Technology-Based Interventions in Mathematics Learning for Students with ADHD: A Scoping Review

Abstract

Students with attention-deficit/hyperactivity disorder (ADHD) often experience difficulties in mathematics related to attention, working memory, and executive functioning. Although technology-based interventions (TBIs) and emerging artificial intelligence (AI) technologies offer opportunities for personalized support, evidence concerning their use in mathematics education for students with ADHD remains fragmented. This scoping review mapped the literature on technology-supported mathematics interventions, focusing on intervention types, targeted cognitive barriers, educational outcomes, and evidence gaps. Following JBI scoping review methodology and PRISMA-ScR guidelines, 37 peer-reviewed studies published between 2010 and 2026 were identified and categorized into four tiers according to their alignment across technology, ADHD, and mathematics. Findings revealed a bimodal technology landscape, with established computer-based interventions showing modest empirical support, while AI-supported approaches remained largely theoretical and lacked empirical validation. Reported mathematics-related outcomes were heterogeneous across the included sources and were not directly comparable across intervention classes. Because of substantial methodological heterogeneity and the absence of formal risk-of-bias assessment, no comparative conclusions can be drawn across intervention classes. Major evidence gaps include limited empirical evaluation of AI-driven mathematics interventions, scarce longitudinal research, and underrepresentation of secondary-school populations.

1. Introduction

Attention-Deficit/Hyperactivity Disorder (ADHD) is among the most prevalent neurodevelopmental conditions affecting school-age children, and its core features, such as inattention, impulsivity, and difficulties with self-regulation, intersect directly with the cognitive demands of mathematics learning. Mathematical problem-solving is particularly dependent on executive functions because multi-step procedures require sustained working memory to retain intermediate values, inhibitory control to suppress premature or careless responses, and processing speed to complete calculations before attention drifts [1]. Because these are precisely the domains most disrupted in ADHD, students with the condition are disproportionately likely to fall behind their peers in mathematics achievement, even when general cognitive ability is unaffected [2]. Foundational classroom-intervention research has long recognized this link, identifying mathematics alongside reading as a domain where targeted instructional strategies for ADHD yield some of the clearest academic gains [3].

Over the past two decades, technology-based interventions (TBI) have emerged as a promising avenue for addressing these ADHD-related barriers. Computer-assisted instruction (CAI) offers self-paced repetition, immediate feedback, and reduced extraneous stimuli that many students with ADHD respond well to, and reviews spanning more than two decades of research point to consistent, if heterogeneous, benefits for on-task behavior and academic performance [4]. Early randomized trials demonstrated that computer-based attention and academic interventions could produce measurable gains on standardized mathematics outcomes for children with attention difficulties [5], while school-based computerized attention training showed downstream improvements captured in teacher ratings of academic functioning, including mathematics [6]. Beyond CAI, gamified platforms have been used to support attention and motivation in inclusive elementary classrooms [7], while more recent randomized trials have also reported improvements in attention and academic performance following the use of gamified applications [8]. A parallel line of research has focused directly on underlying cognitive mechanisms. For example, digital working-memory and arithmetic-reasoning training have improved performance in children with ADHD [9,10]. Meta-analytic evidence [11,12] also suggests that digital interventions can lead to modest improvements in mathematics achievement among children with mathematical learning difficulties.

Artificial intelligence (AI) has recently introduced an additional dimension to this field. Rather than delivering fixed content, AI-driven systems, such as intelligent tutoring systems (ITS), pedagogical agents, and adaptive platforms, can personalize instructional pacing and feedback in real time. This capability is particularly relevant for learners whose attentional and working-memory profiles fluctuate within and across sessions. Early evidence for this approach is encouraging: a pedagogical agent designed to scaffold fifth-grade mathematics produced measurable learning gains among students with ADHD [13], and subsequent work has begun to identify intelligent tutoring systems as the dominant AI modality being applied to mathematics instruction for students with special educational needs more broadly [14]. Systematic reviews of AI in K-12 special education document a decade of accelerating activity, with mathematics recurring as one of the most frequently targeted academic domains [15], and scoping reviews of generative AI applications for neurodivergent learners, spanning ADHD, autism, and dyslexia, point to a rapidly expanding but still poorly consolidated evidence base [16]. AI is also reshaping the identification side of this field because machine-learning and explainable deep-learning approaches are increasingly used to screen for and diagnose ADHD itself, suggesting a second, diagnostic strand of AI-related research that sits adjacent to, but distinct from, AI-driven instruction [17,18,19].

Despite growing research on technology-based interventions for ADHD and on AI applications in inclusive education and specifically in mathematics, the literature at their direct intersection remains narrow and fragmented. Broader systematic reviews of technology-based interventions for ADHD tend to report academic outcomes only incidentally, alongside primary outcomes such as attention or behavior [20], while reviews of technology-based mathematics interventions for learning difficulties typically treat ADHD as a comorbid subgroup within a broader dyscalculia or mathematical-learning-difficulty population rather than as a primary inclusion criterion [21,22,23]. As a result, no existing review has systematically examined how technology-based and AI interventions are designed, deployed, and evaluated for mathematics learning in students with ADHD. Specifically, there remains a gap in understanding their formats, the cognitive barriers they target, and the educational and behavioral outcomes that have been assessed. Given the considerable variability in interventions, outcome measures, and study designs identified across this literature, a scoping review was considered one of the most appropriate methodologies for this stage of inquiry rather than a systematic review or meta-analysis. In accordance with the JBI Manual for Evidence Synthesis [24], scoping review methodology maps the breadth and nature of an emerging or diverse evidence base, identifying key concepts and characteristics, and clarifying evidence gaps that may inform future research or more focused evidence syntheses [25]. This scoping review is guided by the following questions (RQs):

• RQ1. What types of technology-based interventions and tools have been used to support mathematics learning among students with ADHD?
• RQ2. Which ADHD-related learning or cognitive difficulties are explicitly targeted by these interventions?
• RQ3. What outcomes have been assessed in studies of these interventions?
• RQ4. What evidence gaps exist regarding technology-based interventions for mathematics learning for students with ADHD?

This scoping review aims to comprehensively map the existing literature on TBI, such as CAI, virtual manipulatives, gamified platforms, and assistive technologies in mathematics education for students with ADHD. In particular, it aims to identify the types of digital tools employed, how they address ADHD-related cognitive and learning needs (such as inattention and working-memory difficulties), and the reported outcomes relating to mathematical achievement, on-task behavior, engagement, and other educational domains.

The main contribution of this scoping review is threefold. First, it provides a comprehensive mapping of a technology field in which established computer-based interventions for mathematics learning among students with ADHD coexist with a growing number of proposed but empirically unsubstantiated AI applications. Second, it makes a conceptual contribution by distinguishing accommodation-oriented approaches that embed ADHD-related supports within mathematics instruction from domain-general cognitive-training approaches. This distinction is conceptual and does not involve cross-study effect size comparisons; no comparative superiority between these approaches is assumed or asserted. Third, it delineates the methodological shortcomings of the existing evidence base. This review directs the agenda for primary research and informs stakeholders (researchers, designers, educators, and policymakers) on how to conduct, implement, and evaluate technology-based interventions for students with ADHD.

2. Methods

This scoping review was conducted in accordance with the methodological guidance provided in the JBI Manual for Evidence Synthesis [24,25] and reported in line with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews (PRISMA-ScR) [26]. The review scope and inclusion criteria were developed using the Population-Concept-Context (PCC) framework recommended by JBI [25]. A scoping review design was appropriate because the available literature is methodologically heterogeneous and encompasses randomized controlled trials, quasi-experimental and single-subject studies, as well as narrative and systematic reviews. The JBI methodology provided the methodological framework for conducting the scoping review, whereas PRISMA-ScR was used as the reporting framework to ensure transparent and complete presentation of the review methods, evidence-selection process, and results. Moreover, the objective was to map the breadth and characteristics of an emerging evidence base rather than to estimate a pooled effect, consistent with the rationale adopted in recent scoping reviews examining AI and digital technologies in special education [27].

A scoping review design was appropriate because the available literature is characterized by substantial heterogeneity in both technological modalities and research designs [28]. The identified studies encompass generative-AI and personalized learning interfaces [29], AI-enabled assistive educational tools [30], gamified videogames for mathematics learning [31], online interventions for learners with ADHD and specific learning difficulties [32], computer-based mathematics interventions [33], and single-subject technology-supported instructional designs [34]. The literature also includes augmented-reality-based mathematics learning systems [35], illustrating the emergence of immersive and interactive technologies in this field. In parallel, broader evidence syntheses have examined neurofeedback-based technologies for children with ADHD and specific learning disorders [36], digital technologies targeting attention and executive functioning [37], and computerized working-memory training for youth with ADHD [38]. This variation in technological modality, instructional purpose, study design, and outcome domain further supports the use of a scoping review to systematically map the breadth and characteristics of the evidence rather than restrict the synthesis to a single intervention category.

Clinical trial registration was not applicable, and ethics approval, informed consent, and consent for publication were not required because the review synthesized findings from previously published studies. After database searching, records were screened by title and abstract, and potentially eligible studies were assessed in full text against the predefined eligibility criteria. Reasons for exclusion at the full-text stage were documented. Disagreements between reviewers (authors) were resolved through discussion, with a third reviewer consulted when consensus could not be reached. The study-selection process was summarized in a PRISMA-ScR flow diagram [26]. A total of thirty-seven (n = 37) peer-reviewed studies met the eligibility criteria and were included in the final synthesis. PRISMA-ScR was used as the reporting framework to support transparent and complete reporting of the review methods, evidence-selection process, and results. The PRISMA-ScR flow diagram and checklist are provided below and in the Supplementary Materials.

2.1. Protocol and Registration

Prior to data charting, a formal protocol was developed to outline the review’s objectives, eligibility criteria, and planned analytical strategy. In accordance with principles of transparency and reproducibility, the study protocol has been documented and is available at the following repository: https://shorturl.at/FkdTh (accessed on 27 July 2026). The protocol was developed in accordance with the JBI scoping review methodology and PRISMA-ScR guidelines, including specification of the PCC elements, eligibility criteria, search strategy, source selection procedures, and data-charting approach. Consistent with the iterative nature of scoping reviews, any substantive deviations from the protocol were to be transparently documented and justified.

2.2. Eligibility Criteria

Eligibility was defined using the PCC framework recommended for scoping reviews [25] and detailed in Appendix B:

• Population: Children and adolescents in formal or informal educational settings (primary through secondary) with a documented diagnosis of ADHD, clinically identified ADHD symptomatology, or ADHD as an explicitly reported comorbid condition within a broader sample (e.g., students with mathematical learning difficulties or other special educational needs).
• Concept: The core Concept was technology-based intervention or technology-supported provision applied to mathematics learning, instruction, assessment, or ADHD-related academic support. This included CAI, gamified digital platforms, virtual manipulatives, cognitive-training software, assistive technologies and accommodations, and artificial-intelligence-driven tools such as intelligent tutoring systems, pedagogical agents, and AI-based diagnostic or screening systems. The Concept also encompassed the characteristics of the technologies, their targeted ADHD-related learning needs, design mechanisms, and reported educational outcomes where these were relevant to the review objectives.
• Context: The Context comprised educational and learning environments in which mathematics learning, mathematics instruction, assessment, or ADHD-related academic support was delivered. This included general education, special education, and relevant clinical or therapeutic settings, as well as in-person, computer-based, and remote/online delivery formats. The context was intentionally broad because the purpose of the review was to map the range of settings in which technology-based approaches relevant to mathematics learning and ADHD have been investigated.

