Social Skills Training Helps Autistic Teens, But the Evidence Is Weaker Than It Looks

For decades, social skills training has been one of the most widely prescribed supports for autistic young people…

The distinction between observer-rated and informant-reported outcomes deserves closer attention, because it speaks to one of the most persistent methodological challenges in autism intervention research. When the meta-analysis found a moderate effect on directly observed social behaviour (SMD = 0.49) alongside smaller effects on parent-reported outcomes (SMD = 0.42), it aligned with a pattern seen across the broader intervention literature: structured sessions, often delivered in group settings with explicit coaching, can produce visible changes in how young people behave during observed interactions, while the translation of those gains into everyday settings perceived by parents appears more modest. The complete absence of a significant effect on teacher reports is particularly noteworthy, since teachers observe young people in the most socially demanding naturalistic environment most face — the classroom and schoolyard — where demands for spontaneous social initiation, peer negotiation, and emotional regulation are high and adult scaffolding is minimal. That findings did not generalise to this setting suggests that skills learned in clinic-based or research-based group formats may remain context-bound, a phenomenon long described in the learning literature as limited stimulus generalisation.

The stark contrast between waitlist comparisons and active control comparisons is arguably the single most important finding for interpreting the evidence base. When social skills interventions were benchmarked against a waitlist, effects appeared across observed behaviour, adolescent self-reported social knowledge and performance, and parent report. When the same interventions were compared against an active control — a condition offering some plausible alternative activity, attention, or structured engagement — none of the reported measures reached statistical significance. This pattern is familiar from psychotherapy trials more broadly, where waitlist designs systematically inflate effect estimates because they capture not only the specific ingredients of an intervention but also expectancy effects, therapist attention, group belonging, structured time, and simple regression to the mean in a group that has been waiting without support. The authors’ call for future research to prioritise active control conditions therefore reflects a growing consensus that the field needs designs capable of isolating the specific contribution of social skills training content from the generic benefits of participating in any structured, socially engaging programme.

It is also worth considering what the very large effect on adolescent-reported social knowledge (SMD = 2.03) does and does not mean. An effect of that magnitude is unusually large for behavioural intervention research, and the most likely explanation is that the measures involved — instruments such as the Test of Adolescent Social Skills Knowledge and its young adult counterpart — assess explicit, teachable content that maps closely onto what the interventions directly instructed. A participant who has just completed a curriculum covering conversation rules, emotion recognition, or perspective-taking will predictably answer more knowledge questions correctly. Knowing what to do, however, is not the same as doing it in real time under the cognitive and emotional load of an actual social encounter. The much smaller effect on adolescent-reported social performance (SMD = 0.45) within the same meta-analysis illustrates this gap between declarative social knowledge and enacted social behaviour, a gap that has shaped theoretical debates about whether social skills training should target knowledge, performance, or the underlying motivational and perceptual processes that influence spontaneous social engagement.

The age range covered by the review, 12 to 25 years, spans a developmental period with distinct social demands. Early adolescence brings rapidly changing peer hierarchies, increased sensitivity to social evaluation, and the emergence of romantic and identity-related social challenges that differ substantially from the playground contexts of childhood. Late adolescence and early adulthood bring transitions into employment, tertiary education, and independent living, each with their own social conventions that are rarely taught explicitly to neurotypical young people, let alone scaffolded for autistic ones. Interventions designed for one segment of this range may not transfer well to another, and the heterogeneity inherent in pooling trials across such a wide age span is a genuine limitation of any meta-analytic synthesis in this area. The finding that effects were detectable at all across this range is informative, but it also raises the question of whether different developmental windows call for fundamentally different intervention targets — for example, peer-network inclusion strategies in early adolescence versus workplace social navigation in the mid-twenties.

The emphasis on observed outcomes measured with instruments such as the Contextual Assessment of Social Skills is significant because direct observation of conversational behaviour — including measures of question-asking, shared topics, and the quality of back-and-forth exchange — provides a window on performance that questionnaires cannot. Observational measures, however, carry their own constraints: they typically capture behaviour in a contrived interaction with a research confederate or unfamiliar peer, in a clinic or laboratory setting, over a brief period. A young person may perform well in such a structured encounter while still struggling in the unscripted, faster-paced, and less predictable interactions of daily life. Conversely, some gains that matter to families — reduced social anxiety, increased willingness to initiate, greater comfort in group settings — may not register on the conversational metrics that observation protocols capture. The moderate observed effect reported in the meta-analysis should therefore be read as evidence of change in a specific, measurable domain rather than as a wholesale transformation of social functioning.

