Augmented Reality (AR) provides useful support for (business) processes in nuclear facilities. The geometrical content to be visualized can be derived from the facility's building information model (BIM) and semantic data or knowledge from a database. To deploy AR, positioning and continuous tracking have to be established. AR positioning in nuclear facilities in particular has hardly been investigated so far, especially not with current AR-capable smartphones and AR headsets. For this research, several AR experiments were carried out in the Mont Terri Underground Rock Laboratory (Switzerland). Using a smartphone and the Meta Quest 3 headset, different positioning and tracking methods were tested for their suitability. With the help of the smartphone, five different marker types were evaluated based on tracking stability, drift, and the influence of external factors such as lighting and occlusion. Using the Meta Quest 3, we propose manual positioning using the existing survey points in Mont Terri. Furthermore, we discuss the extent to which the findings can be transferred to nuclear facilities. Finally, recommendations are provided as to how a tracking-friendly infrastructure can be realized in a nuclear facility, considering the preferred devices and desired AR use cases.