The three PCC elements were operationalized in conjunction with the review’s four-tier evidence-mapping strategy. Because the direct intersection of technology, ADHD, and mathematics was expected to be narrow, the review included sources that addressed at least two of the three core domains while providing explicit conceptual relevance to the third. This approach was used to preserve the breadth required for evidence mapping while maintaining transparent boundaries around relevance. Accordingly, four evidence tiers were predefined: (a) Tier 1 comprised sources addressing the direct intersection of technology-based mathematics interventions and students with ADHD; (b) Tier 2 comprised technology-based ADHD interventions reporting academic or mathematics-related outcomes; (c) Tier 3 comprised technology-based mathematics interventions involving populations with learning difficulties in which ADHD was a documented comorbid condition; and (d) Tier 4 comprised AI-specific reviews and frameworks with explicit conceptual relevance to the ADHD–mathematics intersection. It should be noted that Tier 1 comprises primary empirical studies, whereas Tiers 2 and 3 include a mixture of primary empirical studies, reviews, and meta-analyses, and Tier 4 comprises reviews, frameworks, and technical studies. Tier 4 was treated as a broader contextual evidence tier rather than as equivalent to direct PCC evidence.

To preserve breadth while maintaining transparent boundaries, tier-specific eligibility was applied. Tiers 1–3 were restricted to school-aged populations (ages 5–18) with mathematics education as an explicit context. Tier 4 (contextual AI evidence) was predefined to include reviews, frameworks, and diagnostic or technical studies with broader age ranges when they provided explicit conceptual relevance to the ADHD–mathematics intersection; such sources are clearly labelled as contextual rather than direct evidence.

Eligible source types included empirical primary studies, peer-reviewed conference proceedings, systematic reviews, meta-analyses, and selected evidence syntheses or conceptual frameworks where they met the predefined tier-specific relevance criteria. No lower date limit was applied a priori, reflecting the exploration and mapping aim of the review. The final database searches retrieved no eligible sources published before 2010. Consequently, the included literature spans 2010 to 2026. In particular, the final database searches were conducted on 30 July 2026. Backward and forward citation tracking and supplementary hand-searching were also completed through the end of July 2026.

2.3. Information Sources

Multiple complementary sources were consulted to maximize retrieval sensitivity, including electronic bibliographic databases (e.g., Scopus, Web of Science Core Collection, ERIC, PubMed/MEDLINE, PsycINFO, and IEEE Xplore for AI- and computing-oriented sources), backward and forward citation tracking of included sources, and hand-searching of journals in which relevant work was identified during the initial scoping process through the end of July 2026 (e.g., Journal of Special Education Technology, Journal of Attention Disorders, Computers & Education). The selection of information sources was designed to identify the breadth of evidence relevant to the PCC-defined scope, including literature indexed within education, psychology, health, special education, mathematics education, and computing/AI databases. Detailed database coverage, search dates, and database-specific search strategies are provided in Appendix B.

2.4. Search Strategy

A structured Boolean search strategy combined three concept blocks, including ADHD, mathematics, and technology/AI connected with the AND operator, with synonyms and truncation within each block connected by OR. The search strategy was developed iteratively in accordance with JBI scoping review guidance and was designed to maximize retrieval across the three core domains represented in the review scope. A representative search string, illustrated here for a Scopus title-abstract-keyword search and adapted to the syntax of each additional database, was:

(ADHD OR “attention deficit*” OR “attention-deficit/hyperactivity disorder”) AND (math* OR numeracy OR arithmetic OR “mathematics education” OR “mathematics learning”) AND (technology OR digital OR computer* OR “artificial intelligence” OR AI OR “intelligent tutoring” OR gamify* OR game* OR “virtual reality” OR “augmented reality” OR app OR software).

This structure follows the concept-block approach used in adjacent technology-and-ADHD systematic reviews [20] and digital-intervention meta-analyses for mathematical learning difficulties [11], which similarly combined population, concept, and technology terms rather than relying on a single combined phrase. It is a strategy that improved sensitivity given the fragmentation of terminology (e.g., “computer-assisted instruction” versus “digital learning platform” versus “intelligent tutoring system”) observed across the included literature.

The complete search strings used for each database are provided exactly as executed in (Appendix B.3.), including the field restrictions, Boolean operators, truncation, phrase searching, and any database-specific limits. Supplementary and grey-literature sources were searched to maximize retrieval sensitivity and identify potentially relevant evidence not captured through bibliographic database searches. All records identified through these sources were screened using the same predefined eligibility criteria, and only eligible sources with sufficient full-text information for data charting were included in the final evidence map in Appendix B.4. These details are reported to enable replication of the search process across the information sources.

2.5. Record Screening

Records retrieved from all sources were merged into a single reference-management library and de-duplicated using EndNote software (version 22.0.0.21347). Screening proceeded in two stages, consistent with standard scoping-review conventions [26]: (a) title and abstract screening against the eligibility criteria above, and (b) full-text review of records retained at stage one.

Two reviewers independently conducted both screening phases; articles that appeared to meet the eligibility criteria or remained ambiguous in the initial phase proceeded to full-text evaluation, and disagreements at either stage were resolved through discussion or referral to a third reviewer. Appendix B.4 presents the entire process. The specific grounds for excluding full-text papers were recorded. The overall source-selection process, including the number of records identified, duplicates removed, records screened, records excluded, and reports assessed for full-text eligibility with reasons for exclusion, was reported using a flow diagram. Figure 1 below illustrates the identification, screening, eligibility assessment, and inclusion of sources of evidence in this scoping review.

Because the exact triple intersection of technology, ADHD, and mathematics yielded a narrow evidence base, the predefined four-tier evidence-mapping framework was used to characterize the breadth of the literature within and around the PCC-defined scope. Studies were eligible if they addressed at least two of the three domains with explicit relevance to the third: Tier 1 (n = 10, 27.0%) comprised studies at the direct intersection (technology-based mathematics interventions for ADHD students); Tier 2 (n = 8, 21.6%) comprised technology-based ADHD interventions that measured academic or mathematics outcomes; Tier 3 (n = 10, 27.0%) comprised technology-based mathematics interventions for learning difficulties where ADHD is a documented comorbid condition; and Tier 4 (n = 9, 24.3%) comprised AI-specific reviews and frameworks relevant to the ADHD–mathematics intersection. The tiering strategy therefore functioned as an evidence-mapping device, distinguishing direct evidence from broader contextual evidence rather than treating all tiers as equivalent in their degree of alignment with the PCC framework (Figure 2).

2.6. Data Charting Process

A structured data-charting form was developed and piloted on a subset of included sources before being applied across the full corpus, consistent with the iterative data-charting approach recommended within JBI scoping review methodology. The form was organized around the four-tier classification scheme and a ten-category technology-type taxonomy (spanning, for example, AI or ITS, computer-assisted instruction, gamification, cognitive training, and AI-based diagnostic tools) developed specifically for this review to accommodate the heterogeneity of technologies identified across the corpus. Charting was performed independently with a second reviewer verifying a random sample of charted records for consistency.

Data were charted using a 34-variable extraction form presented in Appendix C. The form was piloted on five randomly selected studies and refined for clarity and completeness before full data charting. Variables included study characteristics (design, country, sample size, ADHD diagnosis method), intervention characteristics (technology type, AI subtype, delivery mode, duration, mathematical domain), and outcome characteristics (academic, cognitive, behavioral, engagement, and social domains; see Appendix D). Two reviewers independently charted 20% of included studies; inter-rater agreement was 91% (Cohen’s κ = 0.88), with discrepancies resolved through consensus discussion. Although 20% of the included studies were independently dual-charted to establish reliability (Cohen’s κ = 0.88), the remaining 80% were charted by a single reviewer. This limitation should be considered when interpreting the extracted data, which represents a best-effort map rather than a fully independently verified dataset.

2.7. Data Items

For each included source, the following data items were charted: (a) bibliographic details (authors, year, journal); (b) tier classification (intersection level); (c) technology type; (d) country or setting; (e) study design; (f) population characteristics, including age or grade level, sample size, and ADHD diagnostic criteria; (g) intervention description; (h) comparator or control condition, where applicable; (i) intervention duration; (j) outcome measures; (k) key findings; (l) citation count at the time of charting, as an indicator of the source’s influence within the field; and (m) risk-of-bias or quality notes. These data items were selected to enable systematic mapping of the Population, Concept, and Context dimensions and to characterize the distribution and nature of evidence across the four predefined tiers. Where a data item could not be determined from the full text, this was recorded as “not reported” rather than left blank, to distinguish non-reporting from incomplete charting.

During charting, several predefined simplifications were applied to maintain consistency across heterogeneous sources. First, sample size was recorded separately as total sample size and, where reported, the number of participants with ADHD; when a source did not provide an ADHD-specific sample, this was recorded as “N/A” or “not reported” rather than estimated. Second, the four-tier classification was assigned according to the predefined degree of alignment among technology, ADHD, and mathematics, with Tier 4 treated as contextual rather than direct evidence. Third, when a source addressed multiple technologies, outcome domains, or mathematical domains, all relevant categories were recorded rather than forcing the source into a single category. These conventions were applied consistently across the included sources.

2.8. Critical Appraisal of Individual Sources of Evidence

In accordance with the JBI approach to scoping reviews, formal critical appraisal of methodological quality or risk of bias was not undertaken uniformly across the included sources because the primary purpose of the review was to map the extent, characteristics, and distribution of the available evidence rather than to determine the certainty or comparative effectiveness of interventions. The heterogeneity of study designs, ranging from randomized controlled trials to single-subject and case-study designs, further limited the appropriateness of applying a single appraisal instrument uniformly across the corpus. Where methodological or quality-related information was reported in the source, it was charted descriptively as a data item rather than used as a criterion for exclusion [24,25]. No formal critical appraisal instrument was applied uniformly across the corpus, given the heterogeneity of study designs. Consequently, the absence of formal risk-of-bias assessment means that this review cannot determine the certainty of reported effects, rank interventions by comparative effectiveness, or establish superiority of any one technological approach over another. All numerical findings reported in the Results and Discussion are descriptive characteristics of the mapped literature.

3. Results

The main analysis below synthesizes evidence from 37 peer-reviewed studies spanning 2010–2026, categorized across four inclusion tiers: Tier 1 (direct intersection: technology × ADHD × mathematics; n = 10), Tier 2 (technology × ADHD with academic/math outcomes; n = 8), Tier 3 (technology × mathematics with ADHD as documented comorbidity; n = 10), and Tier 4 (AI-specific reviews and frameworks; n = 9). The four tiers were defined to operationalize the degree of alignment between the included sources and the review’s PCC-defined scope; they do not represent levels of evidence quality.

Charted data were synthesized narratively and descriptively, organized by tier and technology type rather than pooled quantitatively, consistent with scoping-review conventions [25,28]. In accordance with the JBI scoping review approach, the results were mapped descriptively against the review’s PCC framework, with particular attention to participant characteristics (Population), technology-based mathematics education and associated design mechanisms (Concept), and educational, geographical, and developmental settings (Context). Figure 3 below illustrates the temporal distribution of the 37 included sources, showing a marked acceleration post-2020 and a peak in 2025 (n = 8).

Two complementary analytic lenses were used: first, a cross-tabulation of technology type by publication year, sized by citation count, to characterize the maturity and trajectory of each technology category over time; and second, a tier-based synthesis distinguishing sources that directly studied technology-based mathematics interventions for students with ADHD (Tier 1) from those addressing the topic through ADHD-technology, mathematics-technology-comorbidity, or AI-specific proximities (Tiers 2–4). The tier structure was used as an analytical mapping device to distinguish direct evidence from adjacent evidence and should not be interpreted as a hierarchy of methodological quality or certainty. This dual approach allows the review to characterize not only what has been studied, but how directly the existing evidence base speaks to the review’s central question. It is a distinction that is itself one of the review’s findings, given the comparative scarcity of Tier 1 sources relative to the adjacent literatures mapped in Tiers 2 through 4. Across the results, findings are therefore presented primarily as descriptive characteristics of the evidence base rather than as estimates of intervention effectiveness. This section also presents the findings of the scoping review organized around four research questions.

