Atendency to extract spurious, message-like meaning from meaningless noise was assessed as a risk factor leading to schizophrenia-spectrum disorders by assessing word length of speech illusions elicited by multispeaker babble in 43 people with prodromal symptoms. These individuals were randomised to olanzapine v. placebo groups during year 1 followed by no pharmacological treatment for those with no disorder conversion during year 2. A time-dependent Cox regression analysis of conversion to schizophrenia-spectrum disorder revealed a significant interaction between condition (olanzapine v. no drug) and length of speech illusion, with the latter strongly predicting subsequent conversion during medication-free intervals but not during olanzapine treatment.