The size of the metasurface unit cell increases with the decrease of its center working frequency (CF). This is not conducive to the integrated design of the metasurface. To address the problem, a miniaturization design method based on genetic algorithm (GA) is proposed. Firstly, an element library is established. By randomly selecting and combining these elements, various structures are generated as the initial population of GA. Secondly, a Python-CST joint simulation program is constructed to automatically model the metasurface unit cells and extract their electromagnetic parameters. Finally, after multiple iterations of optimization, the optimized unit cell is obtained. This paper takes the transmissive-type linear-to-circular polarization conversion metasurface (LTCPCM) as a research object. The simulation results show that the size of the designed unit cell is only 0.18 λ of the CF (3.86 GHz). The LTCPCM exhibits broadband and wide-angle characteristics, with a fractional bandwidth (FB) of 71% (incident angle θ = 0°) and an angular stability (AS) of 45°. The experimental validation and the simulation result are consistent, proving the effectiveness of the proposed method. The research results can provide a reference for further miniaturization design of metasurface unit cells.