How Games Help Autistic People: The Evidence, the Design Principles, and How to Build One

Organisations commissioning a game or application for autistic users — special education programmes, therapy clinics, research groups, nonprofits, health systems, and founders building for this population — face the same first question: what does the evidence support, and what does it not?

The evidence is domain-specific. Game-based interventions show measurable, replicated effects in some skill domains and no demonstrated effect in others. That distinction determines scope, claims, instrumentation and, ultimately, whether a product can be sold into a school, clinic or health system. This guide starts with the published numbers and works forward to design, compliance, buyers, funding and cost.

Key Definitions

Autism (ASD)

A lifelong neurodevelopmental difference characterised by differences in social communication and interaction, restricted or repetitive interests and behaviours, and sensory processing. It is a spectrum: support needs, language level, co-occurring conditions and sensory profiles vary widely between individuals. "ASD" (autism spectrum disorder) is the DSM-5 and ICD-11 clinical term.

Game for autistic people

An interactive application designed for autistic users that supports a defined skill, communication need or wellbeing goal. The defining features are not the content but the design: predictability, user control over sensory input and pace, low-stakes failure, structured repetition and progress data for a supporting adult.

Serious game

A game whose primary purpose is beyond entertainment — skill development, education, assessment or therapy — while retaining real game mechanics: challenge, progression, feedback and player agency. The evidence in this guide is largely about serious games.

Gamified app

A non-game task (a flashcard drill, a worksheet, a checklist) with points, stars or badges attached. Gamification can improve motivation for a task but does not by itself reproduce the mechanisms — agency, repetition inside a motivating structure, low-stakes failure — that the meta-analyses describe.

Game-based intervention

The research term for a structured programme in which a game is the delivery mechanism for a defined skill target, usually alongside a practitioner, teacher or parent, with outcomes measured by validated instruments rather than in-game score.

Digital therapeutic (DTx)

Software that delivers a clinical intervention and makes a medical claim (treatment or mitigation of a condition). In the US this is Software as a Medical Device requiring FDA authorisation. EndeavorRx (ADHD attention function, De Novo 2020) is the reference example; no game-based DTx for autism had received FDA marketing authorisation at the time of writing, with FloreoRx under review.

AAC app

An Augmentative and Alternative Communication application: a tool that lets someone communicate without speech, or in addition to it, using symbols, text and speech output. AAC apps are communication devices used in real situations, not practice games, and reliability outranks every other design goal.

Game-based assessment

Software that uses game-like tasks, eye-tracking or observed play to screen for, aid diagnosis of, or measure autism-related characteristics. Any diagnostic or screening claim makes it a regulated medical device (Canvas Dx and EarliPoint are FDA-cleared examples); progress measurement inside a non-claiming product is a process measure, not a diagnosis.

What the Evidence Shows

Where the Evidence Is Strong, Emerging, and Absent

• Strong and replicated: emotion recognition and emotional awareness, motor skill development, cognitive and attention measures. Multiple independent reviews reach consistent conclusions with moderate-to-large effect sizes.
• Solid but heterogeneous: social skills and social behaviours. Positive across reviews, including the 2026 AI-driven games meta-analysis, but studies vary widely in design, duration, outcome measure and participant profile, which the review authors themselves flag.
• Emerging: VR and immersive interventions. An RCT of interactive VR-motion serious games reported benefits for social communication and self-living ability, and a VR platform completed a pivotal trial in 2026. However, a 2026 network meta-analysis found the VR evidence base too heterogeneous to rank one VR approach against another. Immersion appears to matter; which form of immersion matters most is not yet established.
• Emerging: executive function and attention training. Positive pooled results, but pilot studies dominate and larger trials are needed.
• Emerging: adults. A small RCT and follow-up link VR interview training to interview performance and employment outcomes; most other adult evidence is observational or community-reported.
• Absent: reduction of behavioural problems. Pooled SMD of −0.14 is not significant. This is the clearest negative finding in the literature and it should shape every claim a product makes.
• Unresolved: generalisation and durability. Multiple reviews note that improvements observed post-intervention were not always sustained at follow-up, and that transfer from the game to real-world settings is the field's central open question.

