Truace
Subscribe
70/100

Moderate evidence

Cite
More
Export CSVExport JSONEmailLinkedInX

AI games gave autistic children a moderate social-skills boost, a review of 14 studies finds

Study title: Can AI-driven games enhance social skills for autistic children? — A three-level meta-analysis

Evidence type
Systematic review or meta-analysis
Sources
1 sourcePeer-reviewed
Published
3 min read
Truvace record
TRV-2026-2744cite or check it

Study title: Can AI-driven games enhance social skills for autistic children? — A three-level meta-analysis

A three-level meta-analysis pooled 14 publications (37 effect sizes) from nine databases searched up to March 2026 and found that AI-driven games improved social skills in autistic children by a moderate amount (Hedges' g = 0.511, 95% CI 0.346 to 0.677, p < 0.001).

Read the study abstract

Background Gamified interventions for autistic children are becoming increasingly diverse, enhancing their presence, immersion, and sense of accomplishment to improve various social skill indicators. Existing systematic reviews have primarily evaluated intervention effectiveness from the perspectives of game design and intervention objectives. However, these findings are inconsistent and fail to inspect specific improvements in social skills among autistic children. Methods To address this gap, this study adopts…

61Gain side

How the two sides scoreClose call

57Problem side

The gain

A meta-analysis of 14 studies (37 effect sizes) found AI-driven games improved autistic children's social skills by a moderate amount (Hedges' g = 0.511, 95% CI 0.346 to 0.677), with the largest gain (g = 0.963) for games targeting attention and social skills together.

The problem

In the same meta-analysis of 14 AI-game studies for autistic children, games aimed at anxiety and at real-world interaction skills showed effects well below the overall moderate gain, and neurofeedback-based games had a small, non-significant effect (g = 0.199).

A fixed formula sets both scores from the evidence and its sources. Reader votes don’t move them.

What the scores mean

G is the gain side, out of 100. It goes up when the gain is specific, when it measurably helped people and when the sources behind it are strong.

P is the problem side, out of 100. It goes up when the problem is specific, when it measurably hurt people and when the sources behind it are strong.

In brief

Results varied by goal and technology. Games targeting emotion recognition and expression gave g = 0.536 and games targeting attention plus social skills gave g = 0.963. Adaptive learning (g = 0.780) and computer vision with deep learning (g = 0.952) did best; neurofeedback (g = 0.199) was low and non-significant.

The authors call the evidence preliminary. Only 14 studies qualified, the included designs varied from randomized trials to single-group pre-post studies, and games for anxiety and real-world interaction skills fell short, suggesting skills learned on screen may not carry over to everyday life.

Main points

  1. The meta-analysis screened 1,326 records and kept 14 studies published 2019 to 2025, yielding 37 effect sizes.
  2. Overall effect on social skills was g = 0.511 (95% CI 0.346 to 0.677); heterogeneity came mainly from differences between studies (I² = 52.53%).
  3. Dropping the pre-post studies raised the pooled estimate to g = 0.552 (95% CI 0.310 to 0.795) and cut heterogeneity.
  4. Games aimed at emotion recognition and expression gave g = 0.536; attention plus social skills gave g = 0.963.
  5. Only AI technology type was a significant moderator (Q(5) = 7.481); adaptive learning (g = 0.780) and computer vision with deep learning (g = 0.952) did best.
  6. Feedback (85.71%), rewards (71.43%) and level design (57.14%) were the most common of 14 gamification elements.
  7. Medium-frequency sessions (g = 0.692) outperformed low-frequency (g = 0.413) and high-frequency (g = 0.401) ones in subgroup analysis.

What the review found when it combined the studies

These are the review’s own pooled results, copied from its abstract. Truvace did not recalculate them.

  • Emotion recognition and expression in autistic children after AI-driven games

    Hedges' g 0.536 (95% CI 0.221 to 0.850)

    14 studies · search to March 2026

    The review’s own words
    An effect size analysis for different intervention targets showed that, with emotional recognition and expression (DE) as the baseline, the effect size was g = 0.536 (95% CI: 0.221, 0.850, p < 0.01), indicating that AI-driven games produce a moderate and significant effect on this target.