The inclusion of both randomised controlled trials and one quasi-controlled study, totalling 1244 participants, reflects the reality that the adolescent and young adult autism intervention literature remains thin relative to the childhood literature. The authors note that many included studies exhibited low quality, and this is consistent with long-standing critiques of the social skills training evidence base: small samples, absence of blinding where outcomes rely on unblinded parent report, inconsistent use of manualised protocols, short follow-up periods, and limited attention to maintenance of gains after intervention ends. Low study quality does not mean findings are worthless, but it does mean that point estimates from meta-analysis should be interpreted with appropriate caution, since biased smaller studies can inflate pooled effects. The use of GRADE-style critical appraisal in the review signals an awareness of this problem and provides readers with a structured way to weigh confidence in each outcome domain.

Another dimension worth considering is the diversity of intervention models captured under the umbrella term social skills intervention. The trials in this literature range from structured group curricula with explicit lesson sequences and homework tasks, to cognitive behavioural therapy-informed programmes targeting social anxiety alongside skills, to facilitation-based approaches that arrange peer interactions and coach in the moment, to technology-assisted formats using video modelling or virtual environments. These approaches rest on different theoretical assumptions — some treat social skills as learnable procedures to be rehearsed, others as socially mediated behaviours best acquired through naturalistic peer interaction, and others as downstream expressions of anxiety or rigid cognition that must be addressed first. Pooling them in a single meta-analysis yields a useful headline answer but obscures the possibility that different models produce different patterns of benefit, different durability, and different suitability for particular developmental stages or co-occurring profiles.

The question of who benefits most also remains open. Autistic young people differ enormously in language level, intellectual profile, co-occurring anxiety or attention difficulties, and their own goals regarding social connection. Some participants in social skills programmes are highly motivated to build friendships and experience the intervention as a welcome opportunity; others attend under family or clinical pressure and may engage minimally. Averages drawn across such heterogeneous samples can conceal meaningful variation in response. The field increasingly recognises that outcomes should be framed around personally meaningful goals — an orientation reflected in instruments such as the Goal Attainment Scaling included among the review’s abbreviations — which allow individualised targets to be quantified and tracked. Whether group-level standardised mean differences adequately capture the kinds of change that matter most to autistic young people themselves is a question that future trials, ideally co-designed with autistic advisors, will need to address.

Finally, the publication timeline of the review — with the search capturing literature indexed across major behavioural science and medical databases — positions it as a timely synthesis for clinicians, educators, and funders making decisions about service provision for a group whose needs have historically been underserved relative to younger autistic children. The finding that benefits appear robust against waitlist but not against active comparison should not be read as a dismissal of social skills interventions, but rather as a specification of what remains unknown: whether the specific content of these programmes outperforms the generic benefits of structured social engagement, and under what conditions, for whom, and at what developmental moment such programmes deliver durable, generalised gains. Answering those questions will require trials with active comparators, longer follow-up, measurement across home, school, and community settings, and outcome frameworks that include the perspectives of autistic young people themselves.

Subject of Research: Effectiveness of social skills interventions for autistic young people aged 12 to 25

Article Title: Effectiveness of social skills interventions in young people aged 12 to 25 years with autism spectrum disorder: a systematic review and meta-analysis

Article References: Velalagan, R., Wu, W. T., John, J. R., Lin, P.-I., & Eapen, V. (2026). Effectiveness of social skills interventions in young people aged 12 to 25 years with autism spectrum disorder: a systematic review and meta-analysis. BMC Psychiatry. https://doi.org/10.1186/s12888-026-08579-9

Image Credits: AI Generated

DOI: 10.1186/s12888-026-08579-9

Keywords: autism spectrum disorder, social skills interventions, meta-analysis, systematic review, adolescents, young adults, randomised controlled trials, BMC Psychiatry, cognitive behavioural therapy, PEERS, social knowledge, active control groups