What Games Cannot Do

This section matters commercially as much as ethically. Clinical and education buyers evaluate vendors partly on whether their claims match their evidence; families making decisions about a child are entitled to accuracy; and a claim that crosses into diagnosis, treatment or mitigation can move a product into regulated medical-device territory.

• Games do not cure or treat autism. Autism is a neurodevelopmental difference, not an illness. The framing in current research and in autistic self-advocacy is support, skill-building and accommodation — not remediation toward a neurotypical norm.
• Games do not reduce behavioural problems. The pooled effect is SMD −0.14 and not statistically significant. A vendor claiming otherwise should be asked for the trial.
• Screen gains do not automatically transfer to real life. Generalisation is the field's central unresolved limitation. A product that improves in-game emotion recognition has not demonstrated improved playground interaction unless it measured that specifically.
• Gains are not always durable. Reviews note improvements observed post-intervention that were not sustained at follow-up. Maintenance and booster design is an open problem, not a solved one.
• Games do not replace therapy or teaching. They are adjuncts. The strongest deployments sit inside an existing support plan and give the practitioner data and practice repetitions they could not otherwise get.
• Games do not diagnose. The narrow exception is FDA-cleared diagnostic aids such as Canvas Dx (ages 18–72 months) and EarliPoint (ages 16–95 months as of March 2026), indicated for use by qualified clinicians. Those are regulated medical devices, not apps.
• Technology does not fix an inaccessible environment. If a classroom is sensorily overwhelming, the intervention that helps most is changing the classroom. Software that implies otherwise is what critics call technological solutionism, and practitioners recognise it immediately.

Why Games Work: The Mechanism

Understanding the mechanism matters because it tells you what to build. If predictability is doing the work, then a randomised or surprising interface actively destroys the benefit. If reduced social load is doing the work, then forcing multiplayer interaction as a requirement removes the reason the tool helps in the first place. Design decisions that seem trivial in a mainstream game are load-bearing here.

How Games Help Across Ages and Profiles

Co-occurring Conditions Change the Design Brief

Autism rarely presents alone, and a product designed for "autism" without considering the most common co-occurring conditions will fail a large share of its intended users. Prevalence figures below are drawn from CDC surveillance, Medicaid claims analysis and clinical reviews; ranges are wide because samples and methods differ.

Prevalence, the Access Gap, and Why Demand Is Growing

Three features of this data matter for anyone designing a product. First, identification varies enormously by community — from 9.7 per 1,000 children in one Texas site to 53.1 per 1,000 in California — and the report authors are explicit that this reflects differences in availability of services, early detection and diagnostic practice, not differences in underlying risk. Second, prevalence is now higher among Asian and Pacific Islander (38.2 per 1,000) and Black children (36.6 per 1,000) than among white children (27.7 per 1,000), reversing a long-standing pattern of under-diagnosis in those communities. Third, the service system is capacity-constrained on every axis: assessment waitlists, therapist supervision, and public budgets.

All three point the same way. The fastest-growing user populations are in communities that were previously underserved, they are not well served by products designed around a single cultural context, a single language or an assumption of well-resourced home support, and the families and services reaching them need tools that extend practitioner capacity rather than add to it. Multilingual capability, low-bandwidth and low-cost deployment, and home-practice modes that a practitioner can assign and monitor are not optional extras in this category. They determine whether the growing population can reach the product at all.

The Eight Skill Domains Games Can Support

Types of Games and Products in This Category

The Fourteen Design Principles

Sensory Design: The Rules That Decide Whether It Gets Used At All

Accessibility Standards: WCAG, Game Accessibility Guidelines and Platform Requirements

Input diversity deserves specific attention. Beyond touch and keyboard, autistic users and users with co-occurring motor differences may rely on switch access, eye-gaze systems such as Tobii-based trackers, head pointers or large-format adapted controllers. A game that only responds to fast, precise gestures excludes those users entirely; a game whose core mechanic can be completed with one or two switch inputs and adjustable timing includes them without a separate build.

