Background Group cognitive behaviour therapy for psychosis ( GCBT p) has shown to be effective in diminishing symptoms, as well as in improving other psychosocial dimensions such as self‐esteem. But little is known regarding the processes that generate these therapeutic improvements and might be harnessed to further improve its effectiveness. Objectives The current study aimed at investigating these processes, particularly those linked to interpersonal relationships. Design The participants were all assessed at baseline, were given 24 sessions of GCBT p over the course of 3 months and were assessed again at post‐treatment as well as 6 months later (9 months from baseline). Method Sixty‐six individuals with early psychosis took part in a study of GCBT p where therapist alliance and group cohesion were assessed at three time points during the therapy, and punctual (each session) self‐perceptions on symptoms and optimism were collected. Results Improvements in symptoms (BPRS), self‐esteem (SERS‐SF) and in self‐perceived therapeutic improvements (CHOICE) were linked to specific aspects of the alliance, group cohesion, as well as optimism. The variables retained were not always overall scores, suggesting the importance of the variables at key moments during the therapy. Conclusions The results clearly demonstrate the importance of the alliance and group cohesion, together significantly explaining improvements measured at post‐therapy or follow‐up. Practitioner points This study has attempted to focus mostly on relational aspects, as well as on self‐perceptions, in the context of a GCBT p for individuals with early psychosis. This study also showed that these therapeutic relationships are especially useful when they are more stable and at specific moments during the therapy, namely when more difficult psychological work is done.