In an ever-changing environment, flexible control of behaviour is crucial for survival. Across species, adaptive processes like attention and motivation manifest as flexibly fluctuating, brain-wide states. Although these states likely share commonalities across species, classical tasks are highly species-specific, preventing direct cross-species comparisons.Here we examine naturalistic fluctuations of attention across three species—mice, monkeys, and humans—completing the same highly immersive virtual reality (VR) foraging task (Fig.1A). This allows us to continuously track behavioural decision-making as subjects navigate a complex, naturalistic visuomotor challenge that requires minimal training. We trained a Hidden Markov Model (HMM) on the same behavioural readouts across all three species to identify behavioural states. The HMM categorised each trial into one of four distinct states, based on various speed and accuracy metrics (Fig.1C). Both state durations and state transitions were highly consistent across species. Furthermore, transitions between behavioural states with similar speed levels were slightly more likely, than between states with similar accuracy levels (Fig.1B). There were also some differences, with monkeys and humans typically increasing in speed and accuracy towards the end of the session, while mice tend to show a decline.These observations confirm that in a naturalistic context which allows us to quantify behaviour in a way that can be applied across species, we can identify global cognitive states that are shared and conserved across the taxa. At the same time, this approach also highlights differences that possibly reflect genuinely unique strategies employed by each species.