AIEd Knowledge BaseAIEd KB
On this page

Research Article

Generative AI, virtual reality, and beyond: A scoping review of digital assistive technologies for neurodivergent students in higher education

created 2026-09-17 · updated 2026-09-17

Synthesis: A PRISMA-ScR scoping review searched five databases and a decade of publication (2015–2025) to identify 766 records and include 40 empirical studies of digital assistive technologies for neurodivergent students in higher education. The field it maps is small but has reorganised itself around Generative AI (15 of 40 studies) with virtual reality as a second strand (11 studies, 10 of them VR), and its tools cluster by purpose — supporting learning (n = 27), improving the selection of assistive technology (n = 10), and educating neurotypical peers against bias (n = 3). The authors' central finding is a design critique: the literature is organised around individual accommodation and neurotype-specific tools gated by formal diagnosis, when the barriers the tools actually address (reading and writing, study management, attention, social communication) cut across neurotypes. They argue for universal design and participatory development, and flag the equity cost of an evidence base concentrated in the Global North whose most immersive tools are also the least scalable.

Key Findings

  1. Searches of Scopus, Web of Science, ERIC, PubMed and PsycInfo in October 2025, plus forward and backward snowballing in November 2025, returned 766 studies; 239 automatic and 29 manual duplicates were removed, 498 records were screened at title and abstract, 113 went to full text, and 40 met the inclusion criteria.
  2. Screening was reliable by the authors' own measure: intercoder agreement was 88.25% at the title-and-abstract stage and 89.53% at full text, with every discrepancy resolved by discussion among the three reviewers.
  3. AI-based tools featured in 15 of the 40 studies and the review reports a sharp increase in AI-based work from 2020 onward, crediting generative AI with accelerating the field toward scalable, adaptable assistive tools; AI-based studies amount to 37.5% of the corpus, against 24 non-AI studies and one unspecified.
  4. Technologies clustered into three purposes: tools directly supporting neurodivergent students' learning (n = 27), tools improving the selection of assistive technology (n = 10), and tools addressing bias by educating neurotypical peers (n = 3) — the third category being the only one that treats the environment rather than the student as the object of intervention.
  5. Grouped by barrier rather than by diagnosis, reading and writing dominated (n = 13) and study management followed (n = 12), while attention and social communication were the two smallest domains and only 6 studies addressed several barriers at once.
  6. Most studies made a formal diagnosis a requirement for participants — 28 of 40 required one, 11 recruited students who self-identified as neurodivergent, and one did not specify — which the authors read as excluding students who lack institutional recognition of their functional differences.
  7. Reported effects were generally favourable and no study reported an overall harmful effect, but the review documents unintended burdens: cognitive overload, fatigue, distraction, VR-related discomfort, usability and technical failures, inaccurate prompts, and — for generative AI specifically — accuracy, Privacy, over-reliance and loss of authentic writing voice.
  8. Evidence and method are both thin: studies were concentrated in the Global North (United States 35%, Italy 12.5%, United Kingdom 10%), mostly undergraduate, and lacked (quasi-)experimental comparison groups or longitudinal follow-up, with many conducted by the tools' own developers.
  9. The review's own bounds are stated: a ten-year window risks underrepresenting older technologies, the peer-reviewed and English-language filters may miss very recent and non-Anglophone work, and acquired brain differences, mental health conditions, genetic syndromes, epilepsy and hearing impairments were excluded by design.

What the Review Screened and How It Ordered the Field

The review's method is deliberately transparent about its screening arithmetic, and the arithmetic matters here because the field is small enough that counting studies is itself a substantive finding. Five databases plus citation searching produced 766 records in October–November 2025; after deduplication 498 titles and abstracts were screened and 113 full texts assessed, leaving 40 included studies published between 2016 and 2025. One author screened 100% of results and the other two screened roughly 50% each, with discrepancies resolved collectively at both stages.

Two ordering decisions give the review its analytic edge. The first is to categorise tools by purpose (support, selection, bias reduction) and then by the academic barrier they target — reading and writing, study management, attention, social communication, and multi-barrier access ecosystems — rather than by diagnostic label, on the explicit ground that students with the same diagnosis experience different challenges and students with different diagnoses experience the same ones. The second is to treat VR/XR as a cross-cutting format rather than a third technology class, so the 11 immersive studies are counted inside the AI and non-AI groups rather than alongside them. Both decisions are defences against the diagnostic essentialism the authors see in the literature they reviewed.

Where the Technology Actually Works — and What It Costs

The positive findings are concrete but narrow. An accelerated-reading application (Schneps et al.) combined visual augmentation with text-to-speech and brought dyslexic students' reading speed to the level of readers without dyslexia. Machine Learning classifiers in the VRAIlexia and BESPECIAL platforms predicted suitable support tools and learning strategies with 90% to 94% accuracy — though two of those platforms were never tested with students. A custom VR study environment with automated feedback raised concentration, motivation and effort for university students with ADHD (Cuber et al.); a smartphone/smartwatch prompting system increased independent appointment attendance and task completion; a self-monitoring app produced high on-task behaviour; and GenAI was used to convert textbook chapters into audio modules tuned for ADHD, improving academic performance among engaged students.

The costs are equally concrete: VR appeared less beneficial for single-task comprehension, one immersive attention system produced boredom and comprehension difficulty, another was limited by motor noise and had no effect on cognitive load, and self-monitoring tools were accompanied by fatigue and dependence on external reinforcement. Generative AI carried its own list — inaccurate answers, Academic Integrity concerns, fear of over-reliance, and the loss of human contact. The paper places all of this under the umbrella of what Bauer et al. call inversion effects, the case in which adding technology leaves learning worse than before, whether through over-reliance on chatbots or through the cognitive overload of immersive environments.

