With various levels of automated vehicles on the verge, the human will be put in the operator role and this poses historic human-machine interaction challenges regarding sustained attention, mental workload, and engagement. Augmented reality based interfaces may help address some of these problems. We presume that more lifelike interfaces, in contrast to static icon based interfaces, could reduce the operators cognitive strain, by allowing him to store information externally and reducing cognitive switching between reality and interface. To evaluate augmented reality interfaces we introduce naturalism, as an interface property dimension ranging from arbitrary to natural In this research, we question whether a more naturalistic interface could improve performance, vigilance, and subjective evaluation (workload, acceptance, and engagement). For this we tested an arbitrary, a semi- natural and a natural automation status interface. These three interfaces were tested in a driving simulator setup using non-interactive real-life driving videos, to which 28 participants had to respond when an automation status error event occurred, which appeared after a precipitating cue such as appearing road construction. Results seem to suggest that the semi-natural interface improves reaction time performance most, followed by the natural interface. With the natural interface vigilance seems to be improved, and mental demand is decreased, but all results should be interpreted with caution due to small sample sizes and other limitations. It seems that the dynamic response to precipitating cues, an intrinsic property of the semi-natural and natural interfaces, was of influence. The reduced mental demand seems to support our presumed mechanism of reduced cognitive strain. For further research, a more challenging task design and eye-tracking could be promising. To conclude, this research showed that more natural interfaces show potential to improve safety and comfort.