Stanford study: AI speedup is an illusion – users overestimate time savings

A large-scale preregistered study (N=1237) by Yu et al. from Stanford, Princeton, and other universities investigated how people perceive time savings from AI assistance on simple cognitive tasks. Participants completed tasks either independently or with AI help, and also predicted completion times for both modes. Surprisingly, actual completion times did not differ between independent and AI-assisted conditions. However, participants consistently predicted that AI would be significantly faster than independent work, even though they accurately forecasted independent times. This mismatch — termed the 'speedup illusion' — was specific to AI: when imagining help from another human, the bias disappeared.

Beyond time perception, the study found a dissociation between time and effort. Participants reported significantly lower subjective effort when using AI, despite taking the same amount of time to finish tasks. This suggests that the perceived efficiency gains from AI may stem from reduced cognitive load rather than actual speed improvements. For professionals, the findings caution against assuming AI always accelerates workflows – the real value may lie in making tasks feel easier, not faster. Productiviy metrics should consider both time and cognitive effort to capture true gains from AI integration.