The human hypothalamus encodes prediction errors during sensory-nutrient inference

Humans infer the nutrient content of foods from sensory cues, yet the neural and computational basis of these inferences remains unclear. Here we show how oral texture contributes to judgements of fat content by combining purpose-designed foods that dissociated fat from texture with behavioral testing, computational modelling, functional neuroimaging during consumption, and a naturalistic feeding assay. Judgements of fat content reflected established relationships between oral texture and fat and became systematically less accurate when these relationships were violated. A computational model formalized these failures as prediction errors—discrepancies between fat content predicted from texture and actual fat content. During food consumption, the hypothalamus encoded these model-derived prediction errors, responding most strongly when texture predicted more fat than the food contained. Individual differences in hypothalamic sensitivity to texture predicted fat intake during an ad libitum meal. These findings show how sensory inference of fat content engages hypothalamic signals predicting food intake.