Purpose To describe the mediolateral (ML) gap balance in pre-resection arthritic knees undergoing robot-assisted (RA) total knee arthroplasty (TKA) within the nine phenotypes of the Coronal Plane Alignment of the Knee (CPAK) classification. Methods A total of 1124 RA TKA cases were retrospectively reviewed. ML balance was calculated using a digital ligament tensioning device following tibial resection and prior to any femoral resection throughout flexion (10°, 40°, and 90°). Lateral distal femoral angle (LDFA) and medial proximal tibial angle (MPTA) were landmarked intra-operatively and adjusted for wear based on hip-knee-ankle angle (HKA). Knees were then subdivided into CPAK categories based on the arithmetic HKA (aHKA) and joint line obliquity (JLO). Differences in balance between CPAK phenotypes were compared. Results For aHKA, a greater medial gap was observed in varus compared to neutral or valgus knees at 10° (5.2 ± 3.0 mm vs 3.3 ± 3.2 mm vs − 0.4 ± 5.0 mm respectively; p < 0.0001). A similar trend was observed for JLO in distal apex compared to neutral and proximal apex knees at 10° (3.4 ± 3.6 vs 1.1 ± 5.0 vs − 2.4 ± 6.3 respectively; p < 0.0001). The greatest ML balance differences were found when combining JLO and aHKA at 10° (type I vs type VI: ∆6.6 mm; p < 0.0001). Conclusion There is a high degree of variability in arthritic ML balance within and between each CPAK phenotype. A single alignment philosophy based on bony landmarks alone may not be sufficient to balance all knees. Utilisation of robotics and digital ligament tensioning devices serves as important adjuncts to characterize the patient-specific soft tissue profile.