Potential of large language model-powered nudges for promoting daily water and energy conservation

TL;DR

This study investigates whether large language model (LLM)-powered nudges can more effectively promote daily water and energy conservation intentions than traditional usage-based nudges. It uses a campus-based randomized experiment (n=1,416 valid) to compare three conditions: no nudges, traditional nudges, and LLM-powered nudges that add personalized conservation tips. The findings show LLM nudges increase conservation intentions more than traditional nudges, with average gains of $8.3 ext%$ and individual gains up to $18.0 ext%$, and they operate by strengthening self-efficacy and outcome expectations while reducing reliance on social norms. Structural equation modeling suggests LLM nudges shift the decision-making process toward intrinsic motivation, and multi-arm causal forests reveal heterogeneous effects across demographics and behaviors, highlighting the potential for scalable, targeted sustainability interventions with significant resource implications.

Abstract

The increasing amount of pressure related to water and energy shortages has increased the urgency of cultivating individual conservation behaviors. While the concept of nudging, i.e., providing usage-based feedback, has shown promise in encouraging conservation behaviors, its efficacy is often constrained by the lack of targeted and actionable content. This study investigates the impact of the use of large language models (LLMs) to provide tailored conservation suggestions for conservation intentions and their rationale. Through a survey experiment with 1,515 university participants, we compare three virtual nudging scenarios: no nudging, traditional nudging with usage statistics, and LLM-powered nudging with usage statistics and personalized conservation suggestions. The results of statistical analyses and causal forest modeling reveal that nudging led to an increase in conservation intentions among 86.9%-98.0% of the participants. LLM-powered nudging achieved a maximum increase of 18.0% in conservation intentions, surpassing traditional nudging by 88.6%. Furthermore, structural equation modeling results reveal that exposure to LLM-powered nudges enhances self-efficacy and outcome expectations while diminishing dependence on social norms, thereby increasing intrinsic motivation to conserve. These findings highlight the transformative potential of LLMs in promoting individual water and energy conservation, representing a new frontier in the design of sustainable behavioral interventions and resource management.

Figures (5)

• Figure 1: Roadmap of this study.
• Figure 2: Descriptive statistics of the questionnaire survey.a. Historical behaviors related to water and energy conservation, measured using a 5-point Likert scale. The upper axis shows the average level of these behaviors, whereas the lower axis displays the percentage of respondents selecting each option. b. Cognition factors related to water and energy conservation, with error bars representing the 95% confidence intervals. c. Attitudes toward nudges, with error bars representing the 95% confidence interval. d. The rated effectiveness of different nudge contents.
• Figure 3: Increases in conservation intentions caused by nudges and their heterogeneity. a. Average increases according to the statistical comparison among the three groups, with error bars representing the 95% confidence intervals. b. Individual-level conservation intention changes and their distributions calculated by the multi-arm causal forest. The error bars represent the 95% confidence intervals. c. Ranges of conservation intentions increase in different demographic groups. d. Ranges of conservation intentions increase in student groups with different daily energy behaviors. e. Ranges of conservation intentions increase in student groups with different living metrics, with error bars representing the 95% confidence interval.
• Figure 4: Decision-making related to conservation intentions. a. The full model of the decision-making of conservation intentions under nudges. b. The decision-making of conservation intentions among samples with no nudge (C). c. The decision-making of conservation intentions among samples with traditional nudges (T1). d. Decision-making of conservation intentions among samples with LLM-powered nudges (T2).
• Figure 5: Measurements and experimental design. a. Groupings and included demonstrations of nudge content used to measure conservation intentions. b. An example of a demonstration of a nudge. c. Measurements used in the questionnaire survey.