We present an adaptive method that enables a General Game Player using Monte-Carlo Tree Search to adapt its use of RAVE to the game it plays. This adaptation is done with a comparison of the UCT and RAVE prediction for moves, that are based on previous playout results. We show that it leads to results that are equivalent to those obtained with a hand tuned choice of RAVE usage and better than a fit-for-all fixed choice on simple ad hoc synthetic games. This is well adapted to the domain of General Game Playing where the player cannot be tuned for the characteristics of the game it will play before the beginning of a match.