Distraction is ubiquitous in human environments. Distracting input is often predictable, but we do not understand when or how humans can exploit this predictability. Here we ask whether predictable distractors are able to reduce uncertainty in updating the internal predictive model. We show that utilising a predictable distractor identity is not fully automatic but in part depends on available resources. In an auditory spatial n-back task, listeners (n = 33) attended to spoken numbers presented to one ear and detected repeating items. Distracting numbers presented to the other ear either followed a predictable (i.e., repetitive) sequence or were unpredictable. We used electroencephalography (EEG) to uncover neural responses to predictable versus unpredictable auditory distractors, as well as their dependence on perceptual and cognitive load. Neurally, unpredictable distractors induced a sign-reversed lateralization of pre-stimulus alpha oscillations (~10 Hz) and larger amplitude of the stimulus-evoked P2 event-related potential component. Under low versus high memory load, distractor predictability increased the magnitude of the frontal negativity component. Behaviourally, predictable distractors under low task demands (i.e., good signal-to-noise ratio and low memory load) made participants adopt a less biased response strategy. We conclude that predictable distractors decrease uncertainty and reduce the need for updating the internal predictive model. In turn, unpredictable distractors mislead proactive spatial attention orientation, elicit larger neural responses, and put higher demand on memory.