Background and aims: Type 2 diabetes (T2D) is considerably heterogeneous due to different pathogenetic mechanisms that may affect response to treatment. Glucagon-like peptide-1 receptor agonists (GLP-1RA) are widely used for the treatment of T2D, with clear evidence of benefits on glucose and body weight (BW) control and cardio-renal protection. However, treatment with GLP-1RA allows to achieve HbA1c target in 40-80% of patients, due to interindividual variability. We aimed to assess the efficacy of GLP-1RA according to the phenotypes of adult-onset diabetes identified by Ahlqvist and colleagues. Materials and methods: In this retrospective mono centric study, every individual attending our outpatient clinic since 2013 was evaluated for eligibility up to October 2023. Main inclusion criteria were age at T2D diagnosis >50 years, T2D duration <5 years, BMI >25 kg/m2, first users of a GLP-1RA with at least one follow-up visit at 6-12 months. Main exclusion criteria were type 1 diabetes, latent autoimmune diabetes in adults, ketoacidosis. The assignment to T2D phenotypes (MARD, mild age-related diabetes; MOD, mild obesity-related diabetes; SIDD, severe insulin deficient diabetes; SIRD, severe insulin resistant diabetes) was performed via the algorithm developed by Bello-Chavolla and colleagues. The primary outcome was difference in HbA1c change from baseline, evaluated with ANOVA. SHapley Additive exPlanations (SHAP) allowed to rank predictors of HbA1c reduction. Kaplan-Meier analysis and log-rank test were used to estimate differences in time to treatment failure, defined as time to HbA1c>7.0%. Results: In 128 patients, the SIDD phenotype was associated with a significantly greater HbA1c reduction (-1.9% vs -0.67% [MARD], -0.75% [MOD], -0.56% [SIRD]; p<0.001) following GLP-1RA initiation after a median follow-up of 8 months. However, SHAP analysis identified baseline HbA1c, rather than SIDD phenotype, as the most relevant predictor of HbA1c change, accounting for ~0.5% HbA1c reduction. Yet, belonging to the SIDD phenotype was associated to earlier treatment failure (p<0.01). Conclusion: Easily available clinical variables, such as baseline HbA1c, might be more useful to predict response to GLP-1RA than T2D subclassifications. However, the diverse pathogenesis underlying T2D phenotypes might account for differences in treatment durability.