To label, or not to label: how culture shapes autonomy and consumer responses to AI-labelled appliances

Recent discussions at CES, the world’s largest home-appliances trade show, highlighted a paradox: “Everything was AI, even when it wasn’t” (David, 2024). AI now operates so seamlessly in appliances and smart home systems that its presence is barely noticeable, yet many brands prominently label their products as “AI”. For example, LG and Samsung routinely embed “AI” into product names (e.g. LG’s “AI DD Washing Machine”, “Dual Inverter AI Air Conditioner” and “QuadWash AI Dishwasher”; Samsung’s “AI EcoBubble Washing Machine”, “Wind-Free AI Air Conditioner” and “Bespoke AI Oven”). Chinese brands like Midea’s “COLMO”, Haier, Hisense and TCL market themselves as AI appliance brands. Yet, other brands choose to incorporate similar AI capabilities without emphasising them (e.g. Electrolux UltimateCare washing machine, Bosch VitaFresh Refrigerator, Mitsubishi Electric MSZ-GL air conditioner and Miele’s G 7000 Dishwasher). The question arises: does the choice to label a home appliance as “AI” change consumers’ perception of this appliance, and if so, how? This question is particularly important, given that the smart home appliances market, which includes AI-enabled devices, is projected to reach approximately $250.6 billion by 2029 (Statista, 2024).

The existing literature provides a mixed bag of evidence regarding the potential effects of AI labels on consumer responses. On the one hand, AI labels may signal technological advancement, accuracy and sophistication (Castelo et al., 2019; Dietvorst et al., 2015; Logg et al., 2019) – attributes that are valued in functional product categories such as appliances (Henard and Szymanski, 2001; Wertenbroch and Dhar, 2000). On the other hand, AI labels may trigger psychological reactance by implying a loss of personal control or a shift in decision-making authority from the user to the technology (Decker et al., 2017; Porter and Heppelmann, 2014). Because autonomy is central to consumer satisfaction (Deci and Ryan, 1985; Ryan and Deci, 2000), AI labels may backfire when they suggest diminished autonomy.

We propose that responses to AI-labelled appliances are culture-dependent. Specifically, we propose that consumers from Western cultures, characterised by an independent self-construal that emphasises independence and autonomy, respond less favourably to AI-labelled appliances than those from Eastern cultures, who place less emphasis on individual agency. We further argue that this cultural difference is driven by varying perceptions of autonomy loss.

Four experimental studies and one secondary-data study test the hypotheses. Studies 1a and 1b show that participants from individualistic cultures (US and UK) evaluate AI-labelled appliances less favourably than identical non-AI-labelled alternatives. Study 2 demonstrates that the effect of AI-labelling is mediated by perceptions of constrained autonomy. Study 3 provides further evidence for the mediating role of autonomy while also showing that a collectivistic culture mitigates resistance to AI-labelling. Study 4 reveals that reframing AI as consumer autonomy-enhancing improves consumer perceptions. Lastly, Study 5 validates the experimental results with real-world data.

The present research contributes to three streams of literature. Firstly, it extends work on consumer responses to AI (Castelo et al., 2019; Chugunova and Sele, 2022; Puntoni et al., 2021) by showing that AI-labelling can be detrimental, even in functional domains where AI should, in principle, be advantageous. Secondly, our research contributes to the literature on autonomy (Brehm and Cole, 1966; Deci and Ryan, 2000; Fitzsimons and Lehmann, 2004), highlighting autonomy as a central psychological mechanism underlying resistance to AI-labelled home appliances. Thirdly, it adds to the growing body of research on culture and consumer technology adoption (Lee et al., 2013; Ma et al., 2014; Sunny et al., 2019) by demonstrating that cultural values moderate responses to AI-labelling, with individualistic consumers showing stronger resistance.

