Introduction

Global statistics indicate that roughly 15% of school-age learners require special-education services, yet the latest National Assessment of Educational Progress still reports persistent achievement gaps (National Center for Education Statistics [NCES], 2023). Conventional print materials and two-dimensional digital media often fail to accommodate the diverse sensory, cognitive, and motor profiles present in special-education classrooms. Augmented reality (AR)technology—the overlay of manipulable virtual objects onto real environments—has emerged as a candidate technology for translating abstract content into embodied experiences(Billinghurst et al., 2015; Carmigniani & Furht, 2011). Because AR can be delivered through marker-based, projection-based, or superimposition techniques on widely available mobile devices(Kesim & Ozarslan, 2012; Yuen et al., 2011), its accessibility has expanded rapidly in educational contexts. However, despite this technological diffusion, educators and policymakers still lack a synthesis that clarifies whether, how, and for whom AR narrows learning disparities in special education—the evidence gap addressed by the present review.

Recent classroom pilots show that AR scaffolds can heighten engagement, self-efficacy, and task accuracy in students with autism spectrum disorder, learning disabilities, or intellectual disability(Mosher et al., 2022; Radu, 2014). By embedding step-wise prompts or multimodal cues, AR lessons have also supported social-emotional growth and daily-living skills acquisition. Yet technical shortcomings—including limited camera resolution, sub-optimal tracking, and steep device costs—often curtail learning time or frustrate users(da Silva et al., 2019; Garzón, 2021). Consequently, it remains unclear which combinations of AR modality, hardware, and instructional design achieve reliable benefits relative to these constraints. A systematic comparison across disability categories is needed to inform evidence-based procurement and teacher training.

Design researchers therefore advocate minimisingextraneous interface elements and interaction steps to reduce cognitive load for learners who may experience working-memory or attentional challenges(Elford et al., 2022). Crucially, the past decade has witnessed a steep drop in hardware prices and the near-ubiquity of rear-facing cameras on smartphones and tablets, enabling wider school- and home-based deployment(María Graciela Badilla-Quintana et al., 2020). Nevertheless, the literature remains fragmented—reporting disparate disability targets, outcome measures, and implementation fidelities—making it difficult togeneralisedesign principles across contexts. A structured synthesis can illuminate patterns and guide the Co-design of low-threshold, high-impact AR interventions.Accordingly, the present reviewsynthesisesthe emerging evidence to clarify when, for whom, and under what conditions AR is most effective for K-12 students with special needs.

For practitioners, an expanding ecosystem of authoring platforms now permits teachers to custom-tailor AR activities, foster peer collaboration, and extend practice into remote or hybrid learning settings(Cerritelli et al., 2021). These affordances are particularly salient for rural or resource-constrained schools where specialist therapists are scarce. However, extant studies seldom report cost–benefit analyses, long-term maintenance issues, or student privacy considerations—factors pivotal for sustainable adoption. Mapping such implementation variables alongside learning outcomes will help stakeholders weigh investment options and anticipate support needs.

Purpose and research questions. Building on the gaps identified above, we conduct a systematic literature review to answer six questions:

RQ1: What is the distribution of the publications over time?

RQ2: What are the geographic and population distributions of the research?

RQ3: What research designs and methods have been adopted?

RQ4: What types of disabilities are covered in the reviewed studies?

RQ5: Which AR technologies/methods have been used?

RQ6: How effective areAugmented Realityinterventions in improving specific outcomes for students with special needs and special education?

Literature Review

Disabilities

Globally, around 15 % of the population—over one billion individuals—live with some form of disability(Krahn, 2011). Students with special educational needs are commonly classified by functional impact rather thanaetiology, yielding categories such as physical impairment, autism spectrum disorder (ASD), intellectual disability, Down syndrome, and mental-health–related learning difficulties(Scott et al., 1998). Yet prevalence data mask substantial heterogeneity in cognitive profiles, sensory processing, and environmental barriers. Recent cross-national surveys reveal that students with ASD and learning disabilities represent over 60 % of special-education enrolments, whereas visual or motor impairments account for less than 5 %(Tang et al., 2022). This disparity is mirrored in research attention, creating a knowledge gap on interventions for low-incidence but high-needs groups. Because disability is now understood as aninteraction between individual impairment and contextual constraints (WHO ICF framework), technology-mediated supports must be evaluated not only for efficacy but also for ecological fit. Addressing this multidimensionality is essential for developing evidence-based, scalable solutions that move beyond single-diagnosis perspectives(Braithwaite et al., 2008; Peters, 2003).Despite promising case studies, most trials remain short-term and rarely address low-incidence disabilities, underscoring the need for a systematic synthesis.

