Apparent neural encoding of future words may arise from the statistical structure of language itself, rather than from predictive computations in the brain.