Co-Design: Building With Autistic People, Not For Them

The research is unusually clear here. A published participatory framework for serious games in autism describes involving users and relevant stakeholders as "design partners" through structured participatory sessions and a design thinking process, and emphasises providing autistic children with practical, meaningful opportunities for interaction and promoting their voices. Reviews of inclusive co-design with autistic children recommend positioning children and parents as partners who calibrate the system to their own sensory preferences.

• Recruit early, at concept. Autistic advisors should shape what you build, not react to what you built. Involving people after the architecture is fixed produces cosmetic changes only.
• Include autistic adults, not only children and parents. Autistic adults can articulate experiences a child cannot yet name, and adult mentors have been used effectively in participatory autism research. For adult-facing products, autistic adults are the primary partners.
• Pay your advisors. Lived-expertise consultation is professional consultation. Unpaid participation skews the sample toward those who can afford to volunteer.
• Offer multiple communication modes. Verbal, written, AAC, drawing, photo-based methods such as photovoice, and pointing at options. Requiring speech excludes exactly the users who most need the product.
• Run short, structured, predictable sessions. Publish the agenda in advance, keep sessions brief, provide a quiet space and an exit, and never make participation contingent on staying.
• Involve practitioners and teachers throughout. The literature notes teacher involvement across the whole process is crucial, because they know how the product must fit into a real session.
• Recruit for diversity of profile. Include minimally speaking users, users with intellectual disability, girls and women, and users from the communities where identification is growing fastest. A panel of verbally fluent boys describes one part of the population.
• Test in the real environment. A product that works in a quiet one-to-one room may fail in a busy classroom. Test where it will actually be used.
• Report back what changed. Tell advisors which of their input was acted on and which was not, and why. This is what distinguishes partnership from consultation theatre.

Fitting Games into Therapy Plans, IEPs and Classrooms

Individualised Education Programs and Goal Mapping

In US schools, and under equivalent frameworks elsewhere (EHC plans in England, IPRC processes in Ontario), an autistic student's supports are organised around measurable goals with a baseline, a target, a method of measurement and a review cadence. A product that reports "Level 4 unlocked" is unusable for that purpose; a product that reports "identified target emotion from a photograph in 8 of 10 opportunities across three sessions, up from 4 of 10 at baseline" can be pasted into a progress report. The practical requirements are: goals defined per user by the adult, in their own words; each activity mapped to one or more goals; accuracy, prompt level and independence recorded per opportunity; date-stamped trend views; and exportable summaries in the formats schools use. Alignment with curriculum standards matters for academic products; alignment with goal language matters for every product.

Therapy Models and What Each Needs From a Game

The Screen-Time Question

Every school leader, clinician and parent will ask it, so the product should answer it before they do. Three points are defensible. First, autistic children already spend more time on screens than their peers, so the realistic comparison is not "screen versus no screen" but "purpose-built, bounded practice versus unstructured entertainment." Second, paediatric guidance consistently distinguishes content quality and co-engagement from raw hours; a 15-minute session completed with a parent or therapist toward a defined goal is a different activity from two hours of unsupervised play. Third, the product must make bounded use easy: default session lengths, hard stops with a designed ending, adult-set limits, no notifications, no engagement-maximising mechanics, and visible usage data so the adult can see exactly how much time was spent. A product that cannot demonstrate its own restraint will lose the argument regardless of its evidence.

AI, Adaptive Difficulty and Sensing Technology

Game-Based Assessment and Screening: A Separate Product Class

The commercial logic for assessment products is strong: assessment waitlists are the single largest bottleneck in the service system, and EarliPoint's 2026 expansion was explicitly justified by outcome measurement over time and by the fact that four in five autistic children are diagnosed after toddlerhood. The technical and regulatory reality is equally clear. An assessment product needs design controls, clinical validation against gold-standard diagnostic assessment, defined sensitivity and specificity, a clinician-in-the-loop indication (both cleared products are for use by qualified clinicians), and a quality management system. That is a different company shape from a skills-practice game, with development costs and timelines an order of magnitude larger. Organisations that want screening capability should decide that at the outset and plan the SaMD path from day one, rather than adding a "screening score" to a practice game later, which creates an unapproved medical claim.

