BACKGROUND: Neonatal sepsis is a major cause of mortality in low-resource settings. We assessed how neonatal sepsis guidelines were used in two Zimbabwean hospitals and one Malawian hospital. METHODS: Using routine data collected with the digital health intervention, Neotree, we retrospectively reviewed doctors' and nurses' agreement with national and World Health Organization (WHO) guideline recommendations for antibiotic prescription for sepsis. We compared clinical features and outcomes of neonates who should have received antibiotics as per guideline with those who actually received them and fitted a logistic regression model to identify features associated with prescription. RESULTS: Data were collected between January 2021 and June 2022 from 10,868 neonates: 6,045 admitted to Sally Mugabe Central Hospital (SMCH), 1,094 to Chinhoyi Provincial Hospital (CPH) and 3,729 to Kamuzu Central Hospital (KCH). Complete implementation of national guidelines would increase antibiotics at admission: from 2,188 (38%) to 3,745 (64%) at SMCH, 472 (44%) to 852 (79%) at CPH, and 1,519 (41%) to 3,043 (82%) at KCH. Clinical features of sepsis were frequently not acted on, but the case-fatality rate was lower in those not prescribed antibiotics despite guideline recommendation. Application of WHO guidelines would increase antibiotic prescription to 91% at SMCH, 88% at CPH, and 77% in KCH. Maternal risk factors for sepsis, male gender, low birth weight, older age at admission, and spontaneous vaginal delivery were associated with higher rate of antibiotic prescription. CONCLUSION: Guideline-recommended clinical signs for sepsis are inconsistently used, with clinicians using other features for antibiotic decision-making. Work is needed to revise clinical diagnostic algorithms in low-resource settings to ensure they are useful, usable and contextually appropriate.