Generative AI as the Field's Reorganising Force

The clearest structural change the review records is the shift toward AI. Earlier work in the window was predominantly non-AI; from 2020 the number of AI-based studies rose sharply, and generative AI is credited with pushing the field toward tools that are flexible enough to serve neurodivergent students without being tied to a single profile. General-purpose GenAI tools — ChatGPT, Grammarly, Quillbot, DeepL — appear in the multi-barrier category precisely because they can summarise dense text, clarify instructions, plan study tasks, rephrase material and initiate writing across contexts. GenAI also shows up as an additive layer inside non-AI tools: it was used to generate audio from written chapters and to support creative thinking within collaborative mind-mapping tools already designed to lower the social threshold for neurodivergent students.

That flexibility is the review's reason for optimism about scale, and its reason for caution about evidence. A tool that adapts to the student rather than to the diagnosis is more plausibly universal, and AI-based tools are cheap relative to headsets — but the corpus offers few controlled demonstrations that the tools improve outcomes rather than perceptions, and several studies report the technology's effects through single-trial ratings rather than measured learning.

Accommodation or Universal Design

The review's argumentative core is a continuum rather than a taxonomy: assistive technology can adapt students to environments, or environments to students. Most of the 40 studies sit at the accommodation end — designed for neurodivergent learners, leaving other students' experience unchanged — and the authors note that the pattern of designing for assumed neurotype profiles risks reproducing the limitations of the diagnostic categories themselves. At the universal end sit tools intended for the whole classroom, such as online presentation platforms and e-portfolios. The example the review returns to is McDowell's e-portfolio Group Work study, in which an autistic student took the lead in collaboration in a way the face-to-face setting had not permitted — a technology that made collaboration possible for everyone, and thereby transformed the environment rather than patching the student.

This is why the three bias-reduction studies matter out of proportion to their number. By aiming at the understanding and behaviour of neurotypical peers, they target ableism at its source instead of asking neurodivergent students to adapt, and the authors present them as an underexplored complement to individually oriented tools rather than a substitute for them. The review's recommendation is not to abolish individual support — some students will always need tailored provision — but to stop treating universal design as the exception in the literature, and to require co-design with neurodivergent students and academics in future tool development.

What the Evidence Cannot Yet Support

The gaps the review names are of four kinds. Geographically, the evidence base is North American and European, and the authors point out that the concept of Neurodiversity itself was developed largely by white scholars in the Global North, so tools designed within that frame risk importing an implicitly assumed student; the recommendation is to include Global South voices, institutions and epistemic traditions as partners rather than sites. By level, undergraduate students dominate, leaving postgraduate and doctoral students — whose challenges involve higher autonomy, less structure and an often isolating culture — largely unaddressed. Methodologically, comparison groups are frequently absent, samples are small and heterogeneous, and developer-run product assessments substitute for independent evaluation, which is why the added value of the technology is hard to establish even when effects look favourable. Technically, the most immersive options are the least scalable, and the review treats cost and specialist hardware as an access question rather than a design afterthought.

For institutions, the practical reading is a procurement and design position: the current trajectory of assistive tools will keep producing narrow, diagnosis-gated and resource-intensive products unless universal design and participatory development are stated as requirements. For researchers, the gap list doubles as an agenda — experimental and longitudinal designs, cross-neurotype functional targets, attention and social communication, postgraduate populations, and scalable immersive tools.

Connected Concepts

  • Neurodiversity — the paper's organising concept, and the one whose diagnostic framing it critiques
  • Assistive Technology — the object of the review and its accommodation-versus-universal-design spectrum
  • Universal Design for Learning — the review's central recommendation for tool development
  • Accessibility — framing of cost, specialist hardware and the digital ecosystem as access questions
  • Inclusive Learning — the broader design agenda the review places itself inside
  • Higher Education — the review's scope, and the setting where contact hours fall and independent study rises
  • Generative AI — present in 15 of 40 studies and credited with reorganising the field
  • Virtual and Augmented Reality — the second dominant strand (11 studies) and the least scalable
  • Equity — Global North concentration, exclusion of students without formal diagnosis
  • Special Education — the diagnosis-gated model the review argues against
  • Digital Divide — scalability and cost of immersive assistive tools
  • Cognitive Offloading — one pathway of the "inversion effects" the review warns about
  • Well-Being — sensory demand, fatigue, anxiety and stigma in the reviewed studies
  • Limitations in AIEd Research — missing comparison groups, tiny samples, developer-run evaluations
  • Stakeholders — the review's audience: students, educators, institutions, developers, policymakers

Connected Articles

Citation

Rempel, C., Heimann, K., & Prilop, C. N. (2026). Generative AI, virtual reality, and beyond: A scoping review of digital assistive technologies for neurodivergent students in higher education. EdArXiv Preprints.

Metadata

Study designMeta-Analysis and Systematic Review, Qualitative Research
LevelHigher Education
Intended audienceInstitutions, Researchers, Instructional Designers
Technologies and techniquesGenerative AI, Virtual and Augmented Reality
Equity, ethics, and responsible useAccessibility, Assistive Technology, Equity, Inclusive Learning, Neurodiversity, Universal Design for Learning
Foundations of AI in educationLimitations in AIEd Research
Page kindSynthesis
SourceHistoryEditChat

Embed this page

Copy the code below to embed a chromeless version of this page in a learning management system or other website. The embedded view hides the site header, navigation, and footer.