Empirical best predictors are studied under area-level Poisson mixed models with time effects. Four cases are considered. The first two cases use independent time random effects. In the second two cases, the time effects follow an autoregressive process of order one. The four models are fitted by the moment-based method and the corresponding empirical best predictors are derived and compared with plug-in predictors. Several simulation experiments investigate the performance of both predictors. A parametric bootstrap procedure is considered for estimating the mean squared error. The developed methodology is applied to estimate the proportion of people under the poverty line by counties and sex in Galicia (a region in north-west of Spain).