Managerially, the findings caution against the indiscriminate use of “AI” labels on consumer products. Although AI may enhance product performance, emphasising it may undermine consumer evaluations, particularly when it signals a potential loss of user control. Marketers should highlight features that preserve or enhance autonomy (e.g. manual overrides, customisation settings) and use framing that highlights consumer agency (e.g. “You stay in charge – AI helps you decide”). Lastly, our findings suggest that the effectiveness of AI-labelling varies across cultures, as what signals innovation in one market may provoke resistance in another.

A growing body of research suggests that consumers tend to react negatively to the use of AI in products and services. The resistance to AI is particularly salient in the services context where AI is compared directly with human providers. For instance, consumers lose confidence in AI more quickly than in humans after observing equivalent mistakes (Dietvorst et al., 2015). They also exhibit compensatory responses (e.g. seeking social affiliation, ordering and eating more food) after interacting with a humanoid service robot (Mende et al., 2019). Resistance towards medical AI is notable: consumers evaluate health practitioners who rely on AI as less professional and competent (Mende et al., 2019; Palmeira and Spassova, 2015), distrust medical advice coming from AI algorithms (vs from human doctors) (Longoni et al., 2019; Önkal et al., 2009) and display lower willingness to pay for AI-based care (Longoni et al., 2019).

The complexity of this resistance is illuminated by the computers are social actors (CASA) paradigm (Moon, 2000; Nass et al., 1994), also known as social response theory, which argues that consumers instinctively treat technology as a social entity. Consumers unconsciously apply social rules and expectations to technological agents even while knowing these agents lack feelings, intentions or a sense of self (Nass and Moon, 2000; Nass et al., 1994). Human-like cues, such as interactivity or reciprocity, automatically activate social scripts typically reserved for human-to-human interaction (Song and Kim, 2022).

Negative responses to AI can be attributed to fundamentally human reactions to this technology. Consumers believe AI lacks quintessentially human abilities such as warmth, intuition or empathy (Haslam et al., 2008), and therefore perceive it as inferior in tasks seen to require such abilities (Castelo et al., 2019). AI agents, including chatbots, are often seen as less empathetic, resulting in less satisfying interactions (Luo et al., 2019). In healthcare, AI’s inability to fully consider individual symptoms or adapt flexibly to novel situations contributes to negative evaluations (Longoni et al., 2019). Moreover, when AI is seen as incompatible with one’s self-concept, particularly in domains involving creativity, care or personal expression, symbolic and identity concerns become more pronounced (Puntoni et al., 2021). That is, consumers view AI not only as a functional tool but also as a social agent that is unfit for certain tasks, or even certain roles in society.

Whereas resistance to AI is strong in subjective tasks that require “a human touch”, evidence shows that consumers trust AI more for objective, analytical or utilitarian tasks (Castelo et al., 2019; Longoni and Cian, 2022; Smith et al., 2016; Wien and Peluso, 2021). Consumers believe AI recommenders are less competent than their human counterparts for hedonic products but are equally or more competent for utilitarian products (Longoni and Cian, 2022; Wien and Peluso, 2021). Similarly, increasing a task’s objectivity increases trust in and use of algorithms for this task (Castelo et al., 2019).

Home appliances are primarily utilitarian products designed to perform practical tasks such as cleaning, cooking or cooling. Consumers typically evaluate them based on objective criteria such as durability, energy consumption, speed, etc. Since consumers trust AI more for objective tasks (Longoni and Cian, 2022), one could expect that consumers would evaluate AI-labelled home appliances more favourably than non-AI versions. However, the CASA paradigm suggests that consumers also apply social norm, such as autonomy and reciprocity, to their interactions with technology. We propose that even in functional product contexts, such as home appliances, consumers would prefer non-AI-labelled products over AI-labelled ones because AI-labelled ones are perceived as constraining their sense of autonomy. The next section outlines the motivation for this proposition.

