Comparing the influence of visual information and the perceived intelligence of voice assistants when shopping for sustainable clothing online

The study addresses how multisensory cues in online retail—specifically visual information about sustainable clothing and the perceived intelligence of voice assistants—shape consumers’ purchase behaviour for sustainable apparel. Against the backdrop of climate change and growing interest in sustainable consumption, prior work has linked eco-friendly product attributes, consumer knowledge, attitudes, and social influences to sustainable purchase intentions, while also noting barriers such as perceived aesthetics/functionality and attitude–behaviour gaps. With increased online shopping (especially during COVID-19) and advances in AI-driven interfaces, the authors propose that multisensory perception via on-site visuals and voice assistants affects sustainable clothing purchase behaviour, mediated by positive attitudes toward sustainable clothing and moderated by consumers’ knowledge of sustainability issues. The study aims to inform multisensory marketing and consumer behaviour theory and practice in online sustainable fashion retailing.

The Related work and hypotheses synthesize research on AI-enabled multisensory marketing and consumer behaviour. AI can deliver personalised, multisensory services (visual and voice) that influence emotions, attitudes, and behaviour. However, incongruent or unnatural voice interactions can reduce trust and willingness to buy. Attitudes toward sustainable clothing are strong predictors of purchase intention and behaviour, though attitude–behaviour gaps can arise due to price and other barriers. Visual information online (e.g., product images, labels, design cues) strongly shapes attention, mental imagery, attitudes, and decisions. Verified sustainability labels and lifecycle information can enhance evaluation and behaviour. Voice assistants may improve information access, interactivity, and perceived usefulness, thereby fostering positive attitudes and behavioural intentions, but concerns about credibility and privacy may hinder acceptance. Knowledge about sustainability correlates with higher intention and behaviour, moderating how consumers translate attitudes into actions. Hypotheses: - H1: Positive attitude toward sustainable clothing positively impacts purchase behaviour (PB). - H2: Visual information (VI) positively impacts positive attitude toward sustainable clothing (ATSC). - H3: VI positively impacts PB. - H4: Perceived intelligence of voice assistants (PIVA) positively impacts ATSC. - H5: PIVA positively impacts PB. - H6: The effect of VI on PB through ATSC is moderated by knowledge of sustainability issues (KSI). - H7: The effect of PIVA on PB through ATSC is moderated by KSI.

Design and data collection: An online survey was administered via the Sojump (Questionnaire Star) platform, with data collected in Shanghai, PR China. A total of 2656 valid responses were obtained. Demographics included 53.7% female and 46.3% male; most participants were aged 18–35; and a majority had associate or bachelor’s degrees. Monthly clothing expenditure varied across predefined ranges. Measures: All items used five-point Likert scales (1 = strongly disagree to 5 = strongly agree) and were adapted from prior validated sources. Constructs included: - Visual information (VI; 3 items: materials/fibres, longevity/robustness, sustainability labels) - Perceived intelligence of voice assistants (PIVA; 5 items: competence, knowledge, relevance, intelligence, accuracy) - Knowledge about sustainability issues (KSI; 4 items) - Positive attitude toward sustainable clothing (ATSC; 4 items) - Purchase behaviour (PB; 3 items) Analysis: Reliability and validity were assessed using SPSS 23.0 and AMOS 23.0. Cronbach’s alpha (>0.60 threshold, all >0.8), composite reliability (CR >0.70), average variance extracted (AVE >0.50), and KMO (>0.70) indicated adequate reliability/validity. CFA showed good model fit. Multiple regression analyses (VIF <5) tested direct effects. Mediation was tested using PROCESS (Hayes, 2013), and moderation by KSI was examined via conditional indirect effects, with confidence intervals not overlapping zero indicating significance.

