Understanding how humans make decisions is a central theme in the behavioural sciences, with several interdisciplinary ramifications. Recent studies in the fields of psychology and neuroscience have shed light on the neural and computational mechanisms via which humans and other animals make sensorimotor decisions in simple laboratory tasks. However, little is known about the mechanisms underlying more complex multialternative, multiattribute decisions. For instance, what are the mechanisms that the brain employs when we try to choose between a small flat with a short commute and a large house with a long commute? Existing models of sensorimotor decisions do not readily apply to multiattribute decisions. This is because human choices in the latter often exhibit patterns that defy rational explanations and challenge the models designed for simpler decisions. For instance, an initial preference for a small city flat can shift to favour a large suburban house once a medium-sized house in the outskirts (i.e., inferior and thus irrelevant) becomes available. What distinguishes simple perceptual decisions—like determining motion direction in a cloud of moving dots—from more complex choices involving multiple alternatives and attributes? In this project, we suggest that in complex scenarios, the brain cannot process all relevant information in parallel. Instead, it serially samples subsets of information as the decision progresses. To understand the mechanisms behind complex decisions and why preferences can irrationally reverse, we posit that we must explore how people sample information during these complex choices. We developed novel approaches towards uncovering the mechanisms of information sampling during complex decisions. We used magnetoencephalography (MEG) recordings of cortical population dynamics of human participants who performed novel multialternative, multialternative choice tasks. Using neural decoding techniques, we continuously traced the locus and strength of attention and uncovered patterns of information sampling that conventional techniques (such as eye-tracking) cannot capture. This approach provided a new window into the natural interplay between selective attention and decision-making. Our goal is to develop a neurophysiologically detailed theory of multiattribute choice equipped with computational mechanisms that dynamically guide attention towards different aspects of a choice problem. Achieving this goal could significantly impact applied behavioural science by informing the design of more precise interventions and the development of consumer protection tools. Additionally, the emerging framework could have clinical implications by offering insights into how information sampling changes in neuropsychiatric disorders. Our work clarified how information sampling adjusts to complex decisions. We revealed consistent sampling patterns in 3-alternative choices, that first focus on the worst alternative and subsequently us it as an anchor (or reference point) to infer the desirability of the two remaining high-valued alternatives. This “anchoring” effectively leads to irrational preference reversals. At the neural level, we found that information sampling fluctuates rhythmically at 11 HZ and uncovered neural mechanisms that distinguish between focused processing of a single alternative vs. comparing across alternatives. Pharmacologically boosting cortical GABA-A slowed-down the 11 HZ rhythmicity and enhanced the elimination of the worst alternative, and at the behavioural level leads to weaker contextual preference reversals.