Can AI agents pre-test sustainability nudges? A validated virtual experiment in hotels

Abstract

Hotels routinely use sustainability nudges, but pre-testing their effectiveness requires costly field experiments. We propose a virtual experimentation framework in which AI agents respond to hospitality scenarios and their outputs are calibrated against intercept survey data from four hotels in Gran Canaria (N=1680). Simulating two nudges–towel reuse and buffet plate waste (a binary measure, included for cross-intervention generalizability)–under three conditioning strategies, we find that agents given episodic, affect-rich context produce estimates closest to published field benchmarks: the calibrated towel-reuse lift is 5.5 percentage points (field benchmark: 6.8 pp). Calibration preserves strategy rankings and reduces estimation error, consistent with situated-cognition perspectives, though not direct evidence of psychological mechanisms. The framework offers a low-cost screening tool for tourism sustainability interventions.