AR for Disabilities

AR technology applications for special education cluster around four learning domains: academic enrichment, social-communication, motor rehabilitation, and daily-living routines(Blanchet & Assaiante, 2022). Across the 16 studies retrieved, 68 % use simulated real-life scenarios—for example, overlaying 3D shopping items onto a classroom shelf to teach money skills(Koumpouros, 2024)or projecting animated mouth shapes to support phoneme articulation. Meta-analytic conversion shows medium effects on immediate task accuracy (g ≈ 0.54) but small effects on transfer to natural settings (g ≈ 0.29), suggesting ecological validity remains limited. Hardware cost and usability continue to constrain implementation; half of the studies report device prices above USD 500, and only two describe teacher training protocols in detail(McGrath et al., 2023; Phan et al., 2022). Moreover, just three papers incorporate universal-design principles or adaptive interfaces to accommodate diverse sensory profiles(Lee et al.,2019). These methodological gaps—limited transfer evidence, sparse cost analyses, and inadequate accessibility features—highlight why the present systematic review compares AR modalities, design features, and outcome robustness across disability categories. Collectively, these methodological gaps justify a quality-rated synthesis comparing AR modalities, disability categories, and outcome robustness. Comprehensive synthesis will aid stakeholders in selecting cost-effective, context-appropriate AR solutions and in formulating research agendas that move beyond proof-of-concept demonstrations(Henne et al., 2024).

Methodology

Search Strategy

A systematic search of four electronic databases (Scopus, IEEE Xplore, Web of Science, and ProQuest) was conducted to identify peer-reviewed articles published in English between January 2010 and May 2024. The search strategy combined keywords related to the intervention ("augmented reality," "AR") and the population ("special needs," "special education," "disabilities"). To ensure precision, the search was limited to the title, abstract, and keyword fields. An exemplary search string, as used in Scopus, IEEE Xplore, Web of Science, and ProQuest was: TITLE-ABS-KEY("augmented reality" OR "AR") AND TITLE-ABS-KEY("special education" OR "special needs" OR "disabilities"). Full search strategies for all databases are provided inTable 1. This string was adapted for the specific syntax of each database to maintain search consistency. Furthermore, a manual snowballing process, including backward searching of reference lists and forward searching of citing articles, was performed on all included studies to identify any additional relevant literature. This review was conducted and reported following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement(Page et al., 2021).

Table 1. Full Electronic Search Strategies

Inclusion and Exclusion Criteria

Inclusion and exclusion criteria were considered in the study selection process to obtain relevant studies within the scope of this study (Liberati et al., 2009). To ensure a successful selection process, reviewers carefully set inclusion criteria. The selected papers had to be review articles published in English that were fully available in the selected databases, as shown in Table 2. Publications that did not meet these criteria were excluded from the review.

Table 2. Inclusion and Exclusion Criteria

Selection Procedure

The PRISMA flowchart (Figure 1) was used to select the primary studies. PRISMA stands for “Preferred Reporting Items for Systematic Reviews and Meta-Analyses’ (Page et al., 2021). From the database searches, a total of 5466 articles (n = 5466) were identified, distributed as follows: IEEE(n = 809), Scopus (n = 567), Web of Science (n = 3930), and ProQuest (n =160). Before screening the articles, all selected articles were checked to ensure no duplicates were retrieved from different databases. In total, 20 duplicate articles (n = 20) were identified. The screening process was conducted independently by reading the titles and abstracts of 5446 articles (n = 5446). After that, they debated and double-checked to determine whether the articles should have been part of the review. The screening process resulted in the rejection of 5304 research articles (n = 5304) as they were unrelated to the research topic. Therefore, more research articles (n = 66) were excluded for various reasons; for instance, they were not peer-reviewed, did not have open access to the full text, or were not written in English. Subsequently, the researchers performed a full-text reading of the remaining articles (n=76). Sixty articles were excluded because they were unrelated to the research questions and deemed irrelevant to this review. The articles must involve individuals with special needs using AR technology. Finally, the remaining 16 articles (Afriantoet al., 2019; Awang et al., 2019; M. G. Badilla-Quintanaet al., 2020; Franco-Arellano et al., 2024; Hashim et al., 2022;Intarapreecha&Sangsawang, 2023;Kaimara, 2023; Kellems et al., 2020;Khoirunnisaet al., 2023; Lazo-Amado et al., 2022; Lorusso et al., 2018;Miundyet al., 2019;Pavlendovaet al., 2024;Sáiz-Manzanares et al., 2024;Widyaningsihet al., 2023;Wiliyantoet al., 2022) were included in the bibliography and reference management system of EndNote21, which will significantly help researchers collaborate and engage in multiple activities simultaneously. In addition, activities can be monitored, and changes can be tracked from the point of selection up to the data extraction stage.

