This paper introduces a new online method for performing control-based continuation (CBC), speeding up the model-less identification of stable and unstable periodic orbits of nonlinear mechanical systems. The main building block of the algorithm is adaptive filtering which can ensure the non-invasiveness of the controller without the need for offline corrective iterations. Two different strategies, termed stepped and swept CBC, are then developed for performing the continuation steps. A beam featuring different artificial stiffness and damping nonlinearities is considered for the experimental demonstration of the proposed developments. The performance of the CBC strategies are compared in terms of running time and identification accuracy.