LLM‑based GUI agents are increasingly acting on behalf of users in digital environments originally designed for human interaction. These graphical interfaces both support users and deliberately steer their behavior and decisions. While behavioral biases in LLM textual outputs are well documented, far less is known about how such influence operates when models act as agents that perceive interfaces and execute actions, especially whether the growing reasoning capabilities make them more robust.
Drawing on Dual‑Process Theory—System 1 (automatic) and System 2 (reflective)—we empirically examine LLM‑based GUI agents' susceptibility to two types of digital nudges: automatic default nudges and reflective social‑influence nudges. We also test how the agents' reasoning configuration moderates this susceptibility. In a randomized online shopping experiment we deployed 3,600 agents, running a total of 21,600 simulations across six frontier models from three providers.
We found that agents are vulnerable to both nudge types. Crucially, the reasoning configuration moderated the effects in opposite directions: extensive reasoning reduced susceptibility to automatic default nudges but heightened it to reflective social‑influence nudges. Thus, more reasoning did not make agents more robust; it redirected the pathway through which choice architecture exerts influence. Exploratory analysis further revealed that this redirection is systematically structured by model scale.
The study establishes nudge susceptibility as a behavioral property of agentic AI and positions interface design as a governance concern for organizations delegating decisions to autonomous agents.
Review