Background Prior evidence suggests that behaviours closely related to the intervention delivered for autism are amenable to change, but it becomes more difficult when generalising treatment effects beyond that immediate context. Objectives The objectives were (1) to test an early autism social communication intervention designed to promote child social communication change in the naturalistic contexts of both home and education, with an additive effect on overall child symptom outcomes, and (2) to conduct a mechanistic study investigating the transmission of treatment effects within and across contexts to an overall treatment effect. Design The trial was a three-site, parallel-group, randomised controlled trial of the experimental treatment plus treatment as usual and treatment as usual alone. The primary intention-to-treat analysis used analysis of covariance. The mechanism analysis used regression models to test mediation of the primary outcome by parent–child and education staff (learning support assistant)–child social interactions. Setting The study took place in three urban/semiurban regions in Manchester, Newcastle upon Tyne and London. Participants Children aged 2–11 years who met the criteria for severe autism. Interventions The Preschool Autism Communication Trial was adapted to parallel components within home and educational settings using in-person and remote delivery. Treatment as usual was the control condition. Main outcome measures The primary outcome was autism symptoms on the Autism Diagnostic Observation schedule-2. The secondary outcomes were Brief Observation of Social Communication Change, dyadic social interaction between child and parent or learning support assistant, reported language, functional outcome and reduction in child disruptive behaviour. Outcomes were measured at baseline and at the 12-month end point in all settings; interim mechanism measurements were taken at 7 months. Results Participants (n = 249; 122 in the PACT-G group and 127 in the treatment-as-usual group; 51 were female and 197 were male) received a median of 10 (interquartile range 8–12) sessions at home and 8 (interquartile range 5–10) sessions in an educational setting. We found no significant treatment effects on the end-point Autism Diagnostic Observation Schedule-2 primary outcome (–0.04, 95% confidence interval –0.26 to 0.18; p = 0.734), on the end-point Brief Observation of Social Communication Change secondary outcome (–0.03, 95% confidence interval –0.31 to 0.25; p = 0.85) or on language, repetitive behaviour, adaptive behaviour and child well-being. We did find significant treatment effects on dyadic interactions (increased parent synchronous response 0.54, 95% confidence interval 0.39 to 0.69; p = 0.001); child initiations with a parent (0.27, 95% confidence interval 0.12 to 0.41; p = 0.001); learning support assistant synchronous response (0.32, 95% confidence interval 0.14 to 0.49; p = 0.001); child initiations with a learning support assistant (0.21, 95% confidence interval 0.06 to 0.36; p = 0.005); and unblinded measures of improved parental well-being and child disruptive behaviour across home and educational settings. Adult (parent/learning support assistant) synchronous responsiveness in a home/education setting improved child dyadic social initiation. The child dyadic social initiation was also associated with child symptoms on researcher Brief Observation of Social Communication Change. Limitations The delivered sessional dosage was 83% of that planned in the home setting and 67% in the educational setting, with 5.5% of home sessions and 5% of educational sessions deemed ‘unacceptable’, particularly for remote delivery. A change of therapy learning support assistant was experienced by over one-third of children by the mid-point of the trial, by another third by the end point, and by one-fifth at both points. Conclusions The multicomponent Paediatric Autism Communication Trial – Generalised (PACT-G) treatment for a child in a home or educational setting did not produce the hypothesised improvement in child autism symptomatology or adaptive behaviour, but did produce significant improvements in proximal adult–child reciprocal dyadic communication. Future work Future work will involve building on these results towards a further understanding of delivery options, dosage and multicomponent extension of social communication interventions for young children with autism in naturalistic settings. Trial registration Current Controlled Trials ISRCTN25378536. Funding This project was funded by the Efficacy and Mechanism Evaluation programme, a Medical Research Council and National Institute for Health and Care Research (NIHR) partnership. This was also part funded by the NIHR Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King’s College London. This will be published in full in Efficacy and Mechanism Evaluation; Vol. 9, No. 3. See the NIHR Journals Library website for further project information.