The Ethereum blockchain is a distributed database of transactions, where the Gas Oracles suggest the users the Gas price's categories to get a transaction recorded. The paper explores the idea that the Gas Oracles are based on a data-centered model which does not provide users with a reliable prediction. We present an empirical study to test the reliability of the existing Gas Oracles from both the points of view of the Gas price predictions and the existing categories. The study reveals that the Gas Oracles' predictions fail more often than advertised and shows that the Gas price categories do not correspond to the categories set by the users. Therefore we propose a user-oriented model for the Oracles' Gas price prediction, based on two Gas price categories actually corresponding to the users' interests and a new method to estimate the Gas price. The new method, performing the Poisson regression at smaller intervals of time, predicts the Gas price to pay with a lower margin of error when compared to the actual one. The predictions based on the user-oriented model thus provide the users with a more effective Gas price to set.