For wavefront sensorless adaptive optics (AO) systems, the efficiency of close-loop control methods is of greate importance. In this paper, a gradient descent algorithm with Hadamard model is proposed. The working principle of AO close-loop based on the algorithm is introduced. The working process and feasibility of this method are analyzed by comparing with the serial gradient descent algorithm and stochastic parallel gradient descent algorithm. The numerical simulation model with this method is setup, and the close-loop corrections with random generated aberrations are performed to verify its precision and convergence speed. The results show the gradient descent algorithm with Hadamard model can be used in wavefrontless adaptive optics, and compared with the stochastic parallel gradient descent algorithm, it is not only easier to be implemented but also more efficient.