3.1. TBI Categories and Evaluation Tools (RQ1)

Across the PCC framework, the Concept domain was represented primarily by technology-supported mathematics instruction, cognitive training, adaptive learning, gamification, pedagogical agents, and emerging AI-enabled tools. The Population domain varied substantially across sources, ranging from students with a documented ADHD diagnosis to broader groups of students with learning difficulties, disabilities, or ADHD-related characteristics. The Context domain included school-based, online, clinical/educational support, and other learning environments across multiple geographical settings. This variation in population and context was particularly important when interpreting the degree of direct alignment with the review question.

The literature reveals a wider technological field with ten distinct categories of digital tools applied to mathematics education for students with ADHD or related learning difficulties (see Figure 4). Because some sources described more than one technological modality, technology-category frequencies were not mutually exclusive and therefore do not sum to the total number of included sources. This distribution is as follows:

Review articles and meta-analyses constitute the largest strand of literature, collectively representing 14 studies (37.8%). These span mixed-technology reviews (n = 6), digital/CAI/executive-function reviews (n = 4), and AI-focused reviews (n = 4). This predominance of synthesis and conceptual work over primary empirical intervention studies is itself a key characteristic of the evidence base.

Among empirical intervention categories, four technology types each accounted for four studies (10.8% each): Computer-Assisted Instruction (CAI) platforms, Cognitive Training (WM/Attention) programs, Gamification/Serious Digital Games, and Digital/Online Learning Platforms. CAI platforms typically deploy drill-and-practice software or structured computer-based instruction delivered via desktop computers in school lab settings. Representative studies include Herzog and Casale’s [39] computer-based mathematics intervention for primary students with ADHD/EBD, and Sánchez-Pérez et al.’s [40] school-based computer training combining mathematics and working memory exercises. CAI platforms are characterized by structured protocols, repetitive exercises, and incremental difficulty scaling. However, they generally lack true adaptive intelligence; difficulty adjustments are typically pre-programmed rather than dynamically responsive to real-time learner performance.

Cognitive Training (WM/Attention) programs (n = 4) represent an established strand of research targeting domain-general executive functions. Sperafico et al. [9] combined working memory training with arithmetic reasoning practice, while Wiest et al. [10] and Chacko et al. [38] employed adaptive computerized exercises targeting working memory and attention. These interventions are grounded in the hypothesis that strengthening domain-general cognitive resources will free capacity for mathematical processing. However, as discussed in Section 3.3, near-transfer cognitive gains have not consistently translated to mathematics achievement.

Gamified digital platforms and serious games (n = 4; 10.8%) leverage game mechanics (e.g., points, levels, narrative immersion, and immediate reward schedules) to sustain attention during mathematical tasks. Dai et al. [8] conducted an 8-week randomized controlled trial (RCT) of a gamified app for Chinese children with ADHD, finding significant improvements in both attention and academic performance. Kusmawati et al. [7] reported increased concentration and math achievement through gamified learning media in an inclusive Indonesian classroom. The theoretical rationale for gamification aligns with ADHD symptomatology: variable reward schedules and high-frequency feedback loops are hypothesized to compensate for dopaminergic dysregulation associated with inattention and motivation deficits.

AI and Machine Learning-driven tools represent the fastest-growing category, as Figure 3 depicts. This category subsumes: (a) intelligent tutoring systems (ITS) with embedded pedagogical agents (n = 2), (b) AI-focused reviews and frameworks (n = 4), (c) AI-enhanced diagnostic and screening tools (n = 3), and (d) generative AI applications (reviewed conceptually). Mohammadhasani et al. [13] provide the clearest empirical example of an ITS directly targeting ADHD learners, deploying a pedagogical agent named “Koosha” that delivered scaffolded arithmetic instruction to 5th-grade Iranian students. The agent reduced cognitive load through attention-guiding cues and immediate feedback, yielding a large effect size (d = 0.82) on math achievement. More recently, Ronksley-Pavia et al. [16] proposed generative AI-enhanced adaptive interfaces for neurodivergent learners, though this remains at the design stage without empirical validation. The AI category is heavily weighted toward reviews, frameworks, and diagnostic applications rather than primary intervention studies e.g., [15,16,30] collectively synthesize AI applications for neurodivergent learners but note a paucity of ADHD-specific mathematics interventions.

Online and digital learning environments (n = 4; 10.8%) encompass web-based instruction, Learning Management Systems (LMS), and virtual classroom tools. Botsas and Grouios [4] compared online versus traditional instruction of mathematical operations for Greek students with ADHD, finding moderate effects (d = 0.55) favoring the digital condition. Lomibao and Tabor [32] described an online intervention for a Filipino learner with comorbid ADHD and math-specific learning disability, delivered through an educational therapy center’s digital platform. The COVID-19 pandemic appears to have accelerated research in this domain, with Bouck et al. [33] examining online math instruction for secondary students with learning disabilities (including ADHD as a major subgroup) in the post-pandemic context.

Assistive Technology and Accommodations (n = 3; 8.1%) include color-supported instructional designs, accessibility tools, and reduced-distraction interfaces. Shafirarossa et al. [34] conducted a single-subject study of theory-informed color-supported early numeracy instruction for a 6-year-old child with ADHD, demonstrating improved number sense through visual scaffolding. Scalise et al. [41] reviewed accommodations in digital interactive STEM assessment tasks, finding that reduced-distraction interfaces were particularly beneficial for students with ADHD.

Neurofeedback and biotechnologies (n = 1; 2.6%) represent an emerging frontier. Patil et al. [36] reviewed EEG-neurofeedback applications for children with ADHD and specific learning disorders, noting that such technologies can target the attentional substrates underlying mathematical learning, though direct math outcome measurement remains rare.

As shown in Figure 4, review articles and meta-analyses collectively constituted the largest category (n = 14, 37.8%), followed by four empirical categories each accounting for four studies (CAI, cognitive training, gamification, and digital/online learning; 10.8% each). AI diagnostic tools (n = 3, 8.1%), assistive technology/accommodations (n = 3, 8.1%), and AI/intelligent tutoring systems (n = 2, 5.4%) were less represented.

A granular analysis of Tier 4 (AI-specific) sources indicates that the AI evidence base relevant to the PCC framework remains predominantly indirect with respect to the core Concept of AI-supported mathematics instruction for the Population of students with ADHD. Of the 9 AI-focused papers, 7 are reviews or frameworks; only 2 report primary data, and these primarily address diagnostic or identification applications rather than AI-mediated mathematics instruction. The AI tools described include:

Machine learning classifiers for ADHD identification [17,18], which use eye-tracking, fMRI, or behavioral data to detect ADHD but do not deliver mathematics instruction.

Deep learning diagnostic frameworks for dyscalculia screening [19], which acknowledge ADHD comorbidity but do not target instruction.

Generative AI and LLM-based tutoring systems [16], proposed as future directions but with limited or no directly identified empirical evidence involving students with ADHD in mathematics learning contexts within the present review scope. This reveals a critical evidence-mapping gap: while the AI-in-education literature for neurodivergent learners is expanding rapidly, the present evidence map contains very limited direct evidence at the intersection of AI, ADHD, and mathematics education. This field, therefore, appears to be characterized by a substantial body of conceptual, review, diagnostic, and adjacent evidence, alongside a much smaller body of directly aligned instructional evidence.

As shown in the heatmap in Figure 5, AI-focused review papers cluster heavily in 2025 (n = 5), while empirical CAI and cognitive training studies are distributed more evenly across the decade, indicating a recent shift in the composition of literature toward AI-related approaches rather than demonstrating a corresponding increase in empirically validated AI-based mathematics interventions for students with ADHD.

3.2. ADHD-Related Learning and Cognitive Difficulties Targeted by the Interventions (RQ2)

Within the Concept domain, the included sources described several ADHD-related cognitive and behavioral characteristics as relevant to technology-supported mathematics learning. These included inattention, working-memory demands, inhibitory-control difficulties, self-regulation, and motivational/engagement difficulties. The analysis for RQ2 focused specifically on the difficulties that interventions explicitly targeted or accommodated, rather than on the technological mechanisms through which these aims were implemented. The evidence map therefore captures the range of ADHD-related learner characteristics that the included sources explicitly identified as targets of technology-supported mathematics interventions; the technological features used to address each barrier are described below as illustrations of how that barrier was targeted, not as an independent analytic dimension.

The interventions identified addressed or explicitly considered three core ADHD-related cognitive barriers, though with differential emphasis. Inattention and difficulties sustaining attention were the most frequently reported barriers (n = 28 papers; 75.7%). Across these sources, attention was primarily treated as a learning-related characteristic relevant to intervention design rather than as an outcome that could be assumed to improve as a consequence of technology use. Interventions addressing inattention employed diverse technological mechanisms:

• Attention-directing cues: Mohammadhasani et al.’s [13] pedagogical agent “Koosha” used gaze-guiding animations and highlighted salient problem features to direct visual attention away from distractions.
• Gamified reward schedules: Kusmawati et al. [7] and Dai et al. [8] embedded variable-ratio reinforcement within math tasks to maintain attentional engagement through intrinsic motivation.
• Reduced environmental distraction: Computer-based training platforms [5,6] remove the social and sensory distractions of the typical classroom, creating a controlled stimulus environment.
• Chunking and micro-learning: Gunnars [37] noted that digital math interventions for students with ADHD typically break tasks into shorter sequences, accommodating limited sustained attention spans.

Working memory (WM) demands were explicitly considered in 14 papers (37.8%). The included sources commonly linked working-memory demands to multi-step mathematical tasks and described technology as either a training mechanism or a means of externalizing task information. WM impairment is a well-documented cognitive correlate of ADHD that can disrupt mathematical reasoning, particularly during multi-step problem-solving and mental arithmetic. The design mechanisms include:

• Dual-task training protocols: Sperafico et al. [9] combined WM training (n-back tasks, digit span exercises) with arithmetic reasoning practice, positing that strengthening WM capacity would free cognitive resources for mathematical processing.
• Adaptive difficulty algorithms: Wiest et al. [10] and Chacko et al. [38] employed computer programs that adjusted task demands in real time based on WM performance, preventing overload while maintaining the zone of proximal development.
• Externalization of working memory: Digital platforms that display intermediate steps, provide virtual manipulatives, or maintain a visible problem state reduce the WM load during calculation [11].

Impulsivity and inhibitory-control difficulties were reported in 9 papers (24.3%), although less frequent than inattention or WM. The corresponding technologies incorporated structured response procedures, immediate feedback, and self-monitoring features. Design mechanisms include:

• Structured response protocols: Computer-based interventions require deliberate clicking or typing rather than impulsive hand-raising, adding a micro-delay that can reduce impulsive responding.
• Immediate error feedback: Digital math platforms provide instantaneous corrective feedback, which may strengthen error-monitoring processes impaired in ADHD [39].
• Self-monitoring dashboards: Schuck et al. [42], although focused on behavior rather than mathematics, demonstrated that self-monitoring applications using iPad improved classroom behavior among students with ADHD; this design principle may also be transferable to mathematics learning contexts.

3.3. Reported Intervention Outcomes (RQ3)

The included sources reported a wide range of outcomes spanning mathematics achievement, cognitive functioning, ADHD-related behavior, engagement and motivation, and socio-emotional domains. Consistent with the scoping-review objective, these outcomes were mapped descriptively to characterize what has been measured across the evidence base rather than to estimate an overall intervention effect. In accordance with RQ3, the analysis mapped the types of outcomes assessed, the constructs measured, and the instruments used across the included evidence. It did not seek to determine whether the interventions were effective or to estimate comparative intervention effects.

3.3.1. Outcome Measurement Landscape

The reviewed studies operationalized intervention-related findings across five broad outcome domains, although substantial heterogeneity was evident in construct definitions, instruments, timing, and analytic approaches. Mathematics achievement constituted the primary outcome domain, measured in 28 studies (75.7%).