There is a legitimate middle ground. A practice game can measure change on the skill it targets — accuracy, latency, prompt level, independence — and present that to a practitioner as progress data for their own clinical judgement. It should not compute or display anything that looks like a diagnostic or risk score, should not use words such as "screen," "detect" or "identify autism," and should administer validated outcome measures alongside the game rather than deriving outcomes from gameplay.

Formats and Modalities

Technology Choices

• Unity — the default for most projects in this category. Cross-platform from one codebase, mature accessibility plugins, good sensor, camera and VR integration, and strong support for the adjustable-settings architecture these products require.
• Unreal Engine — where visual fidelity or immersive VR is central, particularly for daily-living simulation, vocational rehearsal and environment practice. Higher production cost and heavier device requirements.
• HTML5 and WebGL — frequently the correct choice for institutional deployment. No installation, no app-store gatekeeping, runs on locked-down school devices, and updates instantly. For research studies it is often the only practical option.
• Native iOS and Android — essential for AAC and communication tools, which must work offline, launch instantly and integrate with platform accessibility services such as Switch Control, VoiceOver and TalkBack.
• Roblox — where the goal is reaching autistic players inside a community they already use heavily, with the trade-off that you inherit the platform's rules, moderation and monetisation model.
• Gamification layers — where the product is fundamentally an educational or clinical tool that needs motivating structure rather than a full game. Often the right answer for practitioner-facing platforms.
• Sensing and AI components — on-device inference wherever possible (Core ML, TensorFlow Lite, MediaPipe), so biometric and children's data does not leave the device; model versions stamped into every record; camera and microphone features optional, off by default, and separately consented.
• Data layer — an event-level data model designed before the first sprint, with consent state and build version on every record, regional data residency options, and export in formats a researcher or school can use without engineering help.

Compliance and Regulation

Who Buys These Products, and How They Buy

Funding, Reimbursement and Business Models

Where Development Funding Comes From

Most organisations commissioning in this category are not self-funding the build. Common sources, each with its own eligibility, cycle and evidence expectations: public research funders (in the US, the National Institutes of Health including NICHD and NIMH, the National Science Foundation, and the Department of Education's research and special-education programmes; in the UK, NIHR, UKRI and Innovate UK; in the EU, Horizon Europe); small-business innovation schemes such as US SBIR and STTR, which fund early product development with a research partner; autism-focused foundations and research funders, including the Organization for Autism Research, Autism Speaks and the Simons Foundation's SFARI programme; disability and special-education foundations; corporate social responsibility budgets, which in India are a mandated funding stream under the Companies Act; and philanthropic or family-office funding attached to a specific school or clinic. Eligibility and calls change every year, and this guide does not track them; the durable point is that a funder-ready proposal needs a defined domain, a named evidence tier, a co-design plan, a compliance map and an itemised budget, which is exactly the scoping output described in the process below.

How to Develop a Game for Autistic Users: Step by Step

1. Define one target domain and one outcome

   Choose a single skill domain supported by evidence — emotion recognition, social skills, executive function, motor skills, communication, daily living or academic learning — and define the specific, measurable outcome you expect for a defined age group and setting. Products targeting everything demonstrate nothing.

   OutputA one-page brief naming the domain, age band, setting, primary outcome and the adult who will use the data.

2. Map the evidence and set the claims

   Review what the meta-analyses and trials support for that domain and age, decide what the product will and will not claim, and check whether any intended claim crosses into diagnosis, treatment or mitigation. That decision determines whether the project is an educational tool or a regulated medical device.

   OutputA claims register and a regulatory determination, agreed before design begins.

3. Recruit paid autistic co-design partners

   Bring autistic people, families and practitioners in as design partners from concept. Compensate them, offer multiple communication modes, run short predictable sessions, and recruit for diversity of profile including minimally speaking users, users with intellectual disability, girls and women, and adults for adult-facing products.

   OutputA co-design panel with a session plan, compensation budget and feedback-reporting commitment.

4. Design the sensory and accessibility architecture, then prototype

   Specify every user-controllable sensory channel, conservative defaults, the pre-play calibration flow, WCAG 2.2 AA conformance, switch and screen-reader support, and input alternatives. Build a low-fidelity prototype of one activity and test it with the co-design panel in the real environment before committing to full production.

