Air conditioning load, as an important demand response resource, is difficult for dispatch centers to directly aggregate and control due to differences in type, dispersed access, and random user preferences, thereby limiting its potential. Addressing the issue that existing grouping methods only consider air conditioning thermal parameters while ignoring differences in user temperature regulation preferences, this paper proposes a two-layer clustering grouping method and a response potential assessment method based on a cloud model. First, a cloud model is used to generate users’ response willingness to different temperature adjustment instructions. Then, two-layer clustering grouping is performed using temperature change index, characteristic temperature difference, and user response degree as features, and the potential of each group is assessed. To address the issue of significant load drops during the initial phase of temperature adjustment, a preparation time is introduced for group control. The response aggregation is optimized with the objective of minimizing power deviation during the target time period, balancing regulatory potential differences and user thermal comfort. Case studies validated that this method can achieve orderly control of air conditioning clusters, reduce load fluctuations, and provide support for system operation.