Measurement model: All Cronbach’s alpha values exceeded 0.8; factor loadings >0.5. CFA fit indices indicated good fit: χ²/df = 2.914, GFI = 0.992, AGFI = 0.983, CFI = 0.995, IFI = 0.995, NFI = 0.993, RMSEA = 0.027. AVE values exceeded 0.5 and CR values exceeded 0.8 across constructs. Direct and attitudinal effects (multiple regression): - H1 supported: ATSC → PB (β = 0.230, t = 12.197, p < 0.001; model F = 457.234, R² = 0.340). - H2 supported: VI → ATSC (β = 0.334, t = 17.234, p < 0.001; model F = 574.254, R² = 0.302). - H3 supported: VI → PB (β = 0.295, t = 14.862, p < 0.001). - H4 supported: PIVA → ATSC (β = 0.291, t = 15.005, p < 0.001). - H5 supported: PIVA → PB (β = 0.185, t = 9.413, p < 0.001). Additional direct effects (PROCESS outputs): VI → PB (β = 0.510, t = 30.536, p < 0.001); PIVA → PB (β = 0.453, t = 26.210, p < 0.001); ATSC → PB (β = 0.462, t = 26.827, p < 0.001). Mediation (ATSC): - VI → ATSC → PB: significant indirect effect (t = 15.126, p < 0.001); total effect 0.510; direct effect 0.373; indirect effect 0.137 (BootLLCI = 0.114, BootULCI = 0.160). - PIVA → ATSC → PB: significant indirect effect (t = 17.159, p < 0.001); total effect 0.453; direct effect 0.303; indirect effect 0.150 (BootLLCI = 0.127, BootULCI = 0.176). Moderation (KSI) of indirect paths: - H6 supported: KSI moderates VI → ATSC → PB; conditional indirect effects stronger at higher KSI (e.g., +1 SD: LLCI = 0.167, ULCI = 0.278; −1 SD: LLCI = 0.267, ULCI = 0.373). Interaction tests and index show significance (e.g., LLCI = −0.087, ULCI = −0.018, p < 0.001). - H7 supported: KSI moderates PIVA → ATSC → PB; conditional indirect effects stronger at higher KSI (+1 SD: LLCI = 0.180, ULCI = 0.294; −1 SD: LLCI = 0.309, ULCI = 0.416), with significant moderation indices (e.g., LLCI = −0.103, ULCI = −0.033, p < 0.001). Overall, all seven hypotheses (H1–H7) were supported. Visual information and perceived intelligence of voice assistants significantly enhance positive attitudes and purchase behaviour toward sustainable clothing; attitudes mediate these relationships, and sustainability knowledge strengthens the indirect effects.

Findings demonstrate that both visual product information and the perceived intelligence of voice assistants are effective multisensory cues that shape positive attitudes and purchase behaviour for sustainable clothing in online contexts. Positive attitudes are a key mechanism translating sensory perception into behaviour, aligning with reasoned action/planned behaviour frameworks. The moderation by sustainability knowledge indicates that better-informed consumers more effectively convert positive attitudes, induced by visual and voice cues, into actual purchase behaviours. Managerially, retailers should invest in high-quality sustainability visuals (e.g., clear labels, materials, lifecycle cues) and deploy capable, human-like voice assistants that provide accurate, personalised, and timely information to build trust and facilitate decisions. Enhancing databases with rich sustainability content and tailoring voice interactions may reduce information gaps, reinforce perceptions of competence, and encourage sustainable purchases. The results extend multisensory marketing theory to AI-enabled retail, evidencing how crossmodal digital cues and consumer knowledge interact to influence behaviour.

The study introduces and empirically validates a conceptual model linking visual information and the perceived intelligence of voice assistants to sustainable clothing purchase behaviour, mediated by positive attitudes and moderated by sustainability knowledge. Using a large online sample, the results show robust direct, mediated, and moderated effects, offering theoretical contributions to multisensory marketing and AI-enabled consumer decision-making, and practical guidance for online fashion retailers to enhance sustainable purchase outcomes via improved visuals and intelligent voice support. Future research should test generalisability across cultures, incorporate additional sensory modalities in digital retail (e.g., haptic/tactile proxies), and examine demographic factors such as gender and age to refine understanding of heterogeneous effects.

Generalizability is limited by conducting the study in China; findings may differ in other cultural contexts. The sensory scope focused on visual information and voice assistant intelligence, omitting other modalities relevant to online shopping. Demographic analyses (e.g., gender, older age groups) were not emphasised, and the sample skewed younger. Future work should perform cross-cultural comparisons, include broader multisensory variables, and investigate demographic moderators.