The advice gap: adaptive yet suboptimal advice integration

Effective use of advice is critical for sound decision-making, yet people often fail to fully benefit from external input. The extent and causes of this suboptimality, however, remain poorly understood. Combining a novel Bayesian modeling approach with experimental data, we show that individuals adjust their reliance on advice based on the cognitive and metacognitive profiles of artificial advisers, and that advice generally improves performance. Nevertheless, decision-makers display systematic biases: they overweight self-generated judgments and confidence, and misapply knowledge of metacognitive skills by inconsistently translating global traits into local decisions. Strikingly, these biases persist even when both self and adviser attributes are fully transparent. Our findings reveal fundamental bottlenecks in advice integration, not stemming from ignorance, but from biased information processing. By identifying these limits, this work highlights concrete pathways for enhancing collaboration in human–human and human–AI decision-making.