In swarm robotics, it is crucial for swarm members to reach a consensus on a single option from a set of alternatives to complete complex tasks autonomously. Typically, individual mechanisms underpinning such collective behaviour are designed using either hand-coded or automatic approaches. In this paper, we aim to compare the performance of robotic swarms controlled by mechanisms designed using both types of techniques in a site-selection task. The evaluated hand-coded mechanisms are based on the voter model and majority rule, while the automatic design approach involves an evolved dynamic neural network mechanism. The evaluation is conducted in a simulated environment that represents different operating conditions and swarm sizes and follows the same protocol for all decision-making mechanisms. The results reveal that the evolved neural network controller demonstrates better behavioural responses, including more accurate decision-making and increased resilience to varying environments and group sizes, compared to traditional hand-coded approaches.