Abstract Antibiotic-resistant Gram-negative bacteraemias (GNB) are increasing in incidence. We aimed to investigate the impact of empirical antibiotic therapy on clinical outcomes by carrying out an observational 6-year cohort study of patients at a teaching hospital with community-onset Escherichia coli bacteraemia (ECB), Klebsiella pneumoniae bacteraemia (KPB) and Pseudomonas aeruginosa bacteraemia (PsAB). Antibiotic therapy was considered concordant if the organism was sensitive in vitro and discordant if resistant. We estimated the association between concordant vs. discordant empirical antibiotic therapy on odds of in-hospital death and ICU admission for KPB and ECB. Of 1380 patients, 1103 (79.9%) had ECB, 189 (13.7%) KPB and 88 (6.4%) PsAB. Discordant therapy was not associated with increased odds of either outcome. For ECB, severe illness and non-urinary source were associated with increased odds of both outcomes (OR of in-hospital death for non-urinary source 3.21, 95% CI 1.73–5.97). For KPB, discordant therapy was associated with in-hospital death on univariable but not multivariable analysis. Illness severity was associated with increased odds of both outcomes. These findings suggest broadening of therapy for low-risk patients with community-onset GNB is not warranted. Future research should focus on the relationship between patient outcomes, clinical factors, infection focus and causative organism and resistance profile.