Do AI Assistants Promote Greener Recommendations? Agent-Language Fit for Green Consumption Intention database
Description
This dataset contains anonymized online survey data from three between-subject scenario experiments built on the Theory of Planned Behavior, designed to test a set of core research hypotheses regarding AI autonomous environmental benefits and consumer green purchase intention. H1 proposed that AI-driven autonomous environmental benefits exert a double-edged effect: they positively boost green purchase intention indirectly through environmental self-efficacy, while generating a direct negative inhibitory impact on purchase willingness. H2 hypothesized environmental self-efficacy acts as a significant mediator between AI environmental benefits and green purchase intention. H3 predicted consumer innovativeness moderates the overall mediated path, with stronger indirect effects among highly innovative consumers. H4 stated time orientation serves as another boundary factor, such that future-oriented individuals show amplified positive indirect effects compared to present-oriented participants. All three experiments adopted different green product stimuli to enhance result robustness, collecting valid samples of 241, 360 and 362 Chinese adult consumers respectively via Credamo online platform in late 2025. Uniform data cleaning rules removed careless responses with overly short filling time or repeated identical scale selections. The datasets store raw 7-point Likert scale scores, averaged construct values, experimental group dummy codes, manipulation check results and demographic controls (gender, age). Statistical analyses including t-tests, ANOVA and bootstrap moderated mediation tests were conducted via SPSS 31.0 and Hayes’ PROCESS Macro 4.1. Key significant findings fully supported all hypotheses: AI autonomous environmental benefits exhibited the predicted dual positive indirect and negative direct effects; environmental self-efficacy carried a significant mediating role; consumer innovativeness and time orientation each significantly moderated the indirect pathway. All measurement scales achieved satisfactory reliability and convergent validity. Other researchers can replicate all original empirical results by adopting identical experimental scenarios, measurement scales and data screening standards, and conduct secondary analysis on AI sustainable consumption, consumer heterogeneity and green marketing mechanisms. All data are fully anonymized with no identifiable personal information, and all participants provided voluntary informed consent before survey participation.
Files
Steps to reproduce
All data used in this mixed-method research (three online between-subjects scenario experiments plus one offline field experiment) were gathered from adult Chinese consumers via the Credamo professional online survey platform and community physical stores. Four distinct green product stimuli were adopted: degradable cleaning wipes (Study 1), plant-based milk (Study 2), solar desk lamps (Study 3), and organic vegetable gift boxes (field study). Each experiment adopted a 2×2 factorial design manipulating two independent variables: recommendation agent (AI assistant vs. human salesperson) and linguistic style (explicit data-driven vs. implicit emotional narrative). Pre-tests were implemented to refine scenario texts, visual materials and 7-point Likert measurement scales sourced from published peer-reviewed literature, covering manipulation check items, green consumption intention, cognitive trust, affective trust, need for cognition and demographic covariates (gender, age, past green shopping experience). Uniform data cleaning standards were applied across all datasets: questionnaires with extremely short response time or identical answers for five consecutive items were discarded, yielding valid samples of 256, 360, 400 and 267 respectively. All participants submitted voluntary informed consent, and all raw data were fully anonymized without any identifiable personal information. To fully replicate this research, researchers can adopt identical randomized between-subject assignment logic, matched shopping scenarios, standardized measurement scales and the same sample exclusion criteria on any mainstream online survey tool or offline on-site questionnaire channel. Statistical analyses including independent t-tests, two/three-way ANOVA, reliability and validity tests, as well as bootstrap-based parallel and moderated mediation analyses were executed with IBM SPSS 31.0 and Hayes’ PROCESS Macro 4.1; no specialized laboratory instruments or chemical reagents were required for the whole research workflow, and all analytical codes and variable definition rules are attached in the supporting code book file to facilitate full reproduction of all empirical results, interaction graphs and significance outputs.
Institutions
- Yanshan UniversityHebei, Qinhuangdao