  • Social skills of autistic children after AI-driven gamified interventions

    Hedges' g 0.511 (95% CI 0.346 to 0.677)

    14 studies · search to March 2026

    The review’s own words
    The overall effect size of AI-driven gamified interventions on the social skills of autistic children was Hedges’g = 0.511 (95% CI [0.346, 0.677], p < 0.001), indicating a moderate positive effect.

  • Social skills of autistic children after AI-driven games, excluding pre-post studies

    Hedges' g 0.552 (95% CI 0.310 to 0.795)

    14 studies · search to March 2026

    The review’s own words
    The pooled effect size estimate changed from 0.511 (95% CI: 0.346–0.6767, p < 0.001) to 0.552 (95% CI: 0.310–0.795, p = 0.0001).

  • Combined attention and social skills in autistic children after AI-driven games

    Hedges' g 0.963 (95% CI 0.165 to 1.760)

    14 studies · search to March 2026

    The review’s own words
    Studies targeting the dual development of attention and social skills (DA + DS) showed a large effect size ( g = 0.963, 95% CI [0.165, 1.760]).

Code checks every number above against the review’s text before it is shown. The short labels are an AI model’s reading of that text, so the review’s own sentence sits under each one. A pooled figure is only as strong as the studies behind it. How Truvace uses pooled results

The full story

The team searched Scopus, PubMed, ScienceDirect, Web of Science, IEEE Xplore, Springer Nature Link, Engineering Village, ERIC and Google Scholar. Of 1,326 records, 789 were inaccessible and 108 duplicates, and after screening 14 studies with calculable effect sizes remained. Included designs were randomized controlled trials, non-randomized controlled trials and single-group pre-post studies.

Subgroup results differed by goal: attention (g = 0.644), communication (g = 0.650) and daily living skills (g = 0.618) were moderate but their confidence intervals included or approached zero, while anxiety and real-world interaction outcomes fell short. The authors say most games train interaction only in virtual or lab settings, leaving a gap in transfer to real life.

Moderator results by technology: adaptive learning g = 0.780, computer vision with deep learning g = 0.952, machine learning g = 0.412, standalone computer vision g = 0.295 and neurofeedback g = 0.199. The authors attribute the neurofeedback result partly to noisy EEG signals and fatigue from brain-computer interface devices in children.

The authors flag risks including sensory overload from flashing visuals and loud sounds, reduced real-life contact from screen-based training, data privacy concerns from stored video and clinical records, and high development costs. They call for more high-quality randomized trials and more objective outcome measures.

What this doesn’t fix

Only 14 studies were included, designs ranged from randomized trials to single-group pre-post studies, heterogeneity was significant, and the authors describe the moderation analysis as exploratory and underpowered; behavioral observation measures may carry assessor bias.

Sources

1 on file
  1. Frontiers in Psychologydoi:10.3389/fpsyg.2026.1847426
    Peer-reviewed2026-07-03

Talk it over

Four questions to start onThis record found a gain and a problem in the same place. Which matters more to you, and who might answer differently?

Start with a question

AI games gave autistic children a moderate social-skills boost, a review of 14 studies findshttps://truvace.com/item/can-ai-driven-games-enhance-social-skills-for-autistic-children-a-three-level-me

  1. This record found a gain and a problem in the same place. Which matters more to you, and who might answer differently?

    The gain A meta-analysis of 14 studies (37 effect sizes) found AI-driven games improved autistic children's social skills by a moderate amount (Hedges' g = 0.511, 95% CI 0.346 to 0.677), with the largest gain (g = 0.963) for games targeting attention and social skills together.

    The problem In the same meta-analysis of 14 AI-game studies for autistic children, games aimed at anxiety and at real-world interaction skills showed effects well below the overall moderate gain, and neurofeedback-based games had a small, non-significant effect (g = 0.199).

    Answer in the thread

  2. How do we know? What can this kind of evidence show, and what can’t it?

    The evidence This rests on a systematic review or meta-analysis, from Frontiers in Psychology.

    Answer in the thread

  3. What does it leave open? What else would you want to know before acting on it?

    The record’s own caveat Only 14 studies were included, designs ranged from randomized trials to single-group pre-post studies, heterogeneity was significant, and the authors describe the moderation analysis as exploratory and underpowered; behavioral observation measures may carry…

    Answer in the thread

  4. Have you seen AI used this way where you study, work or live? Did it go the same way?

    Answer in the thread

The questions are the same on every record. The details under them are taken from this record as written.

What readers say

More from Health