The screening process was conducted by two independent researchers to minimize bias. Initially, both researchers independently screened the titles and abstracts of all identified articles against the predefined inclusion criteria. Any disagreements regarding the inclusion or exclusion of a study were resolved through discussion to reach a consensus. If a consensus could not be reached, a third senior researcher was consulted for a final decision.

Two reviewers independently screened 5 446 records at the title-and-abstract level. Raw inter-rater agreement was 98.3 %(90 discrepancies), comfortably exceeding the 80 % benchmark commonly considered acceptable. Remaining conflicts were resolved through discussion, with a third reviewer arbitrating when consensus could not be reached.

Substantial concordance between reviewers was maintained throughout study selection, as evidenced by the 98.3 % raw agreement recorded at the initial title-and-abstract screening stage. Only 90 of the 5 446 records required discussion or third-party arbitration, and all disagreements were resolved within the same review session, ensuring an uninterrupted PRISMA flow.

Figure 1. PRISMA 2020 Flow Diagram Illustrating the Study Selection Process.

Quality Assessment

The quality assessment (QA) process refers to evaluating selected studies related to the main research topic to assist in the analysis and synthesis process. All authors meticulously monitored the progress of the review and task allocation of the project to ensure high-quality systematic literature reviews (SLRs) (Shea et al., 2007). Even though the selected studies were of high quality, Kitchenham's assessment method was still applied. The quality of the primary studies was evaluated using the following questions, and the results were categorized into three categories: low, medium, and high (Kitchenham & Charters, 2007). The following quality assessment questions were formulated for this purpose.

QA1: Do the research topics cover AR technology in special education?

QA2: Is the study context clearly described?

QA3: Are there detailed and accurate descriptions of how AR technology is implemented?

QA4: Is the research methodology properly executed?

QA5: Are the study findings pertinent to the goals of this SLR?

Each study was then rated as 1 for a study that fulfilled the quality criteria, 0.5 for one that partly met the criteria, and 0 if one had not met them. Studies with a total score above 3 were considered high quality, and those below 1 were rated low quality. Those with a total score between 1 and 3 were considered medium quality. The quality criteria found that ten articles were classified as high-quality studies. The remaining six articles were appraised for medium quality, as shown in Table 3.

The methodological quality of the included studies was independently assessed by two researchers. To ensure consistency and transparency in judgment, a detailed scoring rubric was developed a priori (as described in the text), defining the criteria for scoring 'Yes' (1), 'Partially' (0.5), and 'No' (0) for each quality question. Any discrepancies in the initial scores assigned by the two researchers were discussed until a final consensus score was reached for each article.

Table 3.Quality Appraisal Table

Note: QA, Quality Appraisal; PS, Primary Study.

Data Analysis

The data analysis process was conducted in two main stages. First, following the systematic extraction of data into a standardized form Table 4, a descriptive analysis was performed. This initial stage involved quantifying and summarizing key characteristics of the included studies, such as the frequency of publication by year, geographical distribution of research, participant demographics, and the types of research designs employed. This descriptive summary provides a broad overview of the research landscape.

In the second stage, a thematic synthesis was conducted to interpret the findings and address the more complex research questions. This involved a qualitative process of identifying, coding, and categorizing patterns and recurring themes across the 16 studies. Key themes related to the types of AR technologies used, the specific skills targeted, and the reported effectiveness and challenges were systematically collated. This approach allowed for a coherent synthesis of the diverse evidence, leading to the conclusions presented in this review.

