We assume that the likelihood of the feature values, x, given an observation is of class y, follows a normal distribution. Due to this assumption, Gaussian Naive Bayes is best used when all our features are continuous.
More about Prior Probabilities → Naive Bayes