design (drafting lesson plans, resource searches, reducing extraneous Cognitive Load Theory) and offers safe, low-risk simulated classrooms in which preservice teachers refine questioning skills, responsiveness, and classroom-management behaviors — a practice–feedback–iteration loop that also boosts Self-Efficacy.

Benefits across five domains. Frequency counts showed the largest benefit clusters in enhancing subject-specific pedagogy and practical teaching skills (26 studies) and instructional design quality/efficiency (21 studies), followed by teaching reflection and critical thinking (15), teaching evaluation (5), and technology integration/pedagogical innovation (5).

Four challenge categories. Technical limitations (28 studies: factual errors, outdated information, fictitious citations, poor non-English processing, mechanical virtual students), ethical and social risks (20: data Privacy, algorithmic bias, plagiarism, copyright, job anxiety), risk of over-reliance (19: adoption of AI-generated lesson plans weakening creative design, and degeneration of reflective depth, evaluation skills, and independent Problem Solving), and implementation barriers (16: infrastructure gaps, the Digital Divide, low user AI Literacy, and uncertainty about long-term effects given mostly short-term studies).

Implication for teacher preparation. The findings argue that teacher-education programs should intentionally design AI-enhanced training that builds AI Literacy and human–AI collaboration while preserving the human educator's role — a core concern for the knowledge base's Professional Development and Teaching concepts.

What this means for practice

• Instructors. Use AI-powered virtual students and simulated classrooms as a low-risk practice–feedback–iteration loop, and make verification of AI output an explicit task, since technical limitations such as factual errors, outdated information and fictitious citations were reported in 28 of the 55 studies and otherwise risk modeling shallow inquiry.
• Faculty developers. Build critical AI Literacy into preservice programs before tool use; low user literacy and infrastructure gaps were the leading implementation barriers across 16 studies.
• Administrators. Protect process- and reflection-oriented Assessment as AI-generated content raises academic-integrity concerns, and recognize teachers' relational and ethical roles as uniquely human.
• Researchers. Fill the arts-and-humanities gap: among the 40 studies with a clear disciplinary context, language education was 42.5% and mathematics 27.5%, leaving 2.5% each for history and chemistry and little for the arts.

Limitations

• The synthesis covers 55 empirical studies (2020–2025) involving 3,098 preservice teachers across 18 countries; 58% used mixed-methods designs and only 11% quantitative, limiting what can be pooled.
• The evidence base is heavily weighted to recent, short-term, lab-based studies, so long-term effects on teacher development and student learning in authentic classrooms remain unknown.
• Search was limited to three databases (Web of Science, ScienceDirect, EBSCOhost) and to peer-reviewed articles published in English from January 2020, excluding non-English work and Gray literature.
• Disciplinary coverage is skewed: among the 40 studies with a clear disciplinary context, language education (42.5%) and mathematics (27.5%) dominate, with minimal work on the arts and humanities.

Citation

Ziying, L., Yongchun, H., & Qiaoping, Z. (2026). Harnessing artificial intelligence for preservice teachers' development: A scoping review of applications, benefits, and challenges. Computers and Education Open, 100330. https://doi.org/10.1016/j.caeo.2026.100330