Within this domain, researchers employed a tripartite measurement strategy: (a) standardized norm-referenced achievement tests, including the Woodcock-Johnson III Tests of Achievement [5,38], the Wechsler Individual Achievement Test [10], and curriculum-based measurement probes [39]; (b) researcher-developed mathematics assessments tailored to specific intervention content, as deployed by Mohammadhasani et al. [13] in their pedagogical agent study and by Sperafico et al. [9] in their combined working memory and arithmetic reasoning trial; and (c) teacher-rated academic performance scales, which—while ecologically valid—introduce rater bias and lack the psychometric precision of standardized instruments [3]. Only 12 studies (32.4%) employed standardized, validated mathematics achievement measures, while the remaining studies relied primarily on researcher-developed assessments, curriculum-based measures, or rating-based indicators. This variation illustrates substantial heterogeneity in how mathematics achievement was operationalized across the evidence base.

Cognitive outcomes were measured in 19 studies (51.4%), reflecting the field’s theoretical grounding in cognitive processing models of ADHD-related mathematical underachievement. Working memory was the most frequently assessed cognitive construct, operationalized through digit span tasks (forward and backward), n-back paradigms, and complex span measures [40,43]. Sustained attention was assessed via continuous performance tests, including the Test of Variables of Attention [6] and the Conners’ Continuous Performance Test [8]. Processing speed, though implicated in ADHD-mathematics comorbidity models [1], was measured in only 4 studies, representing significant assessment gaps given emerging evidence that processing speed mediates the ADHD-mathematics achievement relationship.

Behavioral outcomes, including ADHD symptom severity, on-task behavior, and classroom conduct, were measured in 16 studies (43.2%). The Conners-3 Rating Scales (parent and teacher versions) and the SNAP-IV constituted the dominant instruments [20,38]. Direct behavioral observation coding systems, while offering superior ecological validity, were employed in only 3 studies, with the majority relying on adult-informant rating scales that may conflate attentional improvement with halo effects from academic gains.

Engagement and motivation outcomes, measured in 11 studies (29.7%), remain theoretically underdeveloped in this literature. Most studies operationalized engagement through proxy indicators—time-on-task, task completion rates, or teacher-reported persistence—rather than validated self-report instruments such as the Math Anxiety Rating Scale or the Situational Motivation Scale [44]. This is particularly problematic for ADHD populations, where behavioral engagement (observable task-focused behavior) and emotional engagement (interest, belonging, intrinsic motivation) may dissociate; a student may appear behaviorally engaged due to gamified reward structures while experiencing low intrinsic motivation for mathematics. These findings indicate that engagement and motivation were assessed inconsistently across the literature, with standardized measures used relatively infrequently.

Social and socio-emotional outcomes were the least measured domain (n = 3; 8.1%), reflecting the individual-task orientation of most digital mathematics interventions. This omission is noteworthy given that peer-mediated learning and collaborative problem-solving have demonstrated efficacy for students with ADHD in non-digital contexts [3], yet remain largely unexplored in technology-based mathematics instruction. The limited assessment of these outcomes represents an important gap in the coverage of measured outcomes across the evidence base. The diversity of instruments and outcome definitions limits direct comparison across sources and indicates an important characteristic of the current evidence landscape. This heterogeneity should not be interpreted as evidence for or against intervention effectiveness; rather, it limits the comparability and standardization of outcome measurement across the available evidence.

3.3.2. Outcome Reporting Across Evidence Tiers

The reported findings were examined across the four-tier evidence map to determine how directly the available sources addressed the review’s PCC-defined intersection. Effect sizes and statistical findings reported by individual studies are presented descriptively where available; they are not pooled or interpreted as estimates of overall effectiveness across the evidence base. The distribution of assessed outcomes varied across the four evidence tiers. Tier 1 studies most frequently assessed mathematics achievement together with selected ADHD-related or cognitive outcomes, whereas Tier 2 studies more commonly combined academic outcomes with measures of attention, working memory, or behavior. Tier 3 sources primarily assessed mathematics achievement and related academic outcomes, although ADHD-specific outcome reporting was less consistent. Tier 4 sources, which were predominantly reviews, frameworks, and technical studies, more commonly discussed potential educational, cognitive, or personalization-related outcomes than directly measured student outcomes. These differences indicate variation in the scope and type of outcomes represented across the evidence map. They should be understood as characteristics of the available evidence rather than as evidence of comparative intervention effectiveness.

Tier 1—Direct Intersection Studies: Empirical studies that simultaneously addressed ADHD, technology-based learning, and mathematics outcomes reported a range of positive, mixed, or descriptive findings across different technologies and outcome measures. Mohammadhasani et al. [13] reported a large effect (d = 0.82) of their pedagogical agent “Koosha” on 5th-grade mathematics achievement. Sperafico et al. [9] reported a medium-to-large effect (d = 0.65) for working memory and arithmetic reasoning program delivered via computer. Herzog and Casale [39] reported a medium effect (d = 0.45) for a computer-based mathematics intervention among primary school students with emotional and behavioral difficulties, including ADHD. Botsas and Grouios [4] reported a medium effect (d = 0.55) favoring online mathematics instruction over traditional methods for Greek students with ADHD.

Rabiner et al. [5], in their landmark RCT of two computer-based interventions for students with attention difficulties, found that while both programs improved attentional performance, effects on Broad Mathematics scores on Woodcock-Johnson III were small (d = 0.20–0.30) and did not reach statistical significance. Steiner et al. [6] reported similarly modest academic effects from school-based computer attention training, with math teacher ratings showing improvement but standardized achievement tests remaining unchanged. Dai et al. [8] reported significant improvements in sustained attention (p < 0.05) and academic performance through an 8-week gamified educational application. Kusmawati et al. [7] found significant improvement in concentration and math achievement for ADHD students using gamified learning media. Shafirarossa et al. [34] demonstrated improved early numeracy skills through color-supported visual scaffolding in a single-subject design, while Lomibao and Tabor [32] provided descriptive evidence of improved math engagement and reduced anxiety through individualized online instruction for a learner with comorbid ADHD and math SLD.

Across Tier 1 studies (

Table 1. Characteristics of Tier 1 Studies (n = 10).

Study	Design	N (ADHD)	Technology	Reported Result
Mohammadhasani et al. [13]	Quasi-experimental	30 (30)	Pedagogical agent (Koosha)	d = 0.82
Dai et al. [8]	RCT	80 (80)	Gamified app	p < 0.05
Rabiner et al. [5]	RCT	77 (77)	CBT attention training	d = 0.20–0.30 (ns)
Steiner et al. [6]	Preliminary trial	41 (41)	CBT attention training	Mixed
Herzog & Casale [39]	Quasi-experimental	11 in total	CBT math intervention	d = 0.45
Sperafico et al. [9]	Quasi-experimental	46 (46)	CBT (WM + math)	d = 0.65
Lomibao & Tabor [32]	Case study	1 (1)	Online platform	Descriptive
Shafirarossa et al. [34]	Single-subject	1 (1)	Computer-assisted instruction	Visual analysis
Botsas & Grouios [4]	Quasi-experimental	18 (6 ADHD students + 12 non-disabled peers)	Online math instruction	d = 0.55
Kusmawati et al. [7]	Quasi-experimental	13 (4)	Gamified media	p < 0.05

Note 1. ADHD = attention-deficit/hyperactivity disorder; CBT = cognitive behavioral therapy; RCT = randomized controlled trial; WM = working memory. Note 2. Full extraction details, including age range, outcome measures, and comparator conditions, are provided in Appendix D, Table A4.

), the outcome measures and effect sizes reported varied widely by technology and design. Given the heterogeneity of designs, populations, technologies, comparators, and outcome measures, these values cannot be pooled or read as a generalizable effect for students with ADHD. The small sample sizes across Tier 1 studies (median n = 24; range = 1–80) and the predominance of quasi-experimental designs (n = 5 of 10) further characterize the direct evidence base as relatively small and widespread. Tier 1 comprises primary empirical studies, whereas Tiers 2 and 3 include a mix of primary empirical studies, reviews, meta-analyses, and contextual sources, and Tier 4 comprises reviews, frameworks, and technical studies. Values in the “Reported Result” column (in Table 1,

Table 2. Characteristics of Tier 2 Studies (n = 8).

Study	Design	N (ADHD)	Technology	Reported Result
Wiest et al. [10]	Experimental	43 (26 reading SLD, 6 writing SLD, 4 math SLD, remainder without SLD)	CBT cognitive training	p < 0.05
Wong et al. [20]	Meta-analysis	N/A	Various technology-based	SMD = 0.42
Chacko et al. [38]	RCT	85 (85)	Cogmed WM training	Near-transfer only
Zentall et al. [45]	Review	Varies	Technology Sensory + medication	Varies
DuPaul et al. [3]	Review	N/A	Various classroom-based	Varies
Yegencik et al. [2]	Meta-analysis	N/A	Various school-based	SMD = 0.35
Wolters et al. [46]	Systematic review	N/A	Academic-skills training	Varies
Gunnars [37]	Systematic review	N/A	Digital attention tools	Varies

Note 1. ADHD = attention-deficit/hyperactivity disorder; SLD = specific learning disorder; CBT = cognitive behavioral therapy; SMD = standardized mean difference; WM = working memory. Note 2. Full extraction details are provided in Appendix D, Table A5.

and

Table 3. Characteristics of Tier 3 Studies (n = 10).

Study	Design	n	Technology	Reported Result
Benavides-Varela et al. [11]	Meta-analysis	1073 (N/A)	Digital intervention (multiple)	g = 0.37
Sánchez-Pérez et al. [40]	Quasi-experimental	104 (N/A)	CBT (math + WM)	d = 0.50
Conesa & Duñabeitia [43]	Quasi-experimental	120 (35 ADHD)	Game-based training	d = 0.35
Stalmach et al. [47]	Systematic review	N/A	Digital learning	Varies
Kim & Xin [22]	Meta-analysis	N/A	Technology-based intervention	g = 1.18
Myers et al. [23]	Meta-analysis	N/A	Various (technology subset)	g = 0.45
Monei & Pedro [21]	Systematic review	N/A	Various (technology subset)	Varies
Bouck et al. [33]	Practitioner article	N/A	Online instruction	N/A
Scalise et al. [41]	Review	N/A	Digital STEM assessments	N/A
Polydoros & Antoniou [44]	Review	N/A	Serious digital games	N/A

Note 1. ADHD = attention-deficit/hyperactivity disorder; CBT = cognitive behavioral therapy; WM = working memory. Note 2. Full extraction details are provided in Appendix D, Table A6.

) are reported by individual studies and are presented descriptively; they are not pooled or interpreted as an estimate of overall effectiveness across studies.

Tier 2—Technology × ADHD with Academic Outcomes: Review evidence from this tier reports a broader range of outcome measures (Table 2). Wong et al. [20], in their systematic review and meta-analysis of technology-based interventions for school-age children with ADHD, reported a small-to-medium standardized mean difference for attention outcomes (SMD = 0.42, 95% CI [0.28, 0.56]), but noted that academic transfer effects were inconsistent and typically smaller than near-transfer cognitive effects. Wiest et al. [10] reported significant improvements in WM, attention, and impulsivity (p < 0.05) following 20 weeks of cognitive training, with reading and math achievement also improving; however, the absence of a control group limits causal inference. Chacko et al. [38], in their examination of Cogmed working memory training, found that while near-transfer working memory effects were robust, far-transfer effects to mathematics achievement were negligible and non-durable. Table 2 below includes primary studies, reviews, and meta-analyses; source type is indicated for each entry.