Autonomy, defined as the ability to make choices independently of external forces (Dahl and Moreau, 2007; Deci and Ryan, 1985), is crucial in shaping consumer attitudes. According to self-determination theory (White, 2015), autonomy is one of the basic psychological needs, along with competence and relatedness, and is a core factor for self-fulfilment and psychological well-being. Previous research has shown that autonomy leads to positive affect and stronger motivation towards a task, whereas its absence can elicit psychological reactance (de Bellis et al., 2023; Deci and Ryan, 2000). In a marketing context, likewise, consumers have been shown to resist changes that diminish their sense of agency or control (Beuckels et al., 2019; Brickman, 1982; de Bellis et al., 2023; Nadler and Halabi, 2006).

There is evidence that AI is perceived as undermining consumers’ sense of personal control and agency. For example Puntoni et al. (2021) argue that AI systems, especially when presented as autonomous decision-makers, can provoke psychological discomfort by infringing on consumers’ self-determination. Their conceptual framework positions autonomy as a central axis in the consumer–AI interaction, noting that AI can be perceived as overly deterministic or depersonalised. In line with this notion, consumers resist using AI tools for tasks that reflect personal values or require discretionary judgement. This is because reliance on AI feels like surrendering personal agency, especially when the system is opaque or when consumers cannot easily override its suggestions (Castelo et al., 2019). Consumers are also more likely to reject algorithmic advice when they feel that accepting it implies a loss of moral agency (Chugunova and Sele, 2022).

In the context of home appliances, autonomy reflects the user’s ability to control and personalise an appliance’s settings and routine. When AI-enabled appliances make decisions without direct input from the consumer (e.g. automatically determining temperature and time), they can potentially threaten the consumer’s sense of autonomy (de Bellis et al., 2023). Importantly, this loss of control also changes how consumers assign causality for outcomes. Attribution theory suggests that people tend to attribute outcomes either internally (to the self) or externally (to outside forces) (Menon et al., 1999; Newmeyer and Ruth, 2020). When the causal chain between one’s actions and the outcome becomes less visible, because the appliance “decides” on the consumer’s behalf, consumers may be less able to claim credit for positive outcomes, which can dampen feelings of competence and positive affect. Consistent with this logic, feeling in control facilitates positive self-attributions, whereas reduced control shifts perceived causality away from the self.

For instance, a consumer may cook a chicken dish by choosing the temperature and time themselves or use an oven that detects the chicken’s characteristics and adjusts settings automatically for the perfect meal. While the latter (automation) may improve outcomes, the former enables consumers to control the details, thereby affording a stronger sense of autonomy. Thus, while AI-enabled appliances offer efficiency and convenience compared to standard appliances, consumers may favour them less due to the diminished opportunity for consumer control over task details. Accordingly, although AI-enabled appliances may improve objective performance, highlighting AI-driven features can backfire by diminishing consumers’ perceived autonomy and the self-relevant value they derive from being the agent of outcomes. This leads us to hypothesise the following:

The emphasis individuals place on autonomy during the consumption process may not be universal and may depend on their culture. While cultural differences may manifest along a number of dimensions, the distinction between individualism and collectivism (Hofstede et al., 2010; Kim, 2024) is of most relevance to the present research. Consumers from individualistic cultures (such as those from North America and Western Europe) tend to develop an independent self-construal, characterised by perceiving oneself as a unique entity, separate from others. By contrast, consumers from collectivistic cultures (such as those in East Asian countries) are more likely to develop an interdependent self-construal, defining themselves in relation to the groups they belong to (Markus and Kitayama, 1991, 2010).

Culture and self-construal have been shown to affect a wide range of consumer behaviours and outcomes. For example, research has found that consumers from independent cultures, such as the USA, prefer products that express uniqueness and personal identity (Kim and Markus, 1999), whereas those from interdependent cultures, such as East Asia, prefer products that signify belonging or social acceptance (Aaker and Maheswaran, 1997). Independent consumers view brands as tools for personal identity expression, whereas interdependent ones treat brands as social facilitators or extensions of their group (Aaker and Schmitt, 2001; Escalas et al., 2005). Whereas independent consumers are driven by self-enhancement, hedonic motives and personal fulfilment, interdependent ones are motivated by social responsibilities, group harmony and role fulfilment (Hung et al., 2007; Lee et al., 2000).

