Most of it are the models I use, and Is not an exact cience but usually, if you put the hands doing something they are more clear than if not. If you put them "waving" or "holding something" "pointing" or even simply "hands up", "V" "covering crotch" "finger to mouth" "covering mouth" it usually helps because the model will go and check references for hands for those things (I suppose).

Then again, not an science, they are fucked up lots of times, but as I said if you ask in the prompt something that involves the use of the hands even if its not necessary for what you are looking for they trend to be less messed up.

And then again, the models are usually use are pretty good and stable