As a result of digitalization, demand for newspapers has become more volatile and difficult to predict. The focus of this research is to determine the best forecasting method with the smallest error value to reduce the company's rate of newspaper returns and losses. The holt-winter and long-short term memory approaches are used to analyze the data. In addition, the results of this research's forecasts were compared to the previous research that used the ARIMA method and trend line analysis method. The forecasting method will be chosen based on the smallest Mean Absolute Percentage Error (MAPE) criteria. According to this research and the previous research, the trend line analysis method has a MAPE value of 2,94%, the ARIMA method has a MAPE value of 3,52%, the long short -term memory has a MAPE value of 3,87%, and the holt-winter has the smallest error rate, with a MAPE value of 2%. In conclusion, the best forecasting method for forecasting newspaper demand at PT. XYZ is the holt-winter method.