Study reveals AI efficiency-gain illusion: users overestimate savings, underestimate usage
People waste time on AI for simple tasks—and don't realize it.
In a paper posted to arXiv (26 May 2026), researchers Sunny Yu and colleagues from Stanford and other institutions present the "efficiency-gain illusion." Across three pre-registered user studies with 2,691 participants, they found that people choose to use AI assistance for simple tasks (e.g., arithmetic, spell-check, answering basic questions) even when doing so provides no measurable time or effort savings. The team identified two key miscalibrations: a self-estimate bias where users believe they use AI less than they actually do, and an efficiency-gain illusion where users overestimate both time and effort saved by AI.
Additionally, the study revealed a session-level carryover effect: a participant's prior AI use increases the likelihood of further AI adoption and deepens miscalibration about time savings. This sets up a potential overreliance feedback loop, where people increasingly lean on AI for tasks where it offers minimal benefit. The findings shed light on mechanisms behind AI usage decisions and highlight risks for productivity, learning, and cognitive skill maintenance in an AI-augmented workplace.
- 2,691 participants across 3 pre-registered studies consistently chose AI for simple tasks despite no meaningful time savings.
- Users underestimated their own AI usage and overestimated time saved by 20% or more.
- Carryover effect: prior AI use led to 15% higher adoption of AI on subsequent simple tasks and entrenched miscalibration.
Why It Matters
The illusion drives overreliance on AI for trivial work, risking skill decay and productivity loss.