Importantly, cultural differences influence the significance that individuals place on autonomy (Iyengar and Lepper, 1999; Oishi, 2000). Individuals from Western cultures prioritise personal control and autonomy in decision-making. By contrast, those from Eastern cultures value interdependence and reliance on others within their group (Brewer and Chen, 2007; Drążkowski et al., 2021; Iyengar and Lepper, 1999; Oishi, 2000). Furthermore, personal choice generally enhances motivation more for independent individuals than for interdependent individuals (Iyengar and Lepper, 1999).

The differences in the importance placed on autonomy and personal choice lead to different problem-solving behaviours. For consumers with an interdependent self-construal, seeking a complete solution from an external source is perceived as more instrumental and beneficial than learning to do it independently (Komissarouk and Nadler, 2014). For consumers with an independent self-construal, on the other hand, the ability to choose allows them to express their inner attributes, assert their autonomy and achieve uniqueness (Iyengar and Lepper, 1999). For those consumers, being provided with choice leads to more enjoyment and higher performance, even when the choices are trivial or incidental (Deci and Ryan, 1985; Legault et al., 2017).

A central feature of AI-enabled products is that they offer “a full solution” and make decisions on behalf of the consumer, without requiring their input (de Bellis et al., 2023). Consider, for example, an AI-powered washing machine that autonomously determines the optimal washing cycle by calculating the appropriate amount of detergent and water for each laundry load. When using this washing machine, customers relinquish the ability to manually select the washing cycle, as the AI system directly chooses the most optimal option rather than guiding consumers through the process. When Western consumers, who hold an independent self-view, encounter such technology, they might perceive a constraint on their ability to choose from multiple options to influence the outcome, consequently perceiving a threat to their autonomy. Conversely, for Eastern consumers who hold an interdependent self-view, the use of AI-powered home appliances represents a reliable and efficient means of obtaining assistance, as autonomy holds less significance compared to the pursuit of dependable and safe results. As a consequence, interdependent consumers may respond more positively to AI-labelled appliances, relative to independent consumers.

Therefore, we propose that culture plays a dual moderating role, influencing both the perception of autonomy threats arising from AI-labelled appliances and the extent to which these autonomy perceptions influence consumer responses. Specifically, because Western consumers place greater importance on autonomy, they perceive AI-labelled products as significantly more autonomy-threatening than Eastern consumers, and their responses towards these products are more strongly influenced by autonomy perceptions compared to Eastern consumers. Specifically:

It follows from H2 and H3 that one way of improving independent consumers’ responses towards AI-labelled appliances is to frame these appliances as less autonomy-restricting. For example, one could emphasise user-control features such as manual overrides and customisation settings, or use messaging that highlights consumer agency (e.g. “You stay in charge – AI helps you decide”). By doing so, brands can reframe AI as an enhancement of, rather than as a threat to consumer empowerment. In line with this idea, a product design that allows consumers to intervene in the actions of an autonomous smart product reduces consumers’ disempowerment (Schweitzer and Van den Hende, 2016). Similarly, framing a task as more objective increases the trust in and use of algorithms for that task (Castelo et al., 2019). Thus, we expect that for Western consumers, reframing AI-labelled appliances as enhancing rather than curtailing user control can mitigate resistance. However, for Eastern consumers, for whom autonomy is less critical (Barnes and Shavitt, 2024a), framing AI appliances as autonomy-enhancing should have little to no effect on responses:

To empirically test our hypotheses, we conducted five studies, using diverse methodologies and samples. Studies1a and 1b tested whether AI-labelled home appliances elicit more negative consumer responses than standard appliances and whether perceived autonomy mediates this effect. Using samples from the US and UK, cultures high in individualism, these studies revealed that AI-labelled products were consistently evaluated less favourably, and that diminished autonomy accