The Beck Cognitive Insight Scale (BCIS) is composed of two subscales, self-reflectiveness and self-certainty, assessing reflectiveness and openness to feedback, and mental flexibility. Delusions have previously been associated with low cognitive insight. The aim of this study was to determine whether changes in BCIS scores predict changes in delusional beliefs. The study is a secondary analysis of a previously published randomized controlled trial. All participants had a psychotic disorder diagnosis and received treatment as usual, with half of them also receiving the cognitive restructuring intervention 'Michael's game'. Participants were assessed at three different times: at baseline (T1), at 3months (T2), and at 9months (T3). Cognitive insight was measured with the BCIS, belief flexibility with the Maudsley assessment of delusions schedule (MADS), and psychotic symptoms with the Brief Psychiatric Rating Scale (BPRS). A total of 172 participants took part in the trial. After using generalized estimating equation (GEE) modelling, we observed (1) significant main effects of BCIS self-certainty and Time and (2) significant Time×BCIS self-certainty and Time×treatment group interaction effects on belief flexibility. Improvements in self-certainty (i.e., decrease in scores) were associated with more changes in conviction over time, more accommodation, improved ability in ignoring or rejecting a hypothetical contradiction and increased use of verification of facts. Medication and BPRS total scores were controlled for in the GEE analyses at their baseline values. Overall improvement in BCIS self-certainty scores over time predicted better treatment outcomes as assessed with MADS items. Treatments for patients with psychosis should focus on improving cognitive insight as this seems to improve overall treatment outcomes and recovery. The Beck Cognitive Insight Scale can be used to measure changes during treatment and can predict treatment outcomes.