It depends on the trial design. For a simple RCT like a two independent groups design it could be fine. GPower is pretty flexible. However many RCTs are more complex than this with multiple time points and multiple sites. If so the design is likely to involve a multilevel model or similar and that isn't straightforward to handle without a more sophisticated approach. Usually that means simulation. Even for many simpler designs a simulation approach may be useful in understanding the sensitivity of the analysis to different assumptions about the parameters.

Thank you so much for your answer, Prof Baguley. I want to conduct a three-arm RCT and am wondering if there is any online tool for the sample size calculation. Thom Baguley

Standard practice seems to be to power the study to detect specific between group comparisons on the primary outcome or outcomes. If there is no clustering (i.e., it isn't multi-site etc.) this reduces to one or more t tests or ANCOVA if you have baseline primary outcome measures as covariate (which. typically increases power). So you could in principle treat this as one or more t tests (in the absence of clustering) and use something like GPower. If the outcome is dichotomous you might use a Chi-square test or similar.