Tier 3—Technology × Mathematics with ADHD as Comorbidity: Evidence from populations with mathematics learning difficulties, in which ADHD is a documented comorbid condition, provides the largest evidence base but the least ADHD-specific information (Table 3). Benavides-Varela et al. [11], in their meta-analysis of digital-based interventions for children with mathematical learning difficulties, reported a small-to-medium overall effect (g = 0.37, 95% CI [0.24, 0.50]) on mathematics achievement. Critically, they found that effect sizes were significantly larger when digital tools were compared to no-treatment controls (g = 0.52) than when compared to active teacher-led instruction (g = 0.18). It is a comparator-dependent pattern reported here descriptively as a characteristic of the evidence base. Kim and Xin [22], in their meta-analysis of technology-based word-problem interventions for students with disabilities (including ADHD), reported a larger mean effect size (g = 1.18), though ADHD-specific subgroup analysis was not conducted. Sánchez-Pérez et al. [40] found that a school-based computer training program combining mathematics and working memory exercises produced significant improvements in both cognitive skills and academic achievement in primary school children, with a medium effect on mathematics (d = 0.50). Conesa and Duñabeitia [43] reported comparable effects (d = 0.35) for game-based computerized training on executive functions and academic achievement. Across Tier 3 studies, mathematics-embedded technologies were represented in several sources and were associated with reported mathematics outcomes, whereas ADHD-specific effects were generally unavailable or insufficiently reported. Tier 3, therefore, provides contextual evidence concerning technology-supported mathematics learning in populations with mathematical learning difficulties or broader special educational needs.

Tier 4—AI-Specific Reviews: The AI literature mapped in Tier 4 was predominantly composed of reviews, frameworks, technical studies, and conceptual work rather than directly evaluating AI-mediated mathematics interventions involving students with ADHD (

Table 4. Characteristics of Tier 4 Studies (n = 9).

Source	Type	Technology Focus	Key Finding
Barua et al. [30]	Systematic review	AI/ML/DL assistive tools	AI shows promise; ADHD-math evidence sparse
Yang et al. [15]	Systematic review	Various AI	ITS dominant in math; ADHD underrepresented
Ronksley-Pavia et al. [16]	Scoping review	Generative AI/LLM	No empirical GenAI-ADHD-math studies found
Seung et al. [27]	Scoping review	AI tutoring systems	AI tutoring most researched in STEM broadly
Lessing & Ogbonnaya [14]	Review	AI in math education	ITS most common tool for SEN
Bhushan et al. [19]	Survey/Review	ML/DL screening	ADHD comorbidity noted; no instructional AI
Rehman et al. [18]	Technical study	XAI/deep learning (diagnosis)	F1 = 99% binary; 94.2% multi-class
De Silva et al. [17]	Technical study	ML classifiers (identification)	81% eye-movement; 82% fMRI
Patil et al. [36]	Review	Neurofeedback/EEG	Math outcomes discussed; limited direct evidence

Note: 1. ADHD = attention-deficit/hyperactivity disorder; ML = machine learning; DL = deep learning; LLM = large language model; XAI = explainable artificial intelligence; EEG = electroencephalography; ITS = intelligent tutoring system; fMRI = functional magnetic resonance imaging; SEN = special educational needs. Note 2. Full extraction details are provided in Appendix D, Table A7.

). Tier 4 comprises reviews, frameworks, and technical studies; no primary empirical intervention data are included. Barua et al. [30], in their highly cited review synthesized AI-enabled assistive tools for neurodevelopmental disorders and identified personalized learning algorithms, intelligent tutoring systems, and adaptive feedback mechanisms as promising directions, but noted that empirical validation with ADHD learners in mathematics was absent. Yang et al. [15], in their decade review of AI in special education (2013–2023), identified intelligent tutoring systems as the most frequently utilized AI technology in mathematics education for students with special needs, yet acknowledged that ADHD-specific evidence remained sparse. Ronksley-Pavia et al. [16], in their scoping review of generative AI for neurodivergent students, proposed that large language models could provide personalized mathematical explanations and scaffolded problem-solving support, but reported limited empirical evidence for specific neurodivergent populations and educational outcomes. This tier thus maps a substantial conceptual and technological interest in AI-supported personalization, adaptive feedback, and intelligent tutoring, but comparatively limited direct evidence aligned with all three PCC elements of the present review.

3.3.3. Observed Variation in Reported Effect Sizes

Reported effect sizes varied substantially across studies and were not directly comparable because of differences in study design, population, intervention, comparator, outcome measure, and evidence tier. No cross-study comparisons by developmental level, intervention duration, comorbidity status, or technology type were undertaken. These variables are retained as descriptive characteristics of the evidence base and as potential questions for future confirmatory research.

3.3.4. Durability and Transfer: Evidence Gaps

The durability and generalizability of reported findings were poorly represented in the evidence map. Only 3 out of 37 studies (8.1%) incorporated follow-up assessments beyond the immediate post-intervention period, with the longest follow-up extending to 6 months (Rabiner et al. [5]). This limited reporting prevents firm conclusions regarding the maintenance or transfer of reported gains beyond the intervention period. Rabiner et al. [5] explicitly noted that while their computer-based attention training produced immediate cognitive gains, these did not sustain at 6-month follow-up. Chacko et al. [38] similarly reported robust near-transfer working-memory effects that were not accompanied by durable far-transfer effects on mathematics achievement. The theoretical implications of this transfer pattern for intervention design are addressed in the Discussion.

3.4. Evidence Gaps and Directions for Subsequent Research (RQ4)

This scoping review identified seven critical gaps in the evidence base concerning the use of technology for mathematics education among students with ADHD. These gaps emerged from the mapping of the PCC dimensions and the characteristics of the included sources rather than from formal quality appraisal or evidence grading. They relate to the limited direct empirical evidence on AI, the lack of longitudinal and secondary-school research, insufficient specificity in mathematical domains, limited attention to comorbidity, methodological heterogeneity, and unresolved ethical and implementation issues. These limitations are consistent with broader reviews showing that technology-based interventions for students with ADHD remain heterogeneous in their targets, technologies, and outcomes [20,47]. Therefore, these gaps indicate areas in which the current evidence base does not yet adequately cover the Population, Concept, and Context defined by the review.

3.4.1. Limited Direct Evidence at the AI × Mathematics × ADHD Intersection

Within the sources identified and eligible under the present search and inclusion criteria, the review identified no directly aligned empirical study evaluating an AI-driven instructional intervention specifically designed to teach mathematics to students with ADHD. The identified AI literature consisted primarily of diagnostic applications [17] Rehman et al. [18], broader reviews of AI in special education [15,30], or design proposals without empirical validation [29]. AI applications are represented in the literature primarily through diagnostic applications, broader reviews, conceptual frameworks, and design proposals, while directly aligned empirical studies at the intersection of AI, mathematics, and ADHD remain limited. This gap concerns the limited convergence of the three PCC dimensions rather than the absence of AI-related educational research more broadly. Future studies should therefore investigate AI-supported mathematics instruction in clearly defined ADHD populations and educational contexts, with explicit reporting of the mathematical content, intervention characteristics, and assessed outcomes.

The above finding represents the clearest gap at the intersection of the three PCC elements: the Population of students with ADHD, the Concept of AI-supported mathematics instruction, and the educational Context in which such instruction is delivered. Studies should move beyond conceptual discussions and evaluate AI-supported mathematics instruction using rigorous experimental designs, with outcomes that capture both mathematical achievement and ADHD-related learning processes. Previous research on computer-assisted mathematics instruction and pedagogical agents suggests that technology may support mathematical learning and engagement among students with ADHD [4,13], while broader evidence indicates that technology-based interventions can produce improvements in selected ADHD-related outcomes [20]. Particular attention should be given to whether adaptive feedback, personalization, and real-time support can address specific barriers such as inattention, working-memory demands, and difficulties with self-regulation.

3.4.2. Limited Longitudinal and Follow-Up Evidence

Long-term evidence was notably limited, with only 8.1% of studies including follow-up assessments. Consequently, the evidence map contains limited information regarding maintenance of reported outcomes or transfer to subsequent mathematical learning after the intervention period. This limitation is consistent with the broader technology-based ADHD literature, in which intervention outcomes have been evaluated primarily over relatively short periods, leaving questions regarding maintenance and generalization insufficiently addressed [20,47]. This limitation is particularly important for students with ADHD, for whom difficulties with attention and self-regulation may persist across academic contexts and over time.

Studies should therefore incorporate longitudinal follow-up, ideally extending to at least six months, to examine the maintenance and transfer of intervention effects. Longitudinal designs would also help distinguish temporary improvements associated with intensive intervention exposure from durable changes in academic functioning, particularly given evidence that technology-supported interventions can influence attention and executive-function outcomes without necessarily demonstrating broader transfer [20,48]. Booster sessions and repeated assessments could further determine whether continued technological support is necessary to sustain gains or whether students can independently apply the acquired strategies.

3.4.3. Limited Representation of Secondary-School Populations

The evidence base was strongly concentrated on younger learners, with 28 studies (75.7%) focusing on elementary-school students aged approximately 6–12 years. Only five studies (13.5%) explicitly included adolescents, and none focused specifically on high-school or transition-age students. Thus, the Population domain of the PCC framework was substantially less represented at the secondary-school level. This imbalance limits the applicability of current findings to secondary education, where mathematics increasingly involves algebra, geometry, abstract reasoning, and complex multi-step problem-solving. This gap is notable because research on mathematics interventions for adolescents indicates that secondary students require interventions addressing increasingly complex mathematical content and reasoning demands [23], while technology-supported mathematics instruction has also been explored specifically in secondary students with learning disabilities [33]. The underrepresentation of Global South populations is equally pronounced. Despite UNESCO [49] calls for equitable educational technology access, the reviewed literature was dominated by North American, European, and East Asian contexts, with only limited evidence from South American, African, and South Asian settings where ADHD prevalence and mathematics underachievement are similarly pressing concerns.

The lack of adolescent-focused research is particularly important because the interaction between ADHD-related difficulties and mathematical demands may change as academic complexity increases. Adolescence also brings increasing demands for independent planning, sustained attention, self-monitoring, and self-regulated learning, which are particularly relevant to students with ADHD [1,3]. Future studies should therefore investigate technology-based interventions within authentic secondary mathematics curricula and examine whether different forms of scaffolding or adaptive support are required for adolescents compared with younger learners.

3.4.4. Limited Specificity in Mathematical Domains

Another important finding concerns the lack of specificity in the mathematical domains addressed by the reviewed studies, as illustrated in Figure 6. The largest category consisted of studies reporting non-specific or “other” mathematics outcomes (n = 11, 29.7%), which included diagnostic studies, screening tools, reviews with non-specific math targets, and general cognitive outcomes. Studies measuring broad math achievement constituted the second largest category (n = 9, 24.3%), while engagement in math (n = 8, 21.6%) and word problem solving (n = 7, 18.9%) were also represented. More specific domains were considerably less represented: only one study each addressed arithmetic reasoning, number sense, and math computation. Consequently, 29.7% of the literature did not clearly identify a specific mathematical subdomain being targeted, making it difficult to determine whether technology is differentially effective for procedural skills, conceptual understanding, or higher-order mathematical reasoning. This lack of specificity is important because evidence from mathematics intervention research indicates that instructional effects can depend on the targeted mathematical skill and the instructional components employed [11,50].

The previously reported figures indicated that 40% of studies did not specify mathematical subdomains; however, the corrected extraction shows that 29.7% reported non-specific outcomes, with an additional 24.3% using broad achievement measures that also lack subdomain specificity. Combined, over half the corpus (54.0%) does not permit conclusions about which mathematical skills are most amenable to technology-based support. Existing reviews similarly demonstrate that technology-based mathematics interventions have been developed around particular skills, including arithmetic, number understanding, and word-problem solving, rather than representing mathematics as a single homogeneous construct [11,22]. Future research should therefore identify mathematical content more precisely and examine intervention effects within specific domains. Such specificity would allow researchers to determine whether particular technologies are better suited to procedural learning, conceptual development, or complex problem-solving.

3.4.5. Insufficient Characterization of Comorbidity

Comorbidity represents another important gap. Although ADHD and mathematics learning disability (MLD), including dyscalculia, frequently co-occur, only eight studies (21.6%) explicitly addressed participants with comorbid profiles. Across the included sources, participant characteristics were not consistently reported at a level that allowed the evidence to be reliably mapped according to ADHD-only versus ADHD-plus-comorbidity profiles. The relationship between ADHD and mathematical difficulties is well documented, with research indicating that mathematical difficulties and ADHD symptoms can co-occur and may involve overlapping difficulties in basic numerical skills and executive functioning [12]. Most interventions treated ADHD-related difficulties and mathematics difficulties as relatively separate targets, providing limited evidence about how technology can support students experiencing both attentional and mathematical learning difficulties.

This distinction has direct implications for intervention design. Students with ADHD and MLD may require different forms of technological scaffolding from students with ADHD alone, particularly when attentional difficulties interact with deficits in number processing or mathematical reasoning. Research on interventions for dyscalculia and mathematics learning difficulties similarly emphasizes the importance of targeting specific mathematical processes rather than relying exclusively on broad cognitive training [11,21].

3.4.6. Methodological Limitations and Reporting Heterogeneity

The evidence base was characterized by substantial methodological and reporting heterogeneity. Tier 1 empirical studies had a median sample size of 24 participants (range = 1–80; n = 10), reducing statistical power and limiting the reliability of subgroup analyses. Three RCTs were identified across all tiers: two in Tier 1 (Dai et al. [8]; Rabiner et al. [5]) and one in Tier 2 (Chacko et al. [38]). Only one Tier 1 RCT (Dai et al. [8]) examined a gamified mathematics intervention with an active control condition. Furthermore, most studies used no-treatment or treatment-as-usual comparisons rather than active controls, making it difficult to distinguish technology-specific effects from increased instructional time, attention, or novelty.

Because this was a scoping review, these characteristics are reported to describe the evidence landscape rather than to provide a formal risk-of-bias or methodological-quality ranking. Considerable heterogeneity in outcome measures further limits comparison across studies and prevents meaningful meta-analytic synthesis of the direct-intersection evidence. Similar concerns regarding heterogeneity in intervention targets and outcome measures have been reported in broader reviews of technology-based interventions for students with ADHD and students with special educational needs [20,47]. Future research should prioritize RCTs with active control conditions and standardized mathematics and ADHD-related outcome measures. Clear reporting of intervention dosage, instructional content, participant characteristics, and fidelity of implementation would also strengthen the comparability and reproducibility of findings.

3.4.7. Ethical and Implementation Considerations

The included literature provided limited reporting on ethical and implementation characteristics relevant to emerging AI-based educational technologies. In particular, relatively little information was identified concerning data privacy, informed consent, algorithmic transparency, bias, teacher training, infrastructure, accessibility, cost, and scalability. These issues were therefore identified as evidence gaps rather than as outcomes of the included interventions [15,16,51]. These concerns are especially relevant for students with ADHD, whose behavioral patterns may be continuously collected or inferred by adaptive systems. Ethical safeguards should therefore be incorporated into intervention design rather than considered only after implementation. However, the present scoping review does not provide sufficient evidence to determine how frequently or systematically these safeguards are implemented across the field.

Practical implementation was similarly underexamined, with limited attention to teacher training, AI literacy, infrastructure, accessibility, cost, and scalability. This is consistent with broader research emphasizing that the effectiveness of digital interventions depends not only on their technological features but also on accessibility, instructional integration, and the capacity of educators to use them appropriately [41,47]. Future interventions should evaluate not only whether technology improves student outcomes but also whether it can be implemented sustainably in real classrooms. A particularly promising direction is a teacher-mediated model in which AI provides individualized feedback or recommendations while teachers retain responsibility for instructional decisions and pedagogical judgment. This human-in-the-loop model is consistent with recent frameworks for AI in special education that emphasize supportive and augmentative roles for technology, positioning teachers as responsible for interpreting AI-generated information and making instructional decisions while AI supports personalization and other data-intensive tasks [52,53,54]. Such an approach is consistent with the broader view of AI as augmenting rather than replacing teacher expertise in educational settings [15,51]. The mapping therefore suggests a need for future studies to report not only learner outcomes but also contextual and implementation characteristics that determine how technology-supported mathematics interventions function in authentic educational settings. Overall, other studies should prioritize mathematically specific and individualized interventions within authentic secondary-school contexts and evaluate their effectiveness alongside long-term outcomes, transfer, ethical safeguards, and implementation feasibility.

3.4.8. PCC-Based Synthesis of the Evidence Map

When mapped across the JBI PCC framework, the evidence base showed an uneven distribution across the three dimensions. The Population domain was dominated by children in elementary-school settings, with relatively limited representation of adolescents and inconsistent characterization of ADHD presentation and comorbid learning difficulties. The Concept domain was characterized by substantial technological heterogeneity, including computer-based training, gamification, online learning, pedagogical agents, adaptive technologies, and emerging AI-related applications; however, direct evidence concerning AI-supported mathematics instruction for students with ADHD remained limited. The Context domain included school-based, online, and specialized educational settings across multiple geographical locations, but reporting of contextual characteristics was inconsistent.

Overall, the evidence map demonstrates that the literature surrounding technology, ADHD, and mathematics is considerably broader than the directly aligned evidence at the intersection of all three PCC dimensions. The principal gap is therefore not simply a lack of technology-based mathematics research or a lack of AI research, but the limited convergence of a clearly defined ADHD population, a mathematics-specific technology/AI concept, and an educational context in which the intervention is empirically evaluated.

4. Discussion

This scoping review examined the intersection of technology-based mathematics education and ADHD across 37 peer-reviewed studies identified through a multi-database search and supplementary citation chasing and hand-searching. Given the heterogeneity of the literature and the limited number of studies addressing the exact intersection of ADHD, mathematics, and emerging AI, the findings are best interpreted as a map of the current evidence base rather than as an estimate of intervention effectiveness. The discussion is therefore organized around the four research questions, focusing on the types of technologies investigated, the ADHD-related barriers addressed by these technologies, the outcomes and characteristics of the available evidence, and the gaps that should guide future research.

4.1. Types of TBI and Tools (RQ1)

RQ1 examined the types of technology-based interventions used to support mathematics learning among students with ADHD and the extent to which these interventions incorporated AI. The review identified a diverse technology landscape that included computer-based instruction and training, online mathematics interventions, gamified applications, pedagogical agents, digital learning environments, augmented reality, and technologies designed to support attention or executive functioning. However, the literature was strongly concentrated on conventional or rule-based digital technologies, whereas genuinely AI-driven mathematics interventions for students with ADHD remained largely absent.

This finding is consistent with broader evidence showing that technology-based interventions for children with ADHD have predominantly focused on structured digital training, behavioral support, attention, executive functioning, and gamification rather than advanced AI-supported instructional adaptation. For example, Wong et al. [20] reported considerable variation in technology-based interventions for school-age children with ADHD, while Gunnars [37] similarly identified digital technology as an emerging component of interventions targeting attention and executive functioning. Systematic reviews of digital therapeutics for ADHD further confirm that the evidence base remains concentrated in cognitive training and gamification, with AI-adaptive mathematics instruction still largely absent [20,37]. In mathematics specifically, Botsas and Grouios [4] demonstrated the use of computer-assisted instruction for mathematical operations among students with ADHD, whereas Herzog and Casale [39] reported positive effects of a computer-based mathematics intervention among students with emotional and behavioral difficulties. These studies indicate that technology-mediated mathematics instruction for learners with attention-related difficulties has an established history, whereas the principal unresolved issue concerns the empirical development and evaluation of newer AI-based approaches.

This review also identified evidence for more interactive forms of technology. The pedagogical-agent intervention examined by Mohammadhasani et al. [13] is particularly relevant because the virtual tutor was embedded directly within a mathematics learning environment for students with ADHD. The findings suggest that technology can extend beyond automated practice by incorporating instructional guidance and elements of social presence. This example illustrates how digital systems can integrate instructional guidance and interactive support directly into mathematics learning. Similarly, although evidence remains preliminary, gamified mathematics technologies have been developed specifically for students with ADHD, including a pilot study reported by Castro and Huamanchahua [31]. More recent work has extended this direction through gamification, with Dai et al. [8] reporting an 8-week randomized controlled trial with an active control condition (non-gamified digital platform matched for content and time), significant improvements in attention and academic performance (p < 0.05), and maintained gains at an 8-week follow-up. Kusmawati et al. [7] reported increased attention among students following the use of gamified learning media.

The evidence therefore suggests progression from computer-assisted practice toward increasingly interactive, adaptive, and multimodal learning environments. However, this progression should not be equated with the empirical validation of AI. Reviews of AI in special education describe considerable potential for personalized instruction, learner modelling, prediction, and adaptive support [15,30], while recent reviews have specifically examined generative AI for neurodivergent learners [16] and AI integration in K-12 education more broadly [27]. Nevertheless, the present review indicates that these developments have not yet been translated into sufficiently developed empirical literature testing AI-driven mathematics instruction specifically with students with ADHD. This distinction is important because an adaptive digital system and an AI system are not necessarily synonymous. Adjusting task difficulty according to previous accuracy, for example, represents a relatively basic form of adaptivity and should not automatically be interpreted as AI. The literature therefore currently demonstrates a stronger evidence base for digital and adaptive technologies than for AI-specific mathematics interventions. The principal implication of RQ1 is consequently not that AI has been shown to be effective for mathematics learning in ADHD, but that there is a substantial gap between the technological possibilities described in AI literature and their empirical application to this specific population and domain.

4.2. ADHD-Related Cognitive Barriers and Associated Design Features (RQ2)

RQ2 examined which ADHD-related learning and cognitive difficulties were explicitly targeted or considered by technology-based interventions. Across the reviewed studies, the most recurrent targets were inattention and sustained attention, working-memory demands, and impulsivity or inhibitory-control difficulties, with additional attention to self-regulation and motivational or engagement-related difficulties. These findings should be understood as a mapping of the difficulties addressed in literature rather than as evidence that technology directly modifies the underlying characteristics of ADHD. Although design features such as feedback, multimodal presentation, personalization, and motivational support were frequently described within individual interventions, these features are interpreted here as possible ways of responding to identified learning difficulties rather than as separate findings of RQ2. These features correspond closely to difficulties commonly associated with academic functioning in ADHD, particularly attention regulation, working memory, processing speed, persistence, and self-regulation.

The importance of these mechanisms is supported by research linking working memory and processing speed to academic achievement among children with ADHD. Hulsbosch et al. [1], for example, identified processing speed and working memory as relevant factors in academic performance. These findings provide a rationale for technologies that externalize information, reduce unnecessary working-memory demands, or provide immediate support during mathematical problem-solving, including approaches that incorporate ML/DL techniques [19]. Similarly, DuPaul et al. [3] emphasized the importance of structured instructional and behavioral supports within classroom interventions for students with ADHD. The interpretation of the mapped learning difficulties therefore points toward an accommodation-oriented role for technology. Rather than assuming that digital interventions necessarily remediate the underlying cognitive characteristics associated with ADHD, technology may instead alter the learning environment in ways that reduce unnecessary demands, increase access to task information, and support students during mathematical activity.

Immediate feedback was one of the most prominent mechanisms identified in the review. Digital environments can provide information about correctness, errors, and next steps immediately after a response, thereby reducing the delay between performance and reinforcement. This is theoretically compatible with models emphasizing altered reinforcement mechanisms in ADHD [55]. However, the available evidence does not establish that more frequent or immediate feedback is necessarily superior, as its educational value is likely to depend on timing, informational content, and its alignment with mathematical reasoning. Immediate feedback may support persistence and error correction, but excessive prompts may also interrupt independent reasoning or create dependence on external cues. Thus, the relevant design principle is not simply more feedback, but appropriately timed and instructionally meaningful feedback. However, this finding does not establish immediate feedback as a superior intervention component. Its educational value is likely to depend on how feedback is timed, what information it provides, and whether it supports rather than interrupts mathematical reasoning.

A second mechanism concerned the use of visual, interactive, and multimodal representations to direct and sustain attention. This finding is consistent with the use of pedagogical agents, gamified interfaces, augmented reality, and visually supported instructional designs in the reviewed literature. For example, Mohammadhasani et al. [13] used a pedagogical agent to direct students’ attention during mathematics learning, while Al-Bleheshi et al. [35] described an augmented-reality mathematics learning system specifically designed for students with ADHD. However, multimodality should not automatically be interpreted as an advantage. Cognitive load theory emphasizes that additional information can increase extraneous cognitive load when it is not directly relevant to the learning task [56]. Similarly, principles of multimedia learning emphasize the importance of selecting and coordinating representations rather than simply increasing the number of modalities [57]. The implication is that technology should use multimodal information strategically to direct attention toward mathematically relevant information, rather than increasing stimulation for its own sake.

Personalization and adaptivity represented a third mechanism, but the evidence was substantially less mature than the theoretical discussion surrounding AI. Existing technologies frequently personalize learning by modifying difficulty, sequencing tasks, or providing different levels of support. Most of this personalization reflects rule-based adaptivity (e.g., adjusting item difficulty after a fixed number of errors) rather than AI-driven learner modelling that dynamically infers attention, working memory, or cognitive load from multimodal data. Reviews of AI-enabled educational technologies describe the potential for learner modelling and individualized support [15,30], but the current ADHD–mathematics literature provides little empirical evidence that such systems can reliably detect moment-to-moment changes in cognitive or attentional state. Consequently, personalization represents a promising design principle, but the current evidence does not yet establish AI-based personalization as a validated mechanism for mathematics learning among students with ADHD.

The fourth mechanism was motivational and engagement support, particularly through gamification. This finding is consistent with studies reporting benefits of gamified learning for students with ADHD. Kusmawati et al. [7] reported increased attention through gamification, while Dai et al. [8] found benefits associated with an 8-week gamified educational application. Evidence from students with learning disabilities also suggests that serious digital games may support mathematics performance and motivation [44]. Nevertheless, the theoretical explanation should extend beyond simple reward provision. Self-determination theory emphasizes autonomy, competence, and relatedness as important conditions for sustained motivation [58]. Thus, the relevance of gamification may depend less on the presence of rewards themselves than on whether technology creates meaningful opportunities for competence, autonomy, and sustained engagement with mathematical tasks.

Overall, considering both the barriers targeted under RQ2 and the design-feature discussion above, the strongest rationale for technology in ADHD mathematics education is not that technology directly changes the underlying characteristics of ADHD. Rather, technology can modify the learning environment around those characteristics by reducing unnecessary cognitive demands, directing attention, providing timely feedback, supporting persistence, and making mathematical information more accessible. This interpretation is consistent with an accommodation-oriented perspective on disability [59] and contrasts with approaches that assume technology will necessarily remediate domain-general cognitive deficits.

4.3. Reported Outcomes and Characteristics of Evidence (RQ3)

RQ3 examined the outcomes reported in the literature and the characteristics of the available evidence. The reviewed studies reported outcomes across several domains, including mathematics achievement, procedural performance, problem solving, attention, working memory, engagement, motivation, and self-regulation. However, the evidence was varied, and positive findings were not consistently observed across all outcome domains or intervention types.

One important pattern concerns the distinction between domain-specific mathematical outcomes and domain-general cognitive outcomes. Studies that embedded mathematical content directly within technology-based interventions generally provide a stronger basis for interpreting potential mathematics benefits than interventions focused primarily on attention or working memory. This distinction is theoretically important because improvements in a cognitive process and improvements in mathematics represent different outcome domains and should not be assumed to be interchangeable. For instance, Rabiner et al. [5] examined computer-based interventions for students with attention difficulties and did not establish consistent mathematics benefits, despite improvements in aspects of attention. Similarly, Chacko et al. [38] found evidence of near-transfer following working-memory training but limited evidence of broader academic transfer. More generally, the meta-analytic findings of Melby-Lervåg et al. [48] indicate that working-memory training does not reliably produce far-transfer effects to broader cognitive or academic outcomes. This convergent pattern across Rabiner et al. [5], Chacko et al. [38], and Melby-Lervåg et al. [48] raises a fundamental question about the theoretical model underlying many technology-based ADHD interventions. If cognitive improvements in attention or working memory do not reliably transfer to academic achievement, interventions may need to either (a) directly embed mathematics content within cognitive training, as attempted by Sperafico et al. [9] and Sánchez-Pérez et al. [40], or (b) reconceptualize the target of intervention from cognitive remediation toward instructional accommodation, using technology to bypass rather than remediate underlying cognitive deficits. Framed this way, the near- versus far-transfer distinction is not merely a methodological caveat but a design choice with direct implications for how future ADHD-focused mathematics interventions should be structured. Cortese et al. [60] reached a similar conclusion in their meta-analysis of cognitive training for ADHD, reporting that while computerized interventions improve performance on trained cognitive tasks, effects on untrained academic skills are small and typically fail to reach statistical significance, reinforcing the argument that domain-embedded instructional support may offer a more direct pathway to mathematics improvement than domain-general cognitive remediation.

This near- versus far-transfer pattern has direct implications for future intervention design: cognitive training alone may not reliably generalize to mathematics achievement, whereas interventions that embed mathematical content within technology-supported scaffolds provide a more direct basis for evaluating domain-specific learning. However, this conceptual distinction should not be interpreted as evidence that one intervention class is superior to another. Rather, it highlights the importance of matching the measured outcome to the intervention target. In contrast, interventions that integrate mathematical learning with technological scaffolding provide more direct evidence of domain-specific learning. The pedagogical-agent intervention of Mohammadhasani et al. [13] reported a substantial mathematics learning effect among students with ADHD, and the RCT by Dai et al. [8] reported significant improvements in attention and academic performance under an active-control design. Computer-based mathematics interventions have also demonstrated potential among learners experiencing emotional, behavioral, or mathematical difficulties [23,39]. Evidence from broader populations further supports the potential of technology-supported mathematics instruction. A meta-analysis by Benavides-Varela et al. [11] found positive effects of digital-based interventions for children with mathematical learning difficulties, while Kim and Xin [22] reported evidence supporting technology-based word-problem interventions for students with disabilities.

At the same time, these findings should not be interpreted as evidence that technology is uniformly effective. The broader mathematics intervention literature emphasizes the importance of explicit, systematic, and well-structured instructional components [23,50]. Technology appears to be most promising when it implements or enhances effective instructional principles rather than functioning as an intervention mechanism in isolation. This is an important distinction because an engaging interface does not necessarily constitute effective mathematics instruction. This variation limits direct comparison across studies and reinforces the need for greater consistency in outcome definitions and measurement approaches. This principle is well-established in the mathematics intervention literature: Gersten et al. [50] and, more recently, Powell et al. [61] demonstrated that explicit, systematic instruction with visual representations and cumulative practice is the strongest predictor of mathematics gains for students with learning difficulties, suggesting that technological enhancements should be evaluated primarily by whether they deliver, scaffold, or intensify these evidence-based instructional components.

The evidence also suggests that engagement-related outcomes may be particularly relevant for students with ADHD. Several interventions reported improvements in attention, persistence, motivation, or classroom behavior alongside academic outcomes [7,8,44]. However, the current evidence does not establish whether improvements in engagement consistently mediate mathematics achievement. Future studies should therefore distinguish between engagement as an outcome, a mechanism of learning, and a potential mediator of achievement.

The methodological characteristics of the evidence further constrain interpretation. Many studies used small samples, single-group or quasi-experimental designs, researcher-developed measures, short intervention periods, or limited follow-up. This concern is consistent with the broader ADHD intervention literature, where the quality and design of school-based trials vary considerably [2]. In addition, differences in participant characteristics, mathematical content, intervention duration, comparison conditions, and outcome measures make direct comparison across studies difficult.

The most defensible interpretation, derived implicitly rather than as a direct result of RQ3, is that technology can enhance mathematics learning and engagement for certain students with ADHD, particularly when math instruction is directly embedded within the intervention. However, current evidence does not yet demonstrate that advanced technological platforms, specifically AI-based systems, are necessarily more effective than established digital approaches.

4.4. Evidence Gaps and Future Research Directions (RQ4)

It should be noted that the 37-source evidence map combines primary empirical studies (Tier 1), mixed primary and review/meta-analysis sources (Tiers 2 and 3), and reviews, frameworks, and technical studies (Tier 4). Findings from Tier 4 are interpreted as contextual evidence regarding technological possibilities and should not be given equal empirical weight as Tier 1 direct-intervention studies. RQ4 identified several gaps in the evidence base, with the most important involving the limited empirical investigation of AI, methodological weaknesses, limited attention to learner heterogeneity, and insufficient evidence concerning long-term implementation and transfer. The most prominent gap is the absence of a sufficiently developed empirical evidence base for AI-driven mathematics interventions specifically targeting students with ADHD. This finding is particularly important because recent literature increasingly describes AI as capable of personalization, adaptive support, learner modelling, and assistive educational functions [15,30]. Reviews focusing specifically on generative AI and neurodivergent students likewise suggest considerable potential but emphasize the early stage of the evidence base [16]. The present review identifies a clear promise–evidence gap. AI capabilities and conceptual proposals are developing rapidly, whereas empirical studies examining their application to mathematics learning among students with ADHD remain limited.

A second gap concerns the need to distinguish AI-based personalization from conventional digital adaptivity. Future research should explicitly report the computational mechanism underlying an intervention rather than using the terms “AI”, “adaptive”, and “personalized” interchangeably. Studies should specify whether systems use rule-based adaptation, machine-learning models, natural-language generation, large language models, learner modelling, or multimodal data. This distinction is necessary to identify the computational approach being evaluated, determine which technological mechanisms may contribute to observed learning effects, and allow subsequent evidence syntheses to distinguish genuinely AI-based interventions from conventional adaptive technologies. Future studies should also consider formal moderator analyses where sample sizes and study designs permit. Variables that emerged descriptively across the current evidence including age or developmental level, intervention duration, ADHD-only versus comorbid profiles, technology type, baseline mathematical achievement, executive functioning, and prior experience with digital technologies could be examined as potential moderators in adequately powered primary studies or future quantitative syntheses. At present, however, the scoping evidence is insufficient to determine whether any of these characteristics systematically modify intervention outcomes.

Third, other studies require stronger methodological designs. Larger randomized controlled trials, active comparison conditions, standardized mathematics measures, delayed post-tests, and multi-site recruitment would strengthen causal inference and improve generalizability. Where feasible, longitudinal studies should examine whether gains are maintained and transferred to novel mathematical problems. Research should also examine moderators such as baseline mathematical achievement, executive functioning, ADHD presentation, co-occurring learning difficulties, and prior experience with digital technologies. The evidence map further suggests that developmental level, intervention duration, comorbidity status, and technology type may warrant examination as potential moderators in future research; however, because no formal moderator analyses were undertaken, the observed differences should be regarded as hypotheses for confirmatory research rather than as established moderating effects.

A fourth gap concerns implementation and educational context. Technology-based interventions cannot be separated from teachers, classroom routines, school resources, and the broader instructional environment. Existing research provides relatively limited evidence concerning teacher training, fidelity of implementation, technical barriers, and the conditions under which digital interventions can be integrated into everyday mathematics instruction. This is particularly important because technology should complement rather than replace teacher expertise. The teacher remains responsible for interpreting errors, identifying misconceptions, providing conceptual explanations, and determining when technological support is insufficient.

Finally, future studies should address equity, privacy, transparency, and algorithmic fairness. AI systems designed for students with ADHD may potentially rely on detailed behavioral and learning data, creating additional ethical considerations for students with disabilities. Emerging evidence also suggests that generative AI assistance during problem-solving may reduce durable knowledge retention even when it improves immediate completion rates, raising additional questions about whether AI-supported practice produces lasting mathematical understanding. Research should therefore report not only whether an AI system improves mathematical outcomes, but also how data are collected, how learner models are constructed, how decisions are explained, and whether performance differs across learner groups. The growing literature on AI in special education underscores the need to consider these issues alongside technological innovation [15,16]. These concerns align with broader critiques of AI in education. Williamson and Eynon [62] pointed out that adaptive learning systems can encode algorithmic biases that disproportionately affect students with disabilities, while Selwyn [63] argued that the rush to deploy AI in educational settings has outpaced empirical evidence and ethical safeguards, a pattern that is particularly evident in the ADHD–mathematics literature mapped by this review.

The four RQs reveal a field that has progressed substantially in digital support for attention, engagement, and mathematics learning, but has not yet established a comparable empirical foundation for AI-specific interventions. The evidence therefore supports continued technological development, but with a shift from demonstrating that technology is engaging or feasible toward testing which technological mechanisms improve mathematical learning, for whom, under what instructional conditions, and over what period of time.

5. Implications

5.1. Implications for Theory

The findings support a theoretical shift from viewing educational technology primarily as a mechanism for remediating ADHD-related cognitive deficits toward understanding technology as a context-sensitive instructional scaffold. Across the mapped literature, the strongest conceptual rationale appears when technology is used to reduce unnecessary cognitive demands, structure attention, provide immediate feedback, support persistence, and facilitate engagement with mathematical content.

This interpretation is compatible with an accommodation-oriented perspective in which the learning environment is modified to enable participation and learning rather than requiring the learner’s cognitive characteristics to be changed before successful participation can occur. However, the present scoping review does not establish a single theoretical model of technology-mediated mathematics learning for ADHD. Rather, it identifies mechanisms that can inform the development and testing of such models. This perspective is supported by disability studies scholarship emphasizing that accommodation does not represent a lesser form of support than remediation, but rather a legitimate mechanism for equitable access and participation for learners whose functional limitations may not be readily modifiable [64,65].

The findings also indicate the need for theoretical models that recognize the dynamic and heterogeneous nature of ADHD-related learning difficulties. Attention, motivation, response regulation, and executive functioning may vary across tasks and overtime. Future conceptual models should therefore move beyond static learner classifications and explicitly consider interactions among learner characteristics, mathematical task demands, technological affordances, and instructional context.

A further theoretical implication concerns transfer. The limited evidence that domain-general cognitive training consistently transfers to mathematics supports greater emphasis on domain-embedded support, in which attentional and executive scaffolds are integrated directly into mathematical activity. Future theoretical work should therefore specify the mechanisms through which technology-supported cognitive scaffolding is expected to influence mathematical performance and distinguish proximal cognitive outcomes from actual mathematical learning.

Finally, theoretical models should explicitly account for learner heterogeneity, including age, mathematical proficiency, comorbid learning difficulties, language background, and educational context. The PCC mapping demonstrates that these characteristics are insufficiently represented in the current literature and may substantially influence the applicability of technology-based interventions.

5.2. Implications for Educational Practice

For educational practice, the findings suggest that technology should primarily be used as support embedded within mathematics instruction, rather than as a replacement for teacher-led mathematical instruction or as isolated cognitive training.

Technology-supported mathematics activities may be particularly useful when they provide: (a) immediate and specific feedback; (b) manageable task segments and clearly structured sequences; (c) persistent representations of mathematical information; (d) opportunities for guided and repeated practice; (e) visual or interactive representations that direct attention to relevant mathematical features; (f) adjustable levels of support and task difficulty; and (g) mechanisms that facilitate self-monitoring and sustained engagement.

These features should be implemented selectively rather than treated as universally beneficial. Excessive animation, gamification, notifications, or multimodal information may increase rather than reduce cognitive demands. Teachers therefore remain essential in determining whether a student’s difficulty reflects a mathematical misconception, attentional fluctuation, excessive cognitive load, or another barrier.

Technology-generated learning data may also contribute to formative assessment. Response accuracy, error patterns, response latency, help-seeking, and task completion can provide additional information about students’ interaction with mathematical tasks. Such information should be interpreted by educators rather than treated as an automatic representation of students’ cognitive states.

At the institutional level, schools should adopt evidence-informed approaches to the procurement and implementation of technology-based interventions. Claims that technology is “AI-powered”, “adaptive”, or “personalized” should not be treated as evidence of effectiveness for students with ADHD. Schools should consider the empirical population, intervention target, mathematical content, comparator condition, implementation requirements, privacy provisions, and available evidence before adopting technology-based interventions.

5.3. Implications for Technology Design

The findings suggest several design priorities for future technology-based mathematics systems for students with ADHD. First, adaptive systems should consider more than answer accuracy when determining instructional support. Response latency, error patterns, help-seeking, and interaction behavior may provide useful information about students’ learning processes, although any use of such data should be empirically validated and ethically governed. Second, interfaces should externalize unnecessary working-memory demands. Persistent problem representations, intermediate steps, visual models, worked examples, and easily accessible mathematical information may reduce the need to hold multiple pieces of information in memory while simultaneously performing mathematical operations. Third, engagement mechanisms should support meaningful participation and mathematical mastery rather than relying exclusively on extrinsic rewards. Gamification should be integrated with instructional objectives and evaluated for sustained rather than short-term engagement. Fourth, interfaces should minimize unnecessary visual and interactional complexity. Multimodal features should be used strategically to signal relevant information rather than simply increasing the amount of information presented. Finally, AI-enabled systems should incorporate privacy, transparency, fairness, and accountability by design. Systems intended for students with ADHD should clearly communicate what data are collected, why they are collected, how they are used, and how decisions or recommendations are generated. Given the current lack of empirical validation, AI systems should also be subject to continuous evaluation rather than assumed to be effective because they provide sophisticated personalization.

6. Conclusions

This scoping review mapped the available evidence concerning technology-based approaches to mathematics education for students with ADHD across the PCC dimensions. Thirty-seven peer-reviewed studies (n = 37) published between 2010 and 2026 were identified and organized into four analytical tiers according to their degree of relevance to the direct intersection of technology, ADHD, and mathematics. Four principal conclusions emerge. First, the technology landscape is heterogeneous and has evolved from relatively established computer-based training approaches toward increasingly sophisticated adaptive, intelligent, and AI-related applications. Second, the direct evidence base remains substantially smaller than the broader literatures on technology and ADHD, technology and mathematics, or AI in special education. Third, the mapped evidence indicates that technology-based interventions embedding mathematical content within attentional and executive-function scaffolds were represented across the evidence base alongside domain-general cognitive-training approaches; however, substantial methodological heterogeneity precludes any comparative conclusion between these intervention classes. Fourth, and most importantly, the review identified a pronounced evidence-to-promise gap in AI-supported mathematics education for students with ADHD. Although the AI literature increasingly proposes intelligent tutoring, adaptive personalization, generative AI, and data-driven learner modeling, empirical evidence directly evaluating these approaches for mathematics learning among students with ADHD remains extremely limited within the evidence mapped by this review. Accordingly, the findings should not be interpreted as evidence that AI-based mathematics instruction is already effective for students with ADHD. Rather, they indicate that technological development and theoretical proposals are advancing faster than population-specific empirical validation.

Overall, the evidence supports conceptualizing technology as a context-sensitive instructional support capable of accommodating attentional and executive-function variability during mathematical learning, while recognizing that the effectiveness of emerging AI applications remains an empirical question. The main contribution of the present scoping review is therefore not to establish the superiority of any technology or intervention design, but to clarify the current evidence landscape, identify the limited direct evidence at the ADHD-mathematics-technology intersection, and establish priorities for the next generation of research.

7. Limitations and Future Directions

This scoping review has several limitations that should be considered when interpreting its findings. First, although a comprehensive multi-database search was conducted across Web of Science, Scopus, ERIC, PsycINFO, IEEE Xplore, ACM Digital Library, and PubMed/MEDLINE, the grey-literature search was more restricted, combining ProQuest Dissertations & Theses, OpenGrey, and the first 200 Google Scholar results ranked by relevance, supplemented by backward and forward citation chasing and hand-searching (Appendix B.4). Consequently, some relevant unpublished studies, dissertations, conference proceedings, and other grey-literature sources may have been missed, and the search strategy may therefore have influenced the composition of the mapped evidence base. Future reviews should strengthen grey-literature retrieval through more systematic and targeted searches of relevant repositories and conference proceedings, including comprehensive database exports where feasible. Citation-network analysis and librarian-assisted validation could further improve search sensitivity and reduce the risk of missed evidence.

Second, this scoping review deliberately incorporated evidence with different degrees of proximity to the central ADHD-technology-mathematics intersection. The four-tier framework enabled the review to map a narrow direct evidence base alongside adjacent literature, but it also introduced substantial heterogeneity. Tier 1 sources provide the most direct evidence concerning the review concept and population, whereas Tiers 2–4 provide contextual or adjacent evidence. In particular, findings from Tier 3, where ADHD may occur as a comorbid condition within broader mathematics-learning or special-education populations, should not be interpreted as equivalent to evidence obtained exclusively from students with ADHD. Future research should prioritize generating Tier 1 evidence by designing studies that recruit ADHD-specific samples and report mathematics outcomes as primary endpoints rather than incidental measures. Future reviews, in turn, might adopt stricter tier-specific inclusion criteria or conduct separate syntheses for each tier to reduce interpretive ambiguity. Only 20% of studies were dual-charted for the κ = 0.88 reliability check, whereas the remaining 80% were charted by a single reviewer. This increases the risk of undetected extraction errors and represents a limitation that should be addressed in future reviews by full dual-charting or independent verification.

Third, the included sources varied considerably in methodological design and included randomized and quasi-experimental studies, single-subject research, reviews and meta-analyses, technical studies, and conceptual or framework papers. Therefore, the reported effect sizes and outcome estimates should not be interpreted as directly comparable measures of intervention effectiveness. The purpose of the review was to map the nature and distribution of the evidence rather than to calculate a pooled estimate of effect. Future primary studies should adopt standardized mathematics outcome measures (e.g., curriculum-based measurement, standardized achievement tests) and report effect sizes using consistent metrics (e.g., Cohen’s d, Hedges’ g, or percentage of non-overlapping data for single-subject designs) to enable cross-study comparison. Once a larger homogeneous subset of experimental studies accumulates, future systematic reviews should conduct design-stratified meta-analyses rather than scoping mappings.

Fourth, no formal risk-of-bias assessment or meta-analysis was undertaken. This is consistent with the descriptive and evidence-mapping purpose of the review; however, it means that the study does not provide a formal comparative assessment of methodological quality across individual sources. The absence of such an assessment is particularly important when interpreting findings from small-sample, quasi-experimental, or single-subject studies and when considering apparent differences in reported effect sizes. Future systematic reviews should therefore incorporate validated risk-of-bias instruments (e.g., Cochrane Risk-of-Bias Tool 2.0 for randomized trials, ROBINS-I for non-randomized studies, or What Works Clearinghouse standards for single-case designs) and require primary authors to adhere to reporting guidelines such as CONSORT, PRISMA, or SCRS (Single-Case Reporting Guideline In Behavioral Interventions) to facilitate transparent quality appraisal.

Fifth, substantial heterogeneity was observed in participant characteristics, mathematical content, technology type, intervention duration, comparator conditions, outcome measures, and educational contexts. This heterogeneity limited quantitative synthesis and reduced the generalizability of findings across populations and educational settings. It also reinforces the importance of interpreting the four tiers separately rather than treating the 37 studies as a homogeneous evidence base. Future research should therefore employ standardized taxonomies to classify technology types (e.g., intelligent tutoring systems, gamified platforms, adaptive algorithms) and mathematical domains (e.g., number sense, procedural fluency, word-problem solving), use common outcome batteries across studies, and conduct multi-site replication trials with active comparator conditions. Particular attention should be given to secondary-school populations and students with different comorbidity profiles, as these subgroups remain underrepresented in the current evidence base. Longitudinal follow-up and explicit examination of transfer and implementation fidelity are also needed to determine the durability and scalability of mathematics-technology interventions for students with ADHD.

Sixth, the absence of formal risk-of-bias assessment means that the review cannot determine the certainty of the reported effects or rank interventions by comparative effectiveness. All effect sizes and statistical findings are presented descriptively as characteristics of the mapped literature, not as evidence that one intervention type is superior to another. Readers should therefore treat the reported ranges and patterns as hypotheses for future confirmatory research rather than as established comparative effects.