The use of Big-data, and more specifically of Google Trend data, in now- and forecasting, has become common practice, even by Institutes and Organizations in charge of producing official statistics around the world. However, such data will have many implications in the model estimation, which can roughly impact final results. In this paper, starting from a MIDAS-AR model with Google Trend covariate, we are focussing on the main issues concerning the sampling error and the time domain context.