Table 4.Data Extraction Summary

Findings/Results

Publication Distribution by Time Frame

Despite a slow start2010 to 2017 averaged fewer than one paper per year—research output has climbed steadily since 2018 (Figure 2), peaking at four publications in 2023. The brief dip in 2020 coincides with global COVID-19 disruptions, after which annual production rebounded. This “<1 → 4 per year” contrast underlines a six-fold growth in scholarly activity, confirming that AR in special education has rapidly transitioned from exploratory pilots to a sustained research stream.

Figure 2.The Distribution of Publications Over Time (n =16)

Geographical and Population Distribution

Regarding the geographical distribution of the studies (RQ2), the findings indicate a stark imbalance, with research being heavily concentrated in developing countries in Asia (Figure 3). Notably, Indonesia (n=4) and Malaysia (n=3) together account for nearly half (44%) of all included studies. In sharp contrast, there is a significant lack of research from North American and European contexts, suggesting a potential regional focus in the application and study of AR for special education.

In terms of population, the reviewed studies exclusively focused on K-12 educational settings,(Maas & Hughes, 2020). The participant ages ranged from 4 to 12 years, and sample sizes varied widely from 2 to 265 participants. Regarding gender, it is crucial to note that a substantial portion of the reviewed articles did not provide a gender breakdown (Shattuck et al., 2012). Therefore, the available data are insufficient to draw meaningful inferences about gender distribution.

Figure 3.The Study will be Distributed Across Different Countries

Research Designs and Methodologies Adopted

The analysis of research methodologies (RQ3) shows a clear preference for non-randomized, pre-post evaluation designs (Figure4).Quality appraisal shows that only 10 of the 16 studies met high methodological standards (Table 2), and none used truerandomisation.Quasi-experimental designs were the most frequently employed method (n=5). While such designs are practical for real-world settings where random assignment is difficult, they can present internal validity challenges such as selection bias and confounding variables (Shadishet al., 2002). The dominance of this approach, alongside a variety of other tailored methods used in the reviewed studies (Yıldız, 2020), highlights a notable absence ofmore rigorous designs, such as Randomized Controlled Trials (RCTs). This indicates a methodological gap in the literature, suggesting that while the field has established preliminary evidence of AR's effectiveness, it has yet to widely adopt more robust methods for confirming causal relationships.

Figure 4. Bar chart of design and methodology and field of education

Disability Categories, AR Technology Interventions, and Effectiveness

In response to RQ4, our synthesis reveals that AR interventions have been disproportionately applied to specific disability categories (Figure 5). A substantial body of research has focused on students with learning disabilities and vision impairments and those with autism spectrum disorder (ASD) (McMahon et al., 2016). The technology has also been explored for improving social skills in students with communication disabilities (Kellems & Morningstar, 2012). These results underscore the broad applicability of AR technology in improving learning outcomes and attention spans for students with various disabilities (Cheng & Tsai, 2013). Conversely, other conditions are significantly underrepresented, highlighting a critical research gap.

Regarding the AR technologies used (RQ5), the dominant approach, detailed in (Figure 6), involves interaction with 3D digital objects and game-based concepts (Baccaet al., 2014). These interventions are delivered on diverse hardware, including handheld equipment like smartphones and tablets and other wearable devices (Dunleavy et al., 2009). The prevalence of common handheld devices likely reflects current cost and accessibility constraints in educational settings.

In terms of effectiveness (RQ6), the reviewed studies consistently report positive outcomes across various learning domains (Figure 7). These include mathematics (improving problem-solving and conceptual understanding ), science education, physical education, and social skills training (improving communication and emotional management )(Chiang et al., 2014). This effectiveness stems from AR's ability to provide immersive and highly interactive learning experiences (Radu, 2014). It successfully enhances student motivation andvisual engagement (Cheng & Tsai, 2013; Shelton & Hedley, 2004) by making abstract concepts more concrete. Furthermore, the path to effectiveness is not without significant implementation barriers. For instance, the study by Kaimara(2023), while developing an inclusive transmedia application, found that teachers expressed considerable hesitancy towards adopting new digital educational games, a reluctance driven primarily by a lack of specific training and expertise. This finding is critical as it suggests that even technologically sound interventions may fail if the crucial human factors within the educational ecosystem are not addressed.A critical finding, however, is that while studies demonstrate skill acquisition within the intervention context, there is a notable lack of evidence for skill generalization to real-world, non-AR settings, a crucial area for future investigation.

Figure 5.The different disability categories.Note: ASD, Autism Spectrum Disorder; ID, Intellectual Dias- abilities; LD: Learning Disabilities; SN: Special Needs; DS: Down Syndrome; VI: Vision Impairments; Cerebral palsy: CP; Mild Mental Retardation: MMR; Motor Impairments; Prematurity; Maturational Delay; Rare Diseases; Communication; Language Delays

Figure 6.Distribution of AR Technology Interventions by Type.

Figure 7.Distribution of Learning Domains Addressed in Reviewed Studies.

Conclusion

This systematic review synthesized evidence on the application of AR technology in K-12 special education, revealing a field characterized by both significant potential and persistent challenges. The findings confirm that AR technology is not merely a supplementary tool but a transformative pedagogical approach that can create flexible and highly interactive learning environments tailored to individual student needs(Bowers et al., 2014). The escalating interest among educators in recent years, as identified in our temporal analysis of publications, appears to be driven by the unique capacity of AR technology to merge engaging, game-like elements with core academic content, thereby enhancing student motivation and participation(Yuen et al., 2011). However, while the benefits are widely reported across the included studies, our synthesis also highlights critical gaps and complexities that warrant deeper discussion. As noted in the literature, numerous challenges remain, and this review has systematically categorized them in terms of geographical disparities, methodological limitations, and a narrow focus on specific disability types(Akçayır & Akçayır, 2017). This discussion will, therefore, move beyond a simple reiteration of findings to interpret their implications for theory and practice.

A significant finding of this review is the pronounced concentration of research in developing Asian nations, with a majority of the included studies originating from Indonesia and Malaysia. This geographical clustering does not appear to be coincidental but rather the result of a confluence of factors. A primary driver may be strong governmental support for educational technology as a means to address specific and pressing national educational challenges, a trend previously observed in the region(Hussin et al., 2012). Furthermore, the increasing affordability and accessibility of mobile devices have made innovations with AR technology more feasible in these contexts. This economic viability, combined with large student populations that provide a robust base for empirical research and testing, creates a fertile ground for such studies. The active pursuit of international collaborations and funding may also play a role in fostering this research-intensive environment. Consequently, the leadership of these nations in the field could be interpreted as a strategic response to local needs, enabled by a combination of policy support, economic accessibility, and demographic advantages(Yuen et al., 2011).See also(Koumpouros, 2024)for recent Pacific-Asia policy incentives that accelerate immersive-learning adoption. 

This review’s finding that research samples are predominantly focused on the K-12 level is significant, as this stage represents a critical developmental window for students with special needs(Akçayır & Akçayır, 2017). Intervening during these formative years is essential for helping these students acquire foundational skills necessary for future independence, underscoring the importance of early and sustained support(Odom & Wolery, 2003). The application of AR technology during the K-12 stage is particularly strategic, as it not only provides a mechanism for effective early intervention but also has the potential to promote long-term educational equity and socioeconomic benefits(Cakir & Korkmaz, 2019). Moreover, the inherent multimedia and game-based elements of AR technology are exceptionally well-suited to the cognitive and motivational profiles of younger learners, effectively increasing their engagement and participation in educational activities(Wu et al., 2013). Therefore, the research community’s focus on this age group can be seen as a targeted effort to maximize developmental impact where it is most needed.

The review's findings also reveal a significant concentration of research on specific disability categories, namely learning disabilities and autism spectrum disorder, while other conditions remain underexplored. This focus may be pragmatic, as these groups often present cognitive and behavioral challenges that are well-suited for intervention with visually-based, structured tools like AR technology(Chen et al., 2015). Concurrently, our analysis shows a clear preference for quasi-experimental research designs. This methodological choice is logical for a field in its exploratory phase, as it allows researchers to effectively demonstrate intervention efficacy in real-world educational settings, even without the stringent controls of randomized trials. Taken together, the thematic focus on certain disabilities and the methodological inclination towards quasi-experimental designs suggest that the field of AR technology in special education is still solidifying its evidence base. Researchers appear to be prioritizing the demonstration of foundational effectiveness in receptive populations before expanding to more diverse disabilities and employing more complex research methodologies(Neely et al., 2024).

Researchers have found that the most prominent field shown in the research outcomes is "learning," as the advantages of AR technology lie in providing immersive, highly interactive, and game-based learning methods, which easily capture the attention of special education students.Besides the “learning” field, studies in the area of physical skills have also shown significant improvements in personal abilities among students with special needs(Baccaet al., 2014). Students with special needs benefit from sustainable training programs, as AR and VR technologies offer numerous advantages to physically disabled students in special education. One of the main benefits is the ability to simulate physical activities and practice skills in a controlled and safe environment, gradually learning and strengthening physical fitness(Y. Chen et al., 2018). AR technology can create virtual daily living scenarios, such as shopping, cooking, and cleaning, allowing special education students to practice these skills in a safe and controlled environment(Polyxeni Kaimara et al., 2021). AR technology can also simulate social situations, providing step-by-step guidance and immediate feedback and helping special education students cope with challenges in various social contexts while offering instant feedback for timely improvement (Schlosser & Sigafoos, 2006). Through the application of AR technology in different learning domains,children with special needs immediately receive essential and customized learning content according to their needs, allowing them to obtain instant feedback (Kamarainen et al., 2013).

The findings of this review, which indicate a research focus on learning disabilities and autism spectrum disorder, can be contextualized within broader epidemiological and educational data. Globally, the prevalence of disability is significant and growing, affecting an estimated 1.3 billion people, or 16% of the world's population(World Health Organization, 2022). Among children, the total number of individuals with disabilities is around 240 million, a figure that reflects rapid growth and makes the issue increasingly urgent(Krahn, 2011; Olusanya et al., 2022). More specifically, data from the U.S. National Center for Education Statistics for the 2022–2023 school year reveals that among the 7.5 million students receiving services under the Individuals with Disabilities Education Act (IDEA), "specific learning disabilities" constituted the largest single category, with approximately 32% of students classified as such(Anaby et al., 2013; Irwin et al., 2023). This thematic prevalence in the research literature, therefore, appears to directly mirror the reality in educational systems, where learning disabilities represent one of the most diagnosed and serviced conditions(Hallahanet al., 2014). This alignment suggests that the research community is directing its efforts with AR technology toward the largest segments of the special needs population, likely to maximize potential impact and address the most widespread challenges.

A central advantage of AR technology, evident across the reviewed studies, lies in its ability to provide immersive and highly interactive learning experiences through contextualized simulation. This is most prominent in the academic learning domain, where interventions frequently utilize game-based concepts and three-dimensional (3D) digital objects to make abstract concepts tangible and engaging(Baccaet al., 2014). The use of 3D objects, in particular, allows developers to layer rich, contextual information onto the real world, a common and effective application strategy(O'Shea, 2011). Beyond academics, the power of simulation extends to critical life skills. AR technology enables the creation of controlled, safe virtual environments where students can practice physical skills, daily living scenarios like shopping or cooking, and complex social situations, receiving immediate feedback for timely improvement(J. Chen et al., 2018; P. Kaimara et al., 2021; Schlosser & Sigafoos, 2006). To maximize the effectiveness of these applications, however, two practical considerations are crucial. First, the user interface and operational steps must be simplified to lower the cognitive load for students and reduce barriers to adoption for educators(Cai et al., 2014). Second, for these tools to be scalable and sustainable in educational settings, they must be developed to be both cost-effective and highly efficient.

Several limitations inherent in this review's methodology warrant careful consideration when interpreting the findings. First, potential publication bias is a significant concern; studies demonstrating positive or statistically significant effects of AR technology are more likely to be published than those with null or negative findings. For instance, an early-phase study that found AR technology did not improve outcomes for a specific task might not have been submitted for publication, leading to a potential overestimation of the technology's effectiveness in the available literature. Second, the search strategy was restricted to four databases and English-language articles. Consequently, significant research from non-English speaking regions—for example, innovative work from research groups in Germany or South Korea—may have been excluded, potentially overlooking culturally specific applications or alternative pedagogical designs. Finally, our focus on peer-reviewed articles meant excluding "gray literature." This means that valuable preliminary findings or practical implementation guides shared at specialized conferences, such as the IEEE International Symposium on Mixed and Augmented Reality (ISMAR), might not be represented in this synthesis, thus providing a potentially incomplete picture of the field's most current state.

Despite these limitations, this review illuminates several critical pathways for future inquiry and practical application. To move the field forward, we propose the following concrete actions:

Expand the Scope of Disabilities and Formulate Targeted Hypotheses: The current research landscape is heavily skewed towards autism and learning disabilities. Future work must systematically explore other populations. For example, researchers could investigate a specific hypothesis such as: "For students with visual impairments, AR technology applications that use high-contrast visual overlays and dynamic auditory cues will significantly improve spatial navigation skills in unfamiliar school environments compared to traditional orientation and mobility training." This approach moves beyond general exploration to testing specific, measurable outcomes for underserved groups.

Conduct Longitudinal Studies on Usability and Cost-Effectiveness: The challenges of hardware cost and complexity cannot be understood through short-term studies alone. There is a pressing need for longitudinal research that tracks the usability and total cost of ownership of AR technology over an entire school year. Such a study could, for example, measure not only the initial hardware investment but also the crucial factors of teacher training time, technical support hours, and software maintenance costs. This would provide a far more realistic economic and practical picture for school administrators and policymakers, moving beyond initial sticker shock to understand long-term value and sustainability.

Implement "Co-Design" Models for Pedagogical Soundness: To ensure AR technology is genuinely effective and user-friendly, the design process must be collaborative. We strongly recommend a shift towards participatory "Co-design" models where developers, researchers, special education teachers, and students work as partnersfrom day one. For instance, a development team could host bi-weekly workshops where students with specific needs interact with early prototypes. Their direct feedback on interface complexity or cognitive load would then immediately inform the next design iteration, ensuring the final product is not just technologically advanced but also pedagogically grounded and practically useful in a real, and often resource-constrained, classroom setting.

The completed PRISMA 2020 checklist is provided in supplementary file.

This systematic review confirms that AR technology is a rapidly emerging and highly promising field in K-12 special education, yet its current state is best characterized by a significant gap between isolated successes and widespread, evidence-based practice. The principal contribution of this review lies in its systematic mapping of this landscape. By synthesizing disparate studies, we have provided a novel, integrated insight: the field remains in a foundational, exploratory phase, evidenced by its concentration of research in specific geographical regions, its focus on a narrow range of disabilities like autism and learning disabilities, and its reliance on quasi-experimental designs to establish preliminary efficacy.

The critical challenge highlighted by our findings is not merely a matter of technological refinement, but one of pedagogical translation. While studies consistently demonstrate skill acquisition within controlled AR environments, there is a notable lack of evidence for the generalization of these skills to real-world settings. This "transfer problem" represents the next frontier for researchers. We hypothesize that future AR technology interventions which explicitly incorporate principles of community-based instruction and utilize adaptive algorithms to fade virtual scaffolds over time will yield significantly higher rates of skill generalization.

This review advances the field by shifting the discourse from whether AR technology works to a more nuanced discussion of how, for whom, and under what conditions it can be most effective and equitable. It provides a critical evidence base and a strategic roadmap that calls for a more methodologically rigorous, inclusive, and collaborative research ecosystem. Ultimately, the goal is to move beyond proof-of-concept and harness AR technology as an integral tool to create authentic, sustainable, and truly personalized learning opportunities that empower all students with special needs to thrive.

Although Cohen’s κ could not be calculated because the original screening software (Rayyan) log was archived without the necessary dual-code export, the very high raw agreement (98.3 %) indicates that reviewer disagreement introduced minimal selection bias. This constraint stemmed from project timelines and institutional data-retention policies rather than an a priori methodological choice. Future reviews from our team will preserve screening logs in open formats (e.g., CSV) to enable κ orGwet’sAC1 statistics and thereby enhance methodological transparency.

Statements and Declarations

The authors declare no potential conflicts of interest concerning the research and authorship of this study. In addition, this research received no specific grants from any funding agency in the public, commercial, or not-for-profit sectors.

The data supporting this study's findings are available from the corresponding author upon reasonable request.

No acknowledgements.

Conflict of Interest

The authors declare no conflict of interest.

Funding

No funding was received for this study.

Generative AI Statement

As the authors of this work, we used the AI tool ChatGPT (OpenAI) to assist in refining English grammar, rephrasing sentences, and organizing the structure of the abstract and discussion sections. The AI tool was employed as a writing assistant to improve clarity and coherence. After using this tool, we carefully reviewed and revised all content to ensure accuracy and academic integrity. We, as the authors, take full responsibility for the final version and content of our published work.

Authorship Contribution Statement

Chang: Conceptualization, methodology design, data collection, data analysis and interpretation, drafting the manuscript, and supplementary information gathering. Chao: Supervision, conceptual guidance, directional input, critical review and editing, project administration, and final approval. Yang: Statistical analysis, visualization, reviewing and editing.Semwaiko:Contentadvising, supporting data acquisition, and reviewing.