Language arrives at a rate of 150-300 words per minute: a rapid input stream that needs to be decoded at high speed. There is now an emerging consensus that the input is to some extent predictable, and that the brain exploits this predictability by thinking ahead and generating predictions. I use EEG, eye-tracking, and behavioural experimentation to address questions about the contents of predictions, the underlying mechanisms, and their consequences. One line of research, funded by an NWO Veni grant, investigates whether prediction influences what language users ultimately retain in their memory. In the long run, this work could reveal the best way to speak and write if you want information to be remembered, with implications for learning and education.