Haptic steering guidance is an advanced driving assistance system which provides guidance torques on the steering wheel to assist a human driver. Improvements in performance, safety margins and workload have been reported for lane-keeping and curve negotiation tasks. On the other hand, the guidance system instigates increased driver torques, indicating the occurrence of a mismatch between driver and automation intention. A novel, criticality-based control structure was developed, capable of adapting to different driving conditions by using safety margins (operationalized as time-to-line crossing, TLC) as input. Twenty-four participants drove through a single-lane, 10.8 km long road, with as independent variables the road width (normal road width of 3 m, wide road width of 5 m) and the type of guidance received: no guidance (manual), haptic steering guidance based on lateral deviation from the center-line (performance-based guidance, PBG) and criticality-based guidance (CBG). On the normal road, results show similar benefits for both guidance systems compared to manual control, in terms of performance and safety margins. Workload was reduced by PBG, but both guidance systems yielded increased driver torques. On the wide road, participants drove closer to the center-line with PBG, but at the cost of significantly more guidance torques than CBG. Interestingly, this reduction in lateral deviation did not result in a significant improvement for lowest safety margins encountered. No subjective preference was found for either feedback condition. It can be concluded that criticality-based guidance is capable of adjusting to different driving conditions, without compromising safety margins, whilst greatly reducing guidance torques in situations where they are unneeded.