The objective of this study was to evaluate associations of inline milk fat-to-protein (F:P) data surrounding the detection of subclinical ketosis (SCK) in robotic milking herds. The ratio of fat-to-protein has been proposed as a tool for detecting SCK with moderate accuracy in past studies, however, some producers with robotic milking systems are trusting inline F:P data as their main source of SCK screening. To assess the validity of these data in commercial settings, we monitored 484 cows from 9 robotic milking herds for their first 3 wk of lactation, taking blood samples 1x/wk (n = 1427). Positive cases of SCK were defined by whole blood β-hydroxybutyrate (BHB) concentrations ≥ 1.2 mmol/L. Milk data were collected from the robotic systems on each farm for each cow and converted into 4 different F:P values: 1) value same day of BHB test; 2) 5-d centered moving average (CMA); 3) 5-d backward moving average (BMA); 4) 5-d forward moving average (FMA). In linear regression models, all 4 values were associated with BHB (P < 0.001), but slope estimates varied and R² was low: same day (slope = 0.9, R² = 0.07); CMA (slope = 1.0, R² = 0.07); BMA (slope = 0.7, R² = 0.04); FMA (slope = 1.2, R² = 0.09). In logistic regression models, the odds of having SCK increased with every 0.1 unit increase from the mean (1.16) using same day F:P (OR = 1.35, 95% CI = 1.25–1.47; P < 0.001) and CMA (OR = 1.39, 95% CI = 1.27–1.51; P < 0.001). The same increase in F:P from mean BMA (1.14) and FMA (1.17) were associated (P < 0.001) with 1.22 and 1.49 times the odds of SCK, respectively. For all 4 F:P variations, sensitivities and specificities of different F:P thresholds with SCK status were evaluated using chi-squared tests. As the F:P threshold was raised from 1.15 to 1.22, sensitivity decreased (range: 73 to 45%) while specificity increased (range: 53 to 71%). The F:P cut-offs at which a balance was reached between sensitivity and specificity were 1.17 to 1.20; however, even at these values there were high rates of false positives and negatives (range: 33 to 40%). These results suggest that inline milk F:P data should not be solely used to detect SCK in robotic milking herds.