The Rapid Serial Visual Presentation (RSVP) is an experimental approach to BCIs in which a series of images is displayed at a high speed. Participants are asked to differentiate between a set of target images and a set of non-target images, where the P300 ERPs is evoked by the target image, but not by the non-target image [1, 2]. The automatic identification of this response in a robust way allows this paradigm to produce a functional interface relating a user’s internal brain state to events in the external environment. While the RSVP approach produces impressive results in lab-based environments, translation of this technology into consumer contexts requires a better understanding of performance in ecologically valid settings, for example the use of the BCI to enhance experiences in online worlds, metaverse and gaming contexts. Such application scenarios are characterised by much less constrained user behaviour some of which is entirely necessary for the normal expected interactions we typically encounter in such applications. Examples include talking, head and hand movement. In this study we examine how such interactions induce performance degradations in an RSVP paradigm in a quantified way, explicitly articulate the signal processing challenges to mitigate this and present open datasets for researchers to benchmark performance