   OutputA tested prototype and a sensory and accessibility specification that the build must satisfy.

5. Design the data model and instrumentation before the first sprint

   Choose the evidence tier, select validated outcome measures with a research or clinical partner, and define session, task, settings-state, difficulty-trajectory, generalisation, consent and version data. Instrumentation added later invalidates everything collected before it.

   OutputA data dictionary, an evidence plan and, if a study is intended, ethics and consent documentation.

6. Build against the fourteen principles, with the practitioner view in version one

   Implement predictability, sensory control, removable time pressure, literal language, uncluttered layout, explicit progress, low-stakes failure, personalisation, special-interest theming, structured repetition, transition warnings, optional social features and an always-available exit. Build the adult dashboard, goal mapping and reporting alongside the user-facing game, not after it.

   OutputA release candidate with user-facing activities, adult-facing reporting and compliance controls in place.

7. Pilot in the real environment against pre-agreed success criteria

   Run a six-to-eight-week pilot with eight to twenty users in the actual classroom, clinic or home setting. Measure adult setup time, session completion, sensory-adverse events, practitioner fit rating, a pre- and post-measure on the target skill, and a generalisation log. Change the product based on what the pilot shows.

   OutputA pilot report with go, fix or stop decisions and the first usable evidence for buyers and funders.

8. Launch with compliance documentation and begin evidence generation

   Complete the accessibility conformance report, privacy and consent documentation, data agreements for schools or clinics, and app-store or web deployment. Start the study at the chosen evidence tier, with build and model versions pinned for its duration, and plan follow-up measurement to test durability and generalisation.

   OutputA shipped product with a VPAT or conformance report, a privacy pack, and a running evidence programme.

Proving It Works: Building Your Own Evidence

What to Instrument From Release One

• Session-level: start and end timestamps, duration, completion versus abandonment, and where in the flow abandonment occurred.
• Task-level: per-item accuracy, response latency, retries, prompt level required, and error patterns rather than just error counts.
• Settings state: which sensory and accessibility settings each user has active. This is research data in its own right and one of the most valuable datasets in this category, because almost nobody collects it.
• Difficulty trajectory: how the adaptive system moved, so you can distinguish skill growth from the algorithm simply making things easier.
• Validated outcome measures administered alongside the product, not derived from it. In-game score is not an outcome measure; it is a process measure.
• Generalisation and follow-up: a structured way for a parent, teacher or therapist to record whether the skill appeared outside the game, and a measurement point weeks after the intervention ends.
• Build and model version stamped on every record, so a mid-study update or a model change does not silently contaminate the dataset.
• Consent state and data-sharing scope attached to every record, so deletion requests can be honoured and research-consented data separated from operational data.

Validated Outcome Measures by Domain

Researchers, ethics committees and clinical buyers expect outcomes on instruments they recognise. The list below is illustrative rather than exhaustive; most are licensed, require purchase, and must be administered or interpreted by qualified professionals. Choosing the measure is a clinical or academic decision made with the study partner, but the product must be designed to schedule, record and export it.

Cost and Timeline

Outside the build budget: paid autistic advisor compensation, clinical or educational advisory board, independent accessibility audit and VPAT production, evidence-generation studies and ethics approval if outcome claims are intended, licensed outcome instruments, app-store and platform fees, sensor or headset hardware, AI model licensing and inference costs, ongoing content expansion, and translation and cultural adaptation for each additional language. A Software as a Medical Device programme — design controls, quality management system, multi-site pivotal trial, regulatory submission — typically adds several hundred thousand to several million dollars and one to three years, and should be scoped as a separate programme from the outset.

Twelve Mistakes That Waste Budgets

• Building for all eight domains at once. The product ends up shallow everywhere, has no primary outcome to evaluate, and no practitioner can say when to use it.
• Treating accessibility as a late polish pass. Sensory controls, contrast, switch access and reduced motion are architectural. Retrofitting them means rebuilding the UI layer.
• Testing with neurotypical proxies. Standard QA finds functional bugs. Only autistic testers find the experiential failures that determine whether the product gets used twice.
• Copying mainstream game conventions. Variable reward, surprise, time pressure, forced social features and loss-aversion mechanics are effective in commercial games and counterproductive here, and they raise the problematic-use risk this population already carries.
• Shipping without a practitioner dashboard. The child is the user; the adult is the buyer. No adult view means no institutional sale.
• Overclaiming outcomes. Claiming behavioural improvement or therapeutic effect without evidence loses clinical buyers and can move the product into regulated device territory.
• Adding instrumentation later. Every session collected before proper instrumentation is unusable for evidence. Decide the data model before the first sprint.
• Ignoring the adult setup journey. If a teacher cannot get twelve children configured in ten minutes, the product does not survive contact with a real classroom.
• Assuming reliable connectivity. Schools have restricted networks, clinics have dead spots, and families have data limits. Offline capability is frequently decisive, especially for AAC.
• Designing a single cultural and linguistic context. The fastest-growing identified populations are in communities previously under-diagnosed. A monolingual, single-culture product excludes exactly the users the data says are arriving.
• Using camera-based emotion inference without checking the AI Act and biometric law. A feature that is legal in a US clinic may be prohibited in an EU classroom and may create biometric-consent obligations everywhere. Make it optional, off by default, and on-device.
• Planning revenue on reimbursement that does not exist for the category. Medicare's digital mental health treatment codes apply only to FDA-cleared devices in a specific classification. A business case that depends on billing codes for an unregulated practice game will not survive due diligence.

Why Capermint for Accessible Game Development

Capermint builds games and interactive applications across Unity, Unreal, HTML5 and native mobile. In this category Capermint works as an engineering partner to organisations that hold the domain expertise — schools, clinics, researchers, nonprofits, adult services and founders — and that need a team able to translate that expertise into a product that is accessible, properly instrumented and compliant with the regimes their buyers require.

Capermint Development Services

Key Terms

ASD (Autism Spectrum Disorder)

The clinical diagnostic term used in DSM-5 and ICD-11. Many autistic people and self-advocacy organisations prefer simply "autistic" or "autism," reserving "ASD" for clinical and research contexts. Sensory processing differences are a core diagnostic characteristic.

Serious game

A game designed with a primary purpose beyond entertainment — skill development, assessment or therapy — while retaining game mechanics such as challenge, progression, feedback and player agency. Distinct from gamification, which adds points and rewards to a non-game task.

Digital therapeutic (DTx)

Software that delivers a clinical intervention and makes a treatment or mitigation claim, requiring regulatory authorisation as a medical device. Distinct from general-wellness software, which makes no medical claim.

Game-based assessment

Use of game-like tasks, eye-tracking or observed play to screen for, aid diagnosis of, or measure a condition. Diagnostic or screening use is a regulated medical-device activity.

Effect size (SMD, Hedges' g)

A standardised measure of the magnitude of a difference between groups, allowing comparison across studies using different scales. Conventionally around 0.2 is small, 0.5 moderate and 0.8 large. Reported in meta-analyses as a pooled estimate across multiple trials.

Meta-analysis

A statistical synthesis pooling results from multiple independent studies to produce a single overall estimate of effect. A network meta-analysis compares several interventions at once; a three-level meta-analysis accounts for multiple effect sizes drawn from the same study. Reliability depends entirely on the quality and comparability of the included studies.

Randomised controlled trial (RCT)

A study in which participants are randomly allocated to the intervention or a comparison condition. A sham or active control (for example, a non-therapeutic VR experience) is stronger than a no-treatment control because it separates the effect of the intervention from the effect of attention and novelty.

AAC (Augmentative and Alternative Communication)

Tools and methods supporting communication for people who do not use speech, or whose speech does not meet all their communication needs. Includes symbol boards, speech-generating apps and text-to-speech systems. Used for real communication, not as practice.

Gestalt language processing

A language development pattern in which a person acquires and uses whole phrases or "chunks" before breaking them into individual words, common among autistic people. Many older AAC tools built around single-word grids support it poorly.

Minimally speaking

A description for people who use few or no spoken words to communicate. Preferred by many advocates over "non-verbal," which wrongly implies an absence of language or communication.

Joint attention

Sharing focus on an object or event with another person, using gaze, gesture or pointing. A core early target in developmental interventions and a common early-years game goal.

Co-design / participatory design

A method in which the people a product is for participate as design partners throughout, shaping decisions rather than reacting to finished work. Published participatory frameworks exist specifically for serious games in autism.

Sensory processing differences

Variation in how sensory input is registered, filtered and responded to. Recognised clinically long before formal inclusion as a core characteristic in DSM-5, and the single most decisive factor in whether a digital product is usable.

Interoception

Awareness of internal bodily signals such as heart rate, hunger or the physical sensations of emotion. A common target in regulation-focused games, including biofeedback designs.

Executive function

The set of cognitive processes covering working memory, planning, task-switching, inhibitory control and self-monitoring. A frequent target domain for game-based intervention, with a pooled effect size of SMD −0.43 in 2025 meta-analysis.

Generalisation

Whether a skill practised in one setting transfers to another — in this context, from the game to real life. The central unresolved limitation across the entire literature and the most valuable thing a new product can measure.

Exergame

A game requiring physical movement as its primary input. A randomised crossover trial found a single 20-minute exergame session improved inhibitory control in autistic children compared with active traditional games and painting.

Biofeedback game

A game in which a physiological signal (usually heart rate or breathing) changes gameplay, so the player practises regulating that signal. Small RCTs report improved control of anger expression in children.

Adaptive difficulty

Automatic adjustment of task difficulty to the player's performance. In this category the adjustment logic must be transparent and logged so that skill growth can be separated from the algorithm easing the task.

Prompting hierarchy

A graded sequence of assistance — from full physical or modelled prompts through gestural and verbal cues to independent performance — used in behavioural teaching. Games serving clinical settings record the prompt level at which each response was achieved.

Errorless learning

A teaching approach that prevents or immediately corrects errors so the learner practises only correct responses, then fades support. The digital expression is Principle 8: low-stakes or errorless failure.

Naturalistic developmental behavioural intervention (NDBI)

A family of play-based, child-led approaches (including the Early Start Denver Model, Pivotal Response Treatment and JASPER) that embed learning in natural routines and target social motivation and joint attention.

IEP (Individualized Education Program)

The US legal document setting a student's special-education goals, services and measurement methods. Equivalent frameworks exist elsewhere (EHC plans in England). Game data becomes usable in schools when it maps to IEP goal language.

Masking

Consciously or unconsciously suppressing autistic traits to fit social expectations. Widely reported in autistic girls and women, associated with later diagnosis and with mental-health cost; relevant to co-design recruitment and to adult-facing product design.

Emotion recognition system (EU AI Act definition)

An AI system for identifying or inferring emotions or intentions of natural persons on the basis of their biometric data. Prohibited in EU workplaces and education institutions except for medical or safety purposes; otherwise high-risk.

SaMD (Software as a Medical Device)

Software intended for a medical purpose without being part of a hardware medical device. If a product claims to diagnose, treat or mitigate a condition, it may meet this definition and require FDA authorisation via 510(k), De Novo or premarket approval, or CE marking under the EU Medical Device Regulation.

WCAG 2.2 level AA

The Web Content Accessibility Guidelines conformance level used as the practical benchmark in most public-sector procurement. Covers contrast, keyboard operability, text alternatives, focus visibility, target size and the three-flashes-per-second threshold.

VPAT (Voluntary Product Accessibility Template)

A standardised document reporting how a product conforms to accessibility standards. Frequently requested during US public-sector and education procurement, and often a gating requirement rather than a nice-to-have.

Single-case design

A research methodology measuring change within individuals across repeated baseline and intervention phases. Well established in special education research and a practical, credible evidence tier for products with modest budgets.

Neurodiversity-affirming design

An approach treating neurological differences as natural human variation requiring accommodation and support rather than correction toward a neurotypical norm. Increasingly an explicit requirement in nonprofit and public-sector tender criteria.

Frequently Asked Questions

References and Evidence Base

---