Abstract Background and Hypothesis Clinical observations of psychotic experiences reveal a predominance of social themes, raising an important question about whether the underlying information-processing alterations show an analogous domain-specificity particularly impacting social inferences, including beliefs about others’ intentions. Study Design We examined how learning and decision-making in relation to an external cue was influenced by whether its source was presented as social or non-social, in healthy controls (HCs) compared to people with first episode psychosis (FEP) or with an at-risk mental state (ARMS). Using computational modelling, we quantified how responses were biased by the external cue, and were differently sensitive to its utility, depending on its being perceived as social versus non-social in origin. Study Results Overall, tendency towards choosing the externally-cued option was lower in HCs than both clinical groups, and higher in FEP than ARMS. In HCs, there was a stronger bias toward the external cue when it was presented as social in origin: by contrast, the bias was unaffected in the ARMS group, and the FEP group conversely showed relatively greater bias towards the non-social cue. The effect of the cue’s perceived origin on sensitivity to its objective value likewise differed between groups: despite being more biased towards the social cue, HCs were relatively less able to use social information to optimise their choices; whereas sensitivity to the cue’s current value was unaffected by social/non-social origin in ARMS and FEP. Conclusions Our findings contradict ideas that psychosis is associated with a simple and specific deficit in social processing.