Exploring Artificial Intelligence in Inclusive Education: A Systematic Review of Empirical Studies

Abstract

This systematic review synthesizes empirical evidence on the role of Artificial Intelligence (AI) in inclusive education. The review aimed to examine (1) the impact of AI technologies on learning outcomes and engagement among students with special needs, (2) barriers and enablers influencing AI adoption by educators, and (3) the role of theoretical frameworks in guiding AI-based interventions. A comprehensive search was conducted in Scopus, Web of Science, DOAJ, and Google Scholar for English-language, peer-reviewed studies published between 2020 and 2025. Sixteen (16) studies met the inclusion criteria and were appraised using the Mixed Methods Appraisal Tool (MMAT). Findings indicate that AI tools enhance personalization, accessibility, and engagement, particularly for learners with disabilities, while barriers such as infrastructure gaps and low digital literacy persist. Enablers include institutional support and teacher training, though theoretical frameworks were inconsistently applied. Limitations include the exclusion of grey literature and reliance on short-term studies. AI can advance inclusive education when integrated with ethical, pedagogical, and institutional strategies, while future research should prioritize longitudinal, theory-driven, and culturally responsive models.

1. Introduction

1.1. Review Context

As a cornerstone for equitable learning, inclusive education has gained worldwide recognition through the UNESCO World Conference on Special Needs Education held in 1994 at Salamanca, Spain. The “Salamanca Statement and Framework for Action”, a byproduct of the conference, has called governments to adopt inclusive educational policies and practices. Over 90 countries adopted the call and affirmed that schools should accommodate all children, regardless of their physical, intellectual, social, or emotional differences [1]. This declaration marked a turning point, emphasizing the importance of inclusive practices, especially for children with special needs. International frameworks, such as the UN Convention on the Rights of Persons with Disabilities [2] and Sustainable Development Goal 4 [3], followed suit reinforcing inclusive education as essential for achieving universal access and social justice in learning.

Since the adoption of the “Salamanca Statement”, the promotion of inclusive education has expanded to encompass diverse areas where students with special needs are given equitable opportunities to learn and thrive. From academic instruction to social participation, inclusive classrooms now integrate tailored strategies, assistive technologies, and individualized support systems that foster meaningful engagement. As highlighted by Resilient Educator [4], inclusive education ensures that students with disabilities are not only present in mainstream settings but are actively supported to participate fully and succeed academically and socially. This outcome became a standpoint for many education stakeholders to continuously seek avenues where students with special needs can maximize all available supports to promote equitable learning. This includes the maximization of technology, especially in the utilization of artificial intelligence.

Artificial intelligence (AI) began influencing education in the late 20th century, initially through intelligent tutoring systems and adaptive learning technologies. Over time, its role expanded to include personalized instruction and data-driven decision-making. As Haenlein and Kaplan [5] note, AI’s evolution reflects a growing interest in enhancing educational processes through intelligent, responsive systems that support diverse learning needs. With this advent, the need to create policies on its use in education became the call of many stakeholders. Since then, international policies on artificial intelligence (AI) in education have significantly enhanced instructional delivery and student learning. UNESCO’s guidance promotes ethical, human-centered AI use to support personalized learning and teacher empowerment [6]. UNICEF emphasizes child-centered AI that ensures fairness, safety, and inclusion in digital learning environments [7]. The European Commission’s ethical guidelines help educators critically and effectively integrate AI into classrooms [8]. Meanwhile, the OECD highlights AI’s role in adaptive learning and digital equity across education systems [9,10]. Collectively, these frameworks foster responsible innovation that improves educational access, engagement, and outcomes globally.

Though AI has been recognized for its significant contributions to education, the call for equitable use was also pronounced. More specifically, the promotion of inclusive education must encompass the equitable use of artificial intelligence (AI) to support students with special needs. AI offers transformative potential to personalize learning, enhance accessibility, and foster autonomy among diverse learners [6]. For instance, technologies such as adaptive content, speech-to-text tools, and AI-driven learning platforms can bridge educational gaps and empower students with disabilities. As emphasized by UNESCO, inclusion and equity must guide AI integration to prevent deepening existing divides. Ensuring that all learners benefit from technological advancements is essential to uphold the principles of inclusive education and achieve global educational equity [6].

Despite the growing integration of artificial intelligence (AI) in education, there remains limited empirical evidence on its actual impact on learning outcomes and engagement among students with special educational needs in inclusive settings. While AI promises personalization and accessibility, most existing studies focus on general education contexts, leaving a gap in understanding its effectiveness for learners who face unique challenges. The first research question addresses this gap by examining how AI technologies influence academic performance and engagement for students in inclusive classrooms. Furthermore, although AI adoption in education is widely advocated, there is insufficient clarity on the key barriers and enablers that affect educators’ ability to implement AI tools effectively in inclusive environments. Issues such as infrastructure limitations, teacher digital literacy, and ethical concerns are acknowledged but not comprehensively analyzed in relation to inclusive education. Research question two responds to this gap by identifying and synthesizing these factors to inform strategies for successful AI integration. Lastly, while theoretical frameworks like Self-Determination Theory (SDT), Technology Acceptance Model (TAM/TAM2), Unified Theory of Acceptance and Use of Technology (UTAUT), and Technological Pedagogical Content Knowledge (TPACK) are recognized as valuable for guiding technology adoption, their application in shaping AI-based interventions for inclusive education remains underexplored. Current literature often overlooks how these models can inform design, implementation, and evaluation processes to ensure pedagogical alignment and inclusivity. Hence, research question three addresses this gap by investigating the role of these frameworks in creating effective AI solutions for diverse learners. By systematically reviewing empirical studies across these dimensions, this research aims to provide a comprehensive and theory-informed perspective that will help educators, policymakers, and developers better address these gaps and advance equitable, inclusive AI-driven education.

This paper is organized into five sections to provide a clear and comprehensive discussion. Section 1 (Introduction) establishes the context of inclusive education and the growing role of AI, outlining global policy frameworks, equity concerns, and the rationale for this review. It also presents the research questions and problem statements that guide the study. Section 2 (Related Works) synthesizes prior research on AI in inclusive education, identifying trends, barriers, enablers, and theoretical foundations, which frame the gaps this review addresses. Section 3 (Methods) explains the systematic approach, including search strategy, inclusion and exclusion criteria, quality appraisal, and data analysis procedures, ensuring transparency and rigor. Section 4 (Results) organizes findings from 16 empirical studies around the research questions: AI’s impact on learning outcomes and engagement (RQ1), barriers and enablers influencing adoption (RQ2), and the role of theoretical frameworks in guiding AI-based interventions (RQ3). Section 5 (Discussion) interprets these findings, highlights implications for educators, policymakers, and developers, and suggests directions for future research. The paper concludes by emphasizing theory-informed, ethically grounded, and context-sensitive strategies for integrating AI in inclusive education. Together, these sections offer actionable insights for advancing equitable, AI-enhanced learning environments.

1.2. Impact of AI on Learning Outcomes and Engagement Among Students with Special Needs

Recent research has shown that Artificial Intelligence (AI) can positively influence learning outcomes for students with diverse educational needs. For instance, Melo-López et al. [11] found that AI technologies, such as intelligent tutoring systems and adaptive learning platforms, significantly improved accessibility and personalization. These tools enabled students with disabilities to engage more effectively with learning materials through features like speech-to-text, image descriptions, and real-time feedback, resulting in increased academic performance and motivation. Similarly, Ayobami et al. [12] reported that AI-driven adaptive learning platforms enhanced cognitive engagement and retention among students with special needs. Their study highlighted how AI systems could analyze individual learning patterns and adjust instructional content accordingly, allowing students to learn at their own pace and receive targeted support. This personalized approach was particularly beneficial for learners with attention deficits and processing disorders, as it reduced frustration and improved focus. However, not all studies have reported successful outcomes. Panjwani-Charania and Zhai [13] reviewed sixteen studies on AI applications for students with learning disabilities and found that many interventions prioritized diagnosis over instructional support. They noted a lack of empirical evidence demonstrating consistent improvements in academic achievement, especially for students with complex or multiple disabilities. Furthermore, the limited number of studies focusing on school-age children and the narrow scope of existing AI tools suggest that the technology’s potential in inclusive education is still underdeveloped. With this to say, it is therefore imperative to scope empirical studies that show concrete results on AI’s impact towards learning outcomes and meaningful engagements of students with special needs.

1.3. Problem Statement

While existing literature highlights AI’s potential to enhance learning outcomes and engagement among students with special needs, findings remain fragmented. Empirical studies such as those by Melo-López et al. [11] and Ayobami et al. [12] demonstrate improvements in personalization and cognitive engagement, yet Panjwani-Charania and Zhai [13] reveal gaps in instructional support and limited evidence for students with complex disabilities. Similarly, the adoption of AI tools is influenced by multifaceted barriers and enablers. Studies underscore the importance of teacher training, ethical literacy, and institutional support, yet the lack of consistent empirical data hinders scalable implementation. Lastly, theoretical frameworks, such as UDL, TAM2, SDT, and UTAUT, are pivotal in guiding AI design and evaluation. Empirical evidence shows that theory-informed interventions yield better engagement and equity, yet many implementations lack such grounding. Realizing the above-stated gaps, a good systematic review of empirical studies can clarify each issue’s true impact. It will help identify patterns in educator readiness and institutional dynamics, providing actionable strategies for effective AI integration. By consolidating empirical findings, the results can offer robust, data-driven insights into AI’s efficacy across varied learner profiles.

1.4. Research Questions

Building upon the challenges and gaps identified in the existing literature, this systematic review is strategically guided by the following research questions. These questions aim to critically explore the nuanced impact of AI technologies in inclusive education, address the multifaceted barriers and enablers affecting their adoption, and examine the theoretical foundations that inform their design and implementation. By anchoring the inquiry in empirical evidence, the review seeks to generate a comprehensive and data-driven understanding of how AI can be effectively leveraged to support diverse learners in inclusive settings. Below are the specific guide questions for this systematic review:

RQ1: 

What is the impact of artificial intelligence (AI) technologies on learning outcomes and engagement among students with diverse educational needs in inclusive settings?

RQ2: 

What are the key barriers and enablers influencing the adoption and effective use of AI tools by educators in inclusive education environments?

RQ3: 

How do theoretical frameworks (e.g., UDL, TAM2, SDT, UTAUT) inform the design, implementation, and evaluation of AI-based interventions in inclusive education?

1.5. Purpose and Significance

The primary purpose of this systematic review is to gather and synthesize empirical evidence to support the development of data-driven recommendations for the integration of Artificial Intelligence (AI) in inclusive education. By focusing on empirical studies, the review aims to ensure that the resulting insights and suggestions are grounded in actual classroom experiences, measurable outcomes, and validated interventions. This evidence-based approach strengthens the credibility and applicability of the recommendations, making them more relevant and impactful for real-world educational settings.

Beyond informing recommendations, the review also seeks to identify patterns, gaps, and emerging trends in the use of AI technologies among students with diverse educational needs. It will explore how AI affects learning outcomes and engagement, what barriers and enablers influence its adoption, and how theoretical frameworks guide its design and implementation. These insights will contribute to a deeper understanding of how AI can be effectively and ethically integrated into inclusive education.

The significance of this review extends to a wide range of stakeholders. Policy makers can use the findings to craft inclusive and forward-thinking educational policies. School administrators may benefit from guidance on infrastructure and training investments. Educators will gain insights into best practices and professional development needs, while technology developers can align their innovations with pedagogical and ethical standards. More so, advocacy groups will find the review valuable for promoting equity and inclusion in education. Ultimately, this review serves as a strategic tool to empower stakeholders in making informed decisions that advance inclusive, AI-enhanced learning environments.

2. Related Works

2.1. Barriers and Enablers Influencing Adoption and Use of AI Tools

The adoption and use of AI tools in education are shaped by a complex interplay of barriers and enablers. In a systematic review of 30 empirical studies, it was found that AI enhances personalization, engagement, and academic performance across disciplines [14]. However, barriers such as a lack of AI literacy, ethical concerns, and institutional resistance persist. For instance, students and faculty often lack critical evaluation skills and ethical frameworks to use AI responsibly [14]. Similarly, McGehee [15] identified predictors of educator acceptance of AI, including self-efficacy, perceived usefulness, and technological complexity. Barriers such as cost, time constraints, and pedagogical shifts were found to amplify resistance, while enablers included institutional support and targeted training. These findings underscore the need for systemic strategies to address both technical and human factors in AI integration.

Teacher training emerges as a pivotal enabler in the effective use of AI tools, especially in diverse classrooms. Daher [16] argues that AI literacy must go beyond technical skills to include ethical understanding and critical engagement, warning that without proper training, educators risk deepening the digital divide. Roshan et al. [17] found that only 5% of surveyed teachers felt confident using AI tools, with 70% lacking professional development. Their study revealed a significant correlation between training and increased confidence, emphasizing the importance of continuous, targeted professional development. The EQUIP Framework proposed by Daher [16] offers a structured approach to empower educators through ethical governance, collaborative partnerships, and implementation readiness. These insights affirm that teacher preparedness is not optional—it is essential for equitable and effective AI integration.

In conclusion, while AI holds transformative potential for education, its benefits cannot be fully realized unless the barriers to adoption are systematically addressed. Without adequate training, ethical guidance, and institutional support, AI tools risk being underutilized or misused. Therefore, maximizing AI’s impact in education demands a proactive commitment to dismantling these barriers and empowering educators with the skills and confidence to lead in an AI-enhanced learning environment.

2.2. Role of Theoretical Frameworks in the Design of AI-Based Intervention

The integration of theoretical frameworks in the design of AI-based interventions is essential for ensuring pedagogical coherence, inclusivity, and effectiveness. Empirical studies have demonstrated that AI tools, when grounded in robust educational theories, significantly enhance learning outcomes. For instance, Bognár et al. [18] applied classical educational theories such as self-efficacy and self-regulation to evaluate student engagement in AI-enhanced environments, revealing that AI tools like ChatGPT-3.5 can amplify autonomy and goal setting when aligned with theoretical constructs. Similarly, Lubbe et al. [19] emphasized the role of constructivist learning theory, Bloom’s Taxonomy, and adaptive learning theory in designing AI-powered curricula. Their study at Vietnam National University showed a 40% increase in student engagement and a 25% improvement in test scores when AI interventions were guided by these frameworks.

Theoretical models not only inform the design but also guide the implementation and evaluation of AI-based interventions, especially in diverse classrooms. Xia et al. [20] utilized Self-Determination Theory (SDT) to structure inclusive AI education, finding that students—regardless of gender or achievement level—benefited equally when their psychological needs were supported through AI-enhanced instruction. This underscores the importance of theory in ensuring equity and responsiveness in AI applications. Without a theoretical foundation, AI risks becoming a purely technical tool, disconnected from the nuanced realities of learner diversity. Theories provide the scaffolding for ethical use, adaptive personalization, and meaningful engagement. As such, educators and developers must prioritize theoretical alignment to ensure AI interventions are not only innovative but also inclusive and pedagogically sound. It is therefore interesting to note that, in the realm of inclusive education, it is worth remembering the adage that: “Though AI is born of algorithms and data, its success in education still depends on the wisdom of theory.”

3. Methods

3.1. Search Strategy

A rigorous and systematic search strategy was implemented to ensure comprehensive identification of relevant studies on artificial intelligence (AI) in inclusive education. The initial search employed the phrase “ARTIFICIAL INTELLIGENCE IN INCLUSIVE EDUCATION” across major academic databases, including Scopus, Web of Science (WOS), Directory of Open Access Journals (DOAJ), and Google Scholar. This preliminary search generated a broad pool of potentially relevant articles. To refine the results and improve precision, an expanded search was conducted using multiple keyword combinations, integrating Boolean operators (e.g., AND) and forced inclusion symbols (e.g., +) to capture studies explicitly addressing AI applications within inclusive educational contexts. Filters were applied to include only peer-reviewed articles published between 2020 and 2025, written in English, and participants who have experienced the application of artificial intelligence in inclusive education, who can be either learners, teachers or educational administrators. The final search was conducted last September 8, 2025. Duplicates were removed, and titles and abstracts were screened for relevance before full-text review. Ultimately, 16 studies met the inclusion criteria and were subjected to quality appraisal using the Mixed Methods Appraisal Tool (MMAT, Manufacturer: National Institute for Health and Care Excellence & Scottish Intercollegiate Guidelines Network, City: London, Country: England), ensuring methodological rigor and transparency.

• ARTIFICIAL INTELLIGENCE + INCLUSIVE EDUCATION.
• ARTIFICIAL INTELLIGENCE AND INCLUSIVE EDUCATION.
• INTELLIGENT TUTORING SYSTEM IN INCLUSIVE EDUCAITON.
• ARTIFICIAL INTELLIGENCE FOR STUDENTS WITH SPECIAL NEEDS.
• ARTIFICIAL INTELLIGENCE + SPECIAL STUDENTS.
• AI + PERSONALIZED LEARNING + STUDENTS WITH DISABILITY.

The decision to focus exclusively on empirical studies was deliberate. While conceptual and theoretical reviews provide valuable insights into frameworks and policy directions, they often lack evidence of real-world implementation and measurable outcomes. Empirical research offers data-driven perspectives on how AI tools function in practice, their impact on learning outcomes, and the contextual factors influencing adoption. By prioritizing studies grounded in classroom experiences and validated interventions, this review aims to generate actionable recommendations for educators and policymakers. This contrasts with existing theoretical reviews, which, although informative, cannot fully capture the complexities of implementation, equity challenges, and pedagogical alignment in diverse educational settings. Thus, emphasizing empirical evidence strengthens the credibility and applicability of the findings for advancing inclusive, AI-enhanced education.

3.2. Inclusion and Exclusion Criteria

Using a PRISMA Flowchart (Figure 1), the identification phase was guided by predefined inclusion and exclusion criteria. Studies were included if they focused on AI applications in inclusive education, addressed accessibility or equity, were empirical in nature, published between 2020–2025, written in English, and indexed in reputable databases. Conversely, studies were excluded if they lacked relevance to inclusive education, were non-peer-reviewed, published before 2020, or were written in languages other than English. These criteria ensured methodological rigor and relevance to the review’s objectives.

Table 1. Inclusion criteria.

No.	Filters	Inclusion Criteria
1	Objective	Must focus on the application of AI technologies in inclusive education. Should address accessibility, personalization, equity, or support for learners with special needs.
2	Study Design	Empirical studies (quantitative, qualitative, mixed methods).
3	Publication Type	Original peer-reviewed journal articles or conference papers. Must include methodological transparency.
4	Indexation	Indexed by either Scopus, Web of Science or Google Scholar.
5	Time Frame	Published between 2020 and 2025 to reflect recent developments in AI and inclusive education.
6	Language	Published in English.

and

Table 2. Exclusion criteria.

No.	Filters	Exclusion Criteria
1	Objective	Studies that do not focus on AI applications in inclusive education. Papers focused solely on mainstream education without addressing inclusion or diversity.
2	Study Design	Non-empirical papers (theoretical papers, literature reviews, methodological papers, conceptual papers). Editorials, philosophical or critical essays, opinion pieces, blog posts, or theoretical essays.
3	Publication Type	Publications from non-peer-reviewed sources or predatory journals.
4	Indexation	Studies not indexed in recognized academic databases such as Scopus, Web of Science, ERIC, etc.
5	Time Frame	Studies covered from 2019 and below.
6	Language	Studies with languages other than English.

below present the specific inclusion and exclusion parameters.

In the identification phase, a total of 28 records were retrieved from three databases: Web of Science (n = 7), Scopus (n = 15), and Directory of Open Access Journals (DOAJ) (n = 6), while 34 studies were retrieved from other methods (Google Scholar). This multi-source strategy ensured a broad and inclusive search scope. Following the identification process, 5 duplicates were removed from the database-borne studies, leaving 23 unique articles, while 34 studies sourced from other methods were left for screening.

In the screening phase, 23 studies taken from the three databases were screened, while 34 articles were included from Google Scholar (Other methods). All 57 articles were retrieved. Out of the 23 database-borne studies assessed for eligibility, 13 were excluded as they are all non-empirical, aligning with the inclusion criteria that prioritized evidence-based research, while 18 duplicate studies and 10 non-empirical papers were also excluded from the 34 studies taken from Google Scholar.

Ultimately, 16 studies were assessed for eligibility, thus included in the review. These studies were comprised of the following: Scopus (n = 5), Web of Science (n = 3), Google Scholar (n = 6), and DOAJ (n = 2). This refined pool represents high-quality, relevant literature suitable for in-depth analysis. Overall, the flowchart demonstrates a transparent and rigorous selection process, ensuring that the final dataset aligns with the review’s goals and maintains scholarly integrity.

3.3. Data Extraction

Based on the extracted data from the 16 reviewed articles, the data extraction process was conducted systematically to ensure consistency and transparency. A standardized data extraction form was developed to capture key elements from each study, including author and year, title, publication type, language, objectives, and study design. The process involved identifying study characteristics such as research methodology (qualitative, quantitative, mixed methods), participant demographics, data collection instruments, and analysis techniques. The author independently extracted the data. The extracted data revealed diverse approaches to investigating the role of artificial intelligence in inclusive education, ranging from case studies and randomized controlled trials to cross-sectional surveys. Common themes included the use of AI for personalized learning, support for students with disabilities, and teacher readiness. This rigorous extraction process provided a reliable foundation for synthesizing findings and drawing meaningful conclusions about AI’s impact on inclusive education. Complete details of the extracted data can be found in Appendix A of this manuscript.

3.4. Quality Appraisal

The quality appraisal of the 16 included studies (

Table 3. Quality appraisal report summary of 16 articles subjected for review.

No.	Author/Year	Methodological Design	Screening Questions	MMAT Criteria Compliance	Inclusion Decision
1	Garg et al. (2024) [21]	Qualitative	2/2	5/5	Included
2	Toyokawa et al. (2023) [22]	Qualitative	2/2	5/5	Included
3	Irawan et al. (2025) [23]	Qualitative	2/2	5/5	Included
4	Reyes et al. (2024) [24]	Qualitative	2/2	5/5	Included
5	El Naggar et al. (2024) [25]	Qualitative	2/2	5/5	Included
6	Kruger (2024) [26]	Qualitative	2/2	5/5	Included
7	Xia et al. (2024) [20]	Quantitative experimental	2/2	4/5	Included
8	Adigun et al. (2025) [27]	Quantitative descriptive	2/2	4/5	Included
9	Alyoussef et al. (2025) [28]	Quantitative descriptive	2/2	4/5	Included
10	Beirat et al. (2024) [29]	Quantitative descriptive	2/2	3/5	Included
11	Cai et al. (2022) [30]	Quantitative quasi-experimental	2/2	5/5	Included
12	Naseer et al. (2025) [31]	Quantitative descriptive	2/2	4/5	Included
13	Coy et al. (2024) [32]	Mixed method	2/2	4/5	Included
14	Amouri et al. (2025) [33]	Mixed method	2/2	5/5	Included
15	Lunavictoria et al. (2024) [34]	Mixed Method	2/2	4/5	Included
16	Alsolami (2025) [35]	Randomized Controlled Trial	2/2	5/5	Included

) demonstrates a rigorous and methodologically sound evaluation process. All studies passed the initial screening questions (2/2), confirming relevance and clarity in research aims and design. The Mixed Methods Appraisal Tool (MMAT) was employed as the basis for assessment, given its suitability for appraising diverse methodological designs, including qualitative, quantitative, mixed methods, and randomized controlled trials. Most studies achieved full compliance with MMAT criteria, scoring 5/5, while others scored 4/5 or 3/5 but still met the minimum quality threshold for inclusion. This reflects an inclusive appraisal approach that values methodological diversity while ensuring rigor and transparency.

The inclusion of all 16 studies underscores the reviewers’ commitment to methodological rigor and transparency, aligning with best practices in systematic reviews [36]. The MMAT is particularly suited for appraising complex educational research, allowing for nuanced evaluation across varied designs. The consistent application of MMAT criteria and the inclusion of studies with strong empirical findings support the credibility and comprehensiveness of the review. This approach enhances the reliability of synthesized insights and ensures that the review captures a wide spectrum of evidence on AI’s role in inclusive education.

To strengthen reliability, an inter-rater review process was implemented. Each article was initially evaluated by a primary reviewer, after which the appraisal results were independently checked by a second reviewer. Discrepancies in scoring or interpretation were resolved through discussion and consensus, ensuring consistency and minimizing bias. This dual-review process enhanced the credibility of the appraisal and aligned with best practices in systematic reviews. By combining MMAT-based evaluation with inter-rater rigor, the review ensured that synthesized insights were grounded in high-quality evidence, capturing a comprehensive and trustworthy picture of AI’s role in inclusive education.

3.5. Data Analysis

This review employed Qualitative Content Analysis (QCA) as the principal analytical strategy to synthesize the findings of the 16 articles included in the study. QCA is a systematic and rule-governed method for interpreting textual data, allowing for both inductive and deductive coding approaches [37]. It is particularly appropriate for reviews that aim to explore complex educational phenomena, such as the impact of artificial intelligence (AI) technologies on learning outcomes and engagement in inclusive settings, across diverse methodological designs. The analysis therefore followed a structured QCA framework consisting of three phases [38] as shown in

Table 4. Three phases of the QCA framework analysis.

Phase	Phase Title	Description
1	Preparation	Involves identifying and extracting relevant textual segments from each article.
2	Organization	A coding frame will be developed based on the research question and applied to the data.
3	Reporting	Synthesize and present emergent themes and patterns across studies.

.

Given the methodological diversity of the included studies, the QCA framework was adapted to accommodate different data types. Below are the analytical procedures for articles belonging to a specific methodological design. Supporting details are presented in

Table 5. Qualitative Content Analysis (QCA) application matrix.

Study Type	QCA Application	Data Segments Analyzed	Integration Strategy
Qualitative Studies	Full-text coding using QCA framework; themes extracted inductively	Results, discussions, conclusions	Themes categorized under parent and child nodes; patterns synthesized across studies
Quantitative Studies	Narrative sections analyzed using QCA; statistical data summarized separately	Discussions, interpretations	Interpretive insights coded; statistical findings mapped to relevant thematic categories
Mixed-Methods Studies	Qualitative components coded via QCA; quantitative findings descriptively summarized	Results (qualitative and quantitative), discussions	Dual analysis: qualitative data coded; quantitative data linked to emergent themes
RCT Study	Narrative interpretations included in QCA; statistical outcomes descriptively summarized	Discussion, results (narrative and statistical)	Triangulation: narrative themes coded; statistical outcomes aligned with thematic patterns

:

• For quantitative studies, while numerical data was not analyzed using QCA, narrative sections, such as discussions and interpretations, were examined to identify author-reported implications and contextual meanings [33]. This approach is consistent with practices in mixed-method synthesis, where qualitative insights are extracted from quantitative papers to support thematic integration.
• For qualitative studies, full-text analysis was conducted, focusing on results, discussions, and conclusions to extract themes related to AI’s influence on student engagement and learning outcomes. These were coded inductively and grouped into thematic categories.
• Mixed-methods studies were analyzed by separating qualitative and quantitative components: the qualitative parts were coded using QCA, while the quantitative findings were summarized and mapped onto emerging themes.
• For the randomized controlled trial (RCT), narrative interpretations of results were included in the QCA, while statistical outcomes were descriptively interpreted to complement the thematic findings.

This integrative approach ensured a comprehensive synthesis of evidence across methodological boundaries. By applying QCA to qualitative and narrative data, and complementing it with descriptive summaries of quantitative results, the review provided a nuanced understanding of how AI technologies affect learning and engagement among students with diverse educational needs. The method was supported by literature emphasizing QCA’s flexibility and its increasing use in systematic reviews to explore intervention complexity [37,38,39]. Furthermore, the alignment between QCA and thematic coding practices, such as those used in NVivo, reinforces the rigor and transparency of the analytical process [40].

To ensure methodological rigor in qualitative content analysis, the data in

Table 6. Articles aligned in addressing each research question.

Author	Method	Variable/s Addressed in RQ1 (Impact of AI on Learning Outcomes and Engagement)
Garg et al. (2024) [21]	Qualitative	Learning outcomes, Engagement, Motivation
El Naggar et al. (2024) [25]	Qualitative	Learning outcomes, Engagement
Lunavictoria et al. (2024) [34]	Qualitative	Learning outcomes, Engagement, Motivation
Kruger (2024) [26]	Qualitative	Learning outcomes, Engagement, Motivation
Cai et al. (2022) [30]	Qualitative	Learning outcomes, Engagement
Xia et al. (2022) [20]	Qualitative	Learning outcomes, Engagement, Motivation
Alsolami (2025) [35]	Quantitative RCT	Learning outcomes, Engagement, Motivation
Naseer et al. (2025) [31]	Quantitative	Learning outcomes, Engagement, Motivation
Author	Method	Variable/s Addressed in RQ2 (Barriers and Enablers to AI Adoption)
El Naggar et al. (2024) [25]	Qualitative	Barriers: lack of AI literacy, ethical concerns Enablers: teacher openness, institutional vision
Toyokawa et al. (2023) [22]	Qualitative (Case Study)	Barriers: limited personalization, lack of inclusive design Enablers: data-enhanced reading tools
Irawan et al. (2025) [23]	Qualitative (Case Study)	Barriers: infrastructure, teacher readiness, policy gaps Enablers: personalization, assistive tools, leadership support
Reyes et al. (2024) [24]	Qualitative (Case Study)	Barriers: lack of training, invisibility of marginalized learners Enablers: metacognitive assessment, AI for accessibility
Amouri et al. (2025) [33]	Mixed Methods	Barriers: low ease of use, lack of experience Enablers: perceived usefulness, institutional support
Adigun et al. (2025) [27]	Quantitative (UTAUT + SEM)	Barriers: low technological self-efficacy, poor facilitating conditions Enablers: effort expectancy, social influence
Alyoussef et al. (2025) [28]	Quantitative (TAM + SEM)	Barriers: lack of trust, low familiarity, infrastructure gaps Enablers: perceived usefulness, ease of use, engagement efficacy
Beirat et al. (2025) [29]	Quantitative (Survey)	Barriers: lack of training, infrastructure, program design Enablers: higher qualifications, private school support
Author	Method	Variable/s Addressed in RQ3 (How Theory Inform the Design of AI-Based Interventions)
Amouri et al. (2025) [33]	Mixed Methods	AI adoption factors: perceived usefulness, ease of use, experience, accessibility
Adigun et al. (2025) [27]	Quantitative	AI adoption predictors: effort expectancy, social influence, facilitating conditions
Alyoussef et al. (2025) [28]	Quantitative	AI adoption: trust, familiarity, engagement, infrastructure
Kruger (2024) [26]	Qualitative	Engagement and motivation in AI-supported inclusive learning
Cai et al. (2022) [30]	Qualitative	Engagement and learning outcomes in AI-mediated environments
Xia et al. (2022) [20]	Qualitative	Engagement, motivation, and learning outcomes in AI-enhanced classrooms

plays a pivotal role by aligning specific articles with each research question (RQ), thereby facilitating the extraction of relevant textual segments. For RQ1, which investigates the impact of AI technologies on learning outcomes and engagement among students with special needs, the key variables identified across studies include learning outcomes, engagement, and motivation. These variables are consistently addressed in both qualitative and quantitative studies, such as those by Garg et al. [21], Xia et al. [20], and Alsolami [35].Their inclusion ensures that the review captures a comprehensive picture of how AI influences student performance and emotional involvement in inclusive settings. This alignment supports the systematic identification of meaningful data segments, a critical step in the first phase of qualitative content analysis.

For RQ2, which explores the barriers and enablers to AI adoption by educators in inclusive environments, the articles highlight a wide range of variables. Barriers include lack of AI literacy, infrastructure gaps, low technological self-efficacy, and policy limitations, while enablers encompass teacher openness, institutional support, perceived usefulness, and assistive technologies. Studies such as those by El Naggar et al. [25], Adigun et al. [27], and Reyes et al. [24] provide rich insights into these factors. By mapping these variables to specific articles, the review ensures that the coding process is grounded in empirical evidence, allowing for nuanced thematic development and enhancing the credibility of the findings.

Lastly, RQ3 examines how theoretical frameworks like UDL, TAM2, SDT, and UTAUT inform the design and evaluation of AI-based interventions. Articles such as those by Amouri et al. [33], Alyoussef et al. [28], and Kruger [26] address variables like effort expectancy, social influence, trust, and accessibility. These frameworks guide the interpretation of AI adoption patterns and pedagogical effectiveness. Identifying these articles ensures that the review adheres to the foundational step of extracting theory-informed textual data, thereby reinforcing the analytical depth and coherence of the systematic review.

4. Results

The following section presents the key findings of the systematic review on the use of artificial intelligence (AI) in inclusive education. To ensure clarity, transparency, and analytical coherence, the results are organized using an informative data presentation format. This approach supports a structured and reader-friendly synthesis of evidence, aligning with best practices in systematic review reporting. In particular, the review adheres to the principles outlined in the “Good Practice in Data Extraction” [41], which emphasize the importance of clear, well-documented, and methodologically sound data presentation to minimize bias and enhance the utility of findings. The extraction process was guided by rigorous qualitative content analysis (QCA), ensuring that themes and patterns were systematically identified and validated across diverse study types. To facilitate comprehension and maintain logical flow, the results are presented in the following order: (1) study characteristics, (2) findings addressing Research Question 1, (3) findings for Research Question 2, and (4) findings for Research Question 3. This structure allows for a coherent narrative that directly responds to the review’s objectives while maintaining methodological integrity.

4.1. Study Characteristics

The study characteristics presented in

Table 7. Study characteristic of the 16 articles subjected for review,.

Author(s) and Year	Country	Study Design	Sample Size and Population	Data Collection Methods	Outcomes Measured	Key Findings
Garg et al. (2024) [21]	India	Quantitative	N = 120, secondary school students	Surveys, performance tracking	Engagement levels, academic performance	AI analytics improved engagement and performance
Toyokawa et al. (2023) [22]	Japan	Qualitative	N = 15, inclusive education teachers	Interviews	Perceived usefulness, barriers	Teachers value AI but need training
Xia et al. (2022) [20]	China	Mixed Methods	N = 200, middle school students	Surveys, interviews, performance data	Learning outcomes, engagement	AI tutors improved learning outcomes
Irawan et al. (2025) [23]	Indonesia	Qualitative Case Study	N = 1 school, teachers and administrators	Interviews, policy analysis	Strategic benefits, challenges	AI supports personalization and accessibility
Coy et al. (2024) [32]	Jamaica	Mixed Methods	Focus group of Deaf community	Conceptual framework, focus group	Community perceptions, feasibility	Cautious optimism, need for cultural representation
Amouri et al. (2025) [33]	Morocco	Mixed Methods	N = 150, preservice teachers	Surveys, interviews	Technology acceptance predictors	Usefulness and social influence drive adoption
Adigun et al. (2025) [27]	Nigeria	Quantitative	N = 300, preservice teachers	Survey	Behavioral intention, perceived ease of use	Positive intention to adopt AI
Alsolami (2025) [35]	Saudi Arabia	Quantitative	N = 100, students with mild intellectual disabilities	Pre-post tests	Academic performance	AI improved academic skills
Reyes et al. (2024) [24]	Spain	Case Study	N = 1 online university	Document analysis, interviews	Opportunities and barriers	AI offers inclusion opportunities
Alyoussef et al. (2025) [28]	Saudi Arabia	Quantitative	N = 250, university students	Survey	Adoption factors	Social influence and ease of use matter
Naseer et al. (2025) [31]	Pakistan	Qualitative	N = 20, teachers and students	Interviews	Support for special needs	AI supports personalized learning
Lunavictoria et al. (2024) [34]	Peru	Quantitative	N = 180, students aged 15–18	Survey	Engagement, learning outcomes	AI improves engagement and outcomes
Beirat et al. (2025) [29]	Jordan	Qualitative	N = 25, special education teachers	Interviews	Teacher experiences	Infrastructure and training are barriers
Cai et al. (2022) [30]	China	Quantitative	N = 90, preservice teachers	Pre-post evaluation	Lesson quality, reflection	ChatGPT improves lesson planning
El Naggar et al. (2024) [25]	UAE	Qualitative	N = 30, educators	Interviews	Educator perspectives	AI enhances inclusive practices
Kruger (2024) [26]	South Africa	Qualitative Case Study	Grade 7 learners and teachers	ITS exploration, interviews, policy analysis	Differentiated learning support	MathU supports differentiation and inclusion

reflect a robust and diverse evidence base that supports the objectives of the systematic review. The articles span a wide geographical range, including countries from Asia, Africa, Europe, and Latin America, offering a global perspective on the integration of artificial intelligence (AI) in inclusive educational settings. This diversity enhances the generalizability of the findings and highlights the relevance of AI across varied socio-educational contexts.

Methodologically, the studies employ a balanced mix of quantitative, qualitative, mixed-methods, and case study designs. This variety allows for both statistical rigor and contextual depth, enabling a comprehensive understanding of how AI technologies influence learning outcomes and engagement. Sample sizes range from small qualitative groups to large-scale surveys, reflecting both micro-level insights and broader trends.

The studies also vary in their focus populations, including preservice teachers, students with disabilities, secondary learners, and educators. This range ensures that multiple stakeholder perspectives are represented. Notably, the interventions examined—such as intelligent tutoring systems, AI analytics, and ChatGPT—are directly aligned with the review’s focus on AI-enhanced learning and differentiation.

Across the studies, outcomes measured include engagement, academic performance, technology adoption, and inclusive practices. These outcomes are central to evaluating the effectiveness of AI in supporting diverse learners. Overall, the featured characteristics demonstrate methodological rigor, contextual relevance, and thematic alignment with the review’s goals. They collectively provide a strong foundation for synthesizing evidence on the impact of AI technologies in inclusive education.

4.2. RQ1—Impact of AI Technologies on Learning Outcomes and Engagement in Inclusive Education Settings

The following results show how AI tools impact both learning outcomes and student engagement in inclusive education settings. Specifically, evidence of improved personalization, motivation, and participation among diverse learners. The results also highlight how AI can reduce cognitive barriers and support differentiated instruction. Additionally, it considers contextual factors that shape the effectiveness of AI in promoting inclusive learning environments.

Prior to presenting the results, a word cloud illustration was created to visualize the most frequently occurring terms across the reviewed articles. This approach allowed for a clearer identification of dominant concepts and recurring themes within the literature.

4.2.1. Word Cloud

Using learning outcomes and engagement as codes, the word cloud for RQ1, as shown in Figure 2, reveals several dominant terms such as “learning”, “students”, “AI”, “engagement”, “inclusive”, “technology”, “support”, “feedback” and “individual.” These words reflect the central tenets of RQ1, which investigates how artificial intelligence technologies impact learning outcomes and engagement among students with diverse educational needs in inclusive settings.

“Learning” and “students” are expected focal points, emphasizing the target population and educational outcomes, while “AI” and “technology” highlight the tools being studied, while “inclusive” and “support” suggest the context and purpose, ensuring equity and accessibility. The prominence of “feedback” and “individual” suggests that personalization and adaptive learning are key mechanisms through which AI enhances engagement. “Engagement” itself strongly validates that many studies directly measured or discussed motivational and participatory outcomes. Overall, the word cloud visually affirms that the literature converges on AI’s role in personalizing instruction, enhancing learner engagement, and supporting inclusive education—the core aspects of RQ1.

4.2.2. Impact of AI Technologies on Learning Outcomes

The integration of artificial intelligence (AI) tools in inclusive education has shown promising impacts on the learning outcomes of students with special needs. Several studies highlighted the role of intelligent tutoring systems (ITS) in enhancing academic performance. For instance, the MathU ITS provided personalized learning pathways and adaptive assessments, allowing students to progress at their own pace while receiving real-time feedback and targeted support [26]. Similarly, Naseer et al. [31] found that ITS significantly improved concept mastery and reduced learning disparities among students with disabilities. These systems not only addressed individual learning gaps but also empowered teachers to deliver differentiated instruction more effectively.

Beyond tutoring systems, AI-powered assistive technologies have improved accessibility and comprehension for students with specific learning needs. Garg et al. [21] reported that tools like Microsoft Translator and Equadex enhanced learning for students with hearing and learning disabilities, while robotic tools such as LIFEisGAME supported emotional understanding in students with autism and dyslexia. Teachers observed notable improvements in retention and engagement when these tools were integrated into classroom instruction. Alsolami [35] further emphasized that AI interventions led to sustained gains in reading, writing, and math skills among students with mild intellectual disabilities, largely due to the adaptive and personalized nature of the tools.

AI also played a significant role in promoting higher-order thinking and inclusive pedagogy. El Naggar et al. [25] found that AI-mediated discussions encouraged gifted learners to engage in critical thinking and explore diverse perspectives. However, the study also noted limitations, such as overly simplistic or biased AI responses, which occasionally hindered deeper intellectual engagement. Meanwhile, Xia et al. [20] demonstrated that AI curricula grounded in Self-Determination Theory (SDT) principles—autonomy, competence, and relatedness—enhanced cognitive development and critical thinking, particularly among underrepresented learners. These findings suggest that the pedagogical design of AI tools is crucial in maximizing their educational impact.

Teachers’ perspectives across studies consistently supported the positive influence of AI on student learning. Lunavictoria et al. [34] reported that 60% of students experienced academic improvement through AI tools like chatbots and adaptive platforms, with teachers affirming enhanced engagement and comprehension. In teacher education, Cai et al. [30] showed that ChatGPT-supported critiques improved lesson planning skills, especially when guided by affective and cognitive prompts. These insights underscore the dual benefit of AI—not only in supporting students directly but also in equipping educators to better meet diverse learning needs.

4.2.3. Impact of AI Technologies on Engagement

AI tools have also significantly enhanced student engagement in inclusive classrooms by offering interactive, personalized, and accessible learning experiences. Intelligent Tutoring Systems (ITS), such as MathU, were particularly effective in promoting active participation through gamified features like leaderboards and interactive exercises [26]. These systems encouraged self-directed learning and frequent interaction with content, which teachers observed led to greater autonomy and motivation among students. Similarly, Naseer et al. [31] reported that the adaptive nature of ITS allowed students to engage with materials in ways that aligned with their learning preferences, resulting in higher levels of attention and participation.

Voice-activated AI tools and gamified applications also played a crucial role in fostering engagement among students with disabilities. Garg et al. [21] highlighted that AI-enabled devices like Siri and Alexa empowered students to interact with content independently, increasing their classroom participation. Alsolami [35] found that gamified apps such as “Busy Shapes” and “My School” maintained student focus and encouraged active learning, particularly among those with mild intellectual disabilities. Teachers and parents alike noted a marked improvement in student attentiveness during AI-supported sessions, attributing this to the interactive and visually stimulating nature of the tools.

Beyond individual engagement, AI also facilitated collaborative and emotionally resonant learning environments. Cai et al. [30] demonstrated that ChatGPT-supported lesson planning fostered peer feedback, idea synthesis, and revision, enhancing both emotional and cognitive engagement. Affective prompts were especially effective in building trust and community, with low prior-knowledge participants benefiting from emotional support and high prior-knowledge participants engaging more deeply in cognitive tasks. This dual-layered engagement contributed to more inclusive and equitable learning dynamics. Similarly, Xia et al. [20] emphasized that learner-centered AI design, incorporating culturally responsive content and project-based tasks, fostered a sense of ownership and increased participation, particularly among marginalized groups.

However, engagement through AI was not without its complexities. El Naggar et al. [25] noted that while students appreciated the flexibility and intellectual stimulation offered by AI-mediated discussions, they also expressed concerns about the lack of human interaction, information overload, and limited multimodal experiences. Despite these limitations, many students found AI to be a safe space for expression, especially those experiencing social anxiety. Lunavictoria et al. [34] supported this by reporting that 85% of students were satisfied with their AI-enhanced educational experience, with 70% of students with disabilities stating that AI tools facilitated their involvement in learning activities.

4.3. RQ2—Barriers and Enablers Influencing the Adoption and Effective Use of AI Tools by Educators in Inclusive Education

This section examines the key barriers and enablers affecting educators’ adoption and use of AI tools in inclusive education. It highlights challenges like infrastructure gaps and limited training, alongside supportive factors such as institutional backing and intuitive design. The discussion draws on empirical evidence and theory to explain these dynamics and inform inclusive, effective AI integration.

4.3.1. Word Cloud

The word cloud for RQ2 highlights two critical educator-related codes: barriers and enablers in adopting AI tools within inclusive education. Prominent words were shown in Figure 3, such as “teacher”, “training”, “support” and “confidence”, suggesting that professional development and capacity-building are essential enablers. These terms indicate that when educators are equipped with the right skills and ongoing support, they are more likely to adopt AI effectively.

Conversely, words like “infrastructure”, “access”, “resources” and “policy” point to systemic barriers. Limited access to devices, unreliable connectivity, and unclear institutional policies often hinder AI integration. The appearance of “attitudes”, “beliefs” and “readiness” underscores the psychological dimension—educators’ perceptions and motivation significantly influence adoption. In general, the word cloud affirms that successful AI implementation in inclusive settings depends not only on technological availability but also on empowering educators through training, supportive environments, and responsive policies—core tenets of RQ2.

4.3.2. Barriers Influencing the Adoption and Effective Use of AI Tools

A prominent barrier to AI adoption in inclusive classrooms is the lack of infrastructure and institutional support, which consistently emerged across multiple studies. Irawan et al. [23], Alyoussef et al. [28], and Adigun et al. [27] highlighted how inadequate technological infrastructure, insufficient access to AI tools, and the absence of clear policies significantly hindered educators’ ability to integrate AI meaningfully. These limitations not only delayed implementation but also created uncertainty among teachers, particularly in environments where inclusive education demands tailored interventions. This suggests that without equitable access, AI could reinforce existing divides, leaving marginalized learners behind. Hence, without robust systems and administrative backing, educators struggled to deliver differentiated instruction to students with special needs.

Another recurring theme is low digital literacy and insufficient training, which directly impacts teachers’ confidence and competence in using AI tools. Beirat et al. [29] and Toyokawa et al. [22] emphasized that educators faced challenges in understanding and applying AI in inclusive settings, especially those in public schools or with limited qualifications. This lack of preparedness led to ineffective use of AI dashboards and tools, particularly for learners with special needs. Similarly, Amouri et al. [33] found that teachers perceived AI tools as complex and unintuitive, which discouraged adoption and raised concerns about AI’s diagnostic capabilities for conditions like ADHD. These barriers not only reduced the likelihood of adoption but also compromised the quality of instruction for students requiring personalized support. exacerbate

Beyond technical and skill-based barriers, psychological and perceptual factors also played a critical role. Teachers expressed skepticism about AI’s usefulness and feared role displacement [37,40]. Adigun et al. [27] noted low technological self-efficacy and negative performance expectancy, which diminished behavioral intention to adopt AI. These perceptions were compounded by weak social influence, where peer and institutional encouragement failed to motivate adoption due to limited exposure and training. Such psychological resistance undermines the potential of AI to enhance inclusive teaching, as educators remain hesitant to trust or rely on these tools.

Lastly, design limitations and learner-related challenges further complicated AI adoption. El Naggar et al. [25] and Reyes et al. [24] pointed out that AI tools often lack social cues and spontaneity, making them less effective in fostering inclusive discussions. Students from marginalized groups were frequently overlooked due to the absence of inclusive design strategies. Toyokawa et al. [22] added that learners with cognitive difficulties experienced erratic engagement and overload due to poorly optimized interfaces. These design flaws not only alienated vulnerable learners but also made it harder for educators to use AI as a supportive tool in inclusive classrooms.

4.3.3. Enablers Influencing the Adoption and Effective Use of AI Tools

The current findings also reveal a set of strategic enablers that actively supported and encouraged AI integration. These enablers not only address the earlier challenges but also create new pathways for inclusive and adaptive teaching practices. A central enabler is the perceived usefulness and personalization potential of AI, which directly counters earlier doubts about its relevance and complexity. Irawan et al. [23] and Alyoussef et al. [28] found that adaptive AI tools enhanced engagement and accessibility, especially for learners with special needs. AI’s ability to tailor content, pace, and format, such as through audio/visual cues or scaffolded learning, empowered teachers to deliver differentiated instruction more effectively [22,25]. These features not only improved comprehension but also fostered learner autonomy and critical thinking [24], helping educators meet inclusive goals more confidently.

Beyond the tools themselves, institutional and professional support emerged as a powerful enabler. Teachers were more likely to adopt AI when supported by clear policies, targeted training, and intuitive interfaces [29,33]. Institutional mandates and trainer endorsements increased willingness to experiment with AI, while policy alignment, such as integrating AI into teacher education, boosted readiness [27]. These supports helped mitigate earlier concerns about low digital literacy and self-efficacy, transforming hesitation into proactive engagement.

Finally, collaborative and reflective practices played a key role in sustaining AI adoption. Toyokawa et al. [22] emphasized how shared dashboards and learning analytics facilitated meaningful discussions among educators and parents, enhancing instructional strategies. This collaborative approach, combined with AI’s capacity to create safe, flexible learning spaces [25], reinforced inclusive teaching by ensuring that diverse learner needs were continuously addressed. In sum, these enablers not only responded to the barriers but also redefined the possibilities of inclusive education through AI.

4.4. RQ3—Role of Theoretical Frameworks in Informing the Design, Implementation and Evaluation of AI-Based Interventions in Inclusive Education

The empirical papers employed in this systematic review have elicited robust data on the use of theoretical frameworks in guiding AI-based interventions. These studies not only demonstrate the relevance of theory in shaping inclusive educational technologies but also reveal favorable outcomes when theory-informed designs are applied. The consistent integration of pedagogical, psychological, and technological models across diverse contexts underscores the value of theory in supporting the adoption and effectiveness of AI in inclusive education.

4.4.1. Word Cloud

The word cloud for RQ3 reveals key concepts that illustrate how theoretical frameworks guide the design, implementation, and evaluation of AI-based interventions in inclusive education. As revealed in Figure 4, prominent terms such as “framework”, “theory”, “design”, “implementation” and “evaluation” reflect the structural underpinnings of these studies. The frequent appearance of “TPACK”, “UDL” and “model” suggests that these frameworks are actively used to shape pedagogical strategies and technology integration. Notably, the term “MathU” stands out, referencing the Intelligent Tutoring System explored in Kruger (2024) [26], which exemplifies theory-informed design in practice. Words like “learner”, “support” and “differentiated learning” emphasize the personalized and inclusive nature of these interventions. The inclusion of “lesson plan” highlights how theoretical models influence instructional planning and delivery. Altogether, the word cloud affirms that effective AI integration in inclusive education is deeply rooted in theoretical guidance, learner-centered design, and structured implementation.

4.4.2. How Theory Informs the Design of AI-Based Intervention

One prominent example in the application of theory is the use of Self-Determination Theory (SDT) in designing AI curricula for K-12 learners [20]. SDT informed the structure of learning activities to meet students’ psychological needs for autonomy, competence, and relatedness. This resulted in personalized learning paths and emotionally supportive environments that enhanced motivation and reduced anxiety, particularly among marginalized groups. Similarly, sociocultural theory, combined with epistemic network analysis, guided the design of ChatGPT-supported critique cycles in teacher education [30]. By embedding cognitive, metacognitive, and affective scaffolds, the AI tool facilitated inclusive dialogue and reflective learning, benefiting both novice and experienced pre-service teachers.

In more technology-centered approaches, models like TPACK, TAM, TAM2, and UTAUT provided practical design principles that aligned AI tools with pedagogical goals and user expectations. For instance, Kruger [26] used TPACK to integrate MathU into inclusive math classrooms, ensuring curriculum alignment and adaptive feedback. Meanwhile, Alyoussef et al. [28] and Amouri et al. [33] applied TAM-based models to design intuitive, trustworthy AI systems that fostered collaborative learning and reduced cognitive load. Adigun et al. [27] emphasized the role of effort expectancy and self-efficacy in designing accessible AI tools for teacher education. These theory-driven designs not only improved usability but also promoted equitable access and adoption in diverse educational settings.

4.4.3. How Theory Informs the Implementation of AI-Based Intervention

The implementation of AI-based interventions in inclusive education settings was deeply shaped by diverse theoretical frameworks, each offering unique insights into pedagogical design and teacher practice. For instance, Xia et al. [20] applied Self-Determination Theory (SDT) to guide the delivery of AI programs in K-12 classrooms, emphasizing autonomy-supportive language and competence-building feedback. This theory informed the creation of emotionally supportive environments, where AI tools were tailored to meet students’ psychological needs, thereby enhancing engagement and inclusivity. Similarly, Cai et al. [30] drew on Sociocultural Theory and Epistemic Network Analysis (ENA) to foster collaborative critique among pre-service teachers using ChatGPT. The AI tool facilitated iterative feedback and inclusive dialogue, aligning with Vygotsky’s emphasis on social interaction and scaffolding.

In other cases, implementation was driven by technology integration frameworks. Kruger [26] utilized the TPACK model to embed AI through the MathU tutoring system, enabling teachers to personalize instruction and monitor progress in real time. The gamified features and adaptive feedback mechanisms supported inclusive learning by addressing individual needs. Meanwhile, Amouri et al. [33] and Alyoussef et al. [28] employed variations of the Technology Acceptance Model (TAM and TAM2) to explore AI adoption among pre-service teachers. These studies highlighted the importance of intuitive design, institutional support, and perceived usefulness in promoting inclusive practices, especially for learners with ADHD or in collaborative higher education settings.

A distinct pattern emerged in Adigun et al. [27], where the Unified Theory of Acceptance and Use of Technology (UTAUT) was used to examine behavioral intentions in Nigerian classrooms. Here, ease of use and performance expectancy were critical to AI adoption, with technological self-efficacy acting as a mediator. Across these studies, theoretical frameworks not only guided the technical deployment of AI tools but also shaped the pedagogical strategies and institutional conditions necessary for inclusive education.

4.4.4. How Theory Informs the Evaluation of AI-Based Intervention

The evaluation of AI-based interventions was shaped by theories that emphasized learner motivation, social context, and technology adoption. Xia et al. [20] used Self-Determination Theory (SDT) to assess the impact of autonomy-supportive AI learning environments. Their factorial design revealed that SDT-based interventions reduced anxiety and improved motivation across gender and achievement groups, highlighting the theory’s role in promoting equity. Similarly, Cai et al. [30] applied Sociocultural Theory and Epistemic Network Analysis (ENA) to evaluate ChatGPT’s role in lesson planning. The analysis showed cognitive and affective gains among pre-service teachers, suggesting that AI tools can democratize access to quality education when scaffolded appropriately.

Other studies focused on technology acceptance models to evaluate AI adoption. Kruger [26] used TPACK to assess MathU’s effectiveness in inclusive classrooms, noting its personalized learning benefits but also its limitations in accessibility. Amouri et al. [33] and Alyoussef et al. [28] employed TAM2 and extended TAM, respectively, to evaluate AI’s usefulness and ease of use, emphasizing the influence of institutional support and ethical considerations. Adigun et al. [27], using UTAUT, stressed the importance of aligning AI design with teacher confidence and infrastructure readiness. Collectively, these evaluations underscore that theoretical frameworks not only guide the measurement of AI’s impact but also illuminate the contextual and equity-driven factors essential for inclusive education.

In summary, the articles collectively demonstrate how theoretical frameworks–SDT, TAM/TAM2, UTAUT, TPACK, and Sociocultural Theory—serve as critical lenses for shaping AI-based interventions in inclusive education. In the design phase (

Table 8. Summary of framework contributions across all six articles.

Framework	Informs Design	Informs Implementation	Informs Evaluation
SDT	Needs-based instructional design	Teacher training for autonomy, competence, relatedness	Motivation, engagement, equity outcomes
TAM/TAM2	Usability, usefulness, trust	Interface design, trainer support	Behavioral intention, adoption predictors
UTAUT	Performance and effort expectancy	Institutional support, self-efficacy	Policy alignment, curriculum integration
TPACK	Pedagogy, content, technology alignment	Curriculum-based ITS deployment	Thematic mapping, teacher feedback
Sociocultural Theory	Collaborative scaffolding	ChatGPT-mediated critique cycles	Epistemic network analysis of learning gains

), these frameworks guide the alignment of AI tools with learner needs and pedagogical goals. For instance, SDT [20] informed the creation of autonomy-supportive learning environments, while TAM2 [33] and UTAUT [27] emphasized usability, usefulness, and social influence in AI adoption. TPACK [26] ensured that AI tools like MathU were pedagogically and content-aligned, while Sociocultural Theory [30] shaped collaborative, scaffolded learning through ChatGPT.

In implementation, these theories influenced how AI tools were introduced, supported, and integrated into inclusive settings. SDT and Sociocultural Theory emphasized teacher roles in fostering motivation and dialogue, while TAM2 and UTAUT highlighted the importance of institutional support and user training. Evaluation strategies were also theory-driven: SDT measured motivational outcomes; TAM/UTAUT assessed behavioral intention and adoption; and TPACK and Sociocultural Theory used thematic and network analyses to assess pedagogical impact. Together, these frameworks ensured that AI interventions were not only technologically sound but also pedagogically inclusive and contextually responsive.

5. Discussion

5.1. Converging Insights and Critical Gaps

This systematic review explored the contribution of Artificial Intelligence (AI) to inclusive education, synthesizing evidence from 16 empirical studies and revealing both converging insights and critical gaps. Across contexts, findings consistently affirm that AI technologies, such as intelligent tutoring systems, adaptive platforms, and assistive tools, enhance personalization, accessibility, and learner engagement. These benefits were particularly pronounced among students with disabilities, attention deficits, and learning disorders, where AI-supported interventions reduced cognitive barriers and fostered motivation. However, while convergence on positive outcomes is evident, limitations persist. Several interventions prioritized diagnostic functions over sustained instructional support. This means that many AI tools excel at diagnosing learner needs (e.g., identifying gaps, predicting performance) but fail to provide continuous instructional scaffolding. This imbalance suggests that AI is often used as an assessment tool rather than a teaching partner, which limits its transformative potential in inclusive education. This imbalance further suggests that AI’s potential is not fully realized when its application focuses narrowly on assessment rather than comprehensive learning strategies.

Beyond the impact on learners, the review underscores a tension between technological promise and practical implementation. Barriers such as inadequate infrastructure, low digital literacy among educators, and ethical concerns were recurrent themes, contrasting sharply with enabling factors like institutional support, targeted teacher training, and intuitive design. These findings converge on the notion that successful adoption depends on systemic readiness rather than technology alone. Yet, gaps remain in addressing psychological resistance and equity concerns, particularly in under-resourced settings. While some studies reported strong institutional backing and collaborative practices, others highlighted fragmented efforts and inconsistent policy frameworks, revealing a lack of scalable models for inclusive AI integration. This divergence points to the need for coordinated strategies that combine technical innovation with capacity-building and ethical governance. As an implication, these underrepresented regions face compounded challenges, such as limited infrastructure, low teacher digital literacy, and fragmented policy frameworks. This means AI adoption could deepen disparities unless systemic supports (training, infrastructure, policy alignment) are prioritized. In addition, a notable tension emerged between AI’s diagnostic and instructional roles. While diagnostic capabilities were frequently emphasized, such as predictive analytics and personalized assessments, many interventions lacked robust instructional scaffolds to translate insights into sustained learning gains. This imbalance risks reducing AI to an evaluative tool rather than a pedagogical partner, particularly in inclusive contexts where continuous support is critical.

Theoretical frameworks emerged as another critical dimension, yet their application was uneven across studies. Models such as Self-Determination Theory (SDT), Technology Acceptance Model (TAM2), Unified Theory of Acceptance and Use of Technology (UTAUT), and Technological Pedagogical Content Knowledge (TPACK) were cited as guiding principles, but often in limited or superficial ways. Where theory-informed designs were implemented, outcomes demonstrated greater pedagogical coherence and learner engagement, reinforcing the value of grounding AI interventions in robust conceptual foundations. Conversely, the absence of theoretical alignment in other studies resulted in tools that were technologically advanced but pedagogically disconnected, raising concerns about sustainability and inclusivity. This inconsistency signals a significant gap: the need for frameworks that not only inform design but also guide implementation and evaluation, ensuring that AI serves as a complement to, rather than a substitute for, sound educational practice.

Taken together, these findings suggest that AI holds transformative potential for inclusive education, but its success hinges on more than technological sophistication. Effective integration requires a holistic approach—one that aligns innovation with ethics, pedagogy, and institutional readiness. Future research should move beyond short-term evaluations to examine longitudinal impacts, particularly for students with complex disabilities and in low-resource contexts. Cross-cultural studies and theory-driven interventions are essential to develop adaptable, equity-focused models. Ultimately, this review concludes that AI can advance inclusive education meaningfully when deployed as part of a systemic strategy that respects learner diversity, empowers educators, and embeds theoretical rigor. Without these conditions, AI risks reinforcing existing disparities rather than bridging them.

5.2. Conclusions

This systematic review affirms that Artificial Intelligence (AI) can play a transformative role in inclusive education by enabling personalized learning, improving accessibility, and fostering engagement among students with special needs. Evidence from 16 empirical studies converges on the potential of AI tools to reduce cognitive barriers and support differentiated instruction. However, this promise is lessened by persistent challenges. Barriers such as inadequate infrastructure, low digital literacy, and ethical concerns remain widespread, while theoretical frameworks like SDT, TAM2, UTAUT, and TPACK are applied inconsistently, limiting pedagogical coherence. Furthermore, gaps in longitudinal evidence and cross-cultural applicability raise questions about sustainability and equity, particularly in under-resourced contexts, while a notable tension emerged between AI’s diagnostic and instructional roles.

To fully realize AI’s potential, integration must go beyond technological innovation. It requires systemic strategies that combine ethical governance, institutional readiness, and robust teacher training with theory-driven design and evaluation. When implemented thoughtfully, AI can become a powerful catalyst for transforming inclusive education, bridging learning gaps, empowering educators, and advancing global commitments to equity and social justice. Future research should also explore hybrid AI models that seamlessly link diagnostic insights with adaptive instructional strategies, ensuring that identification of learner needs is coupled with actionable, sustained teaching interventions.

5.3. Limitations

This review is subject to several methodological limitations. First, the exclusion of grey literature and non-English publications may have restricted the breadth of perspectives, potentially omitting innovative practices or culturally specific insights from less formal sources. Second, the reliance on peer-reviewed studies published between 2020 and 2025, while ensuring quality, limits historical comparisons and emerging trends beyond this timeframe. Third, most included studies employed cross-sectional or short-term designs, leaving gaps in understanding the longitudinal impact of AI on inclusive education. Additionally, variations in study contexts and inconsistent application of theoretical frameworks hindered comparability and synthesis. Future research should address these limitations by incorporating grey literature, conducting longitudinal studies, and developing culturally responsive models. Interventions grounded in robust educational theory are essential to ensure that AI integration promotes equity, sustainability, and meaningful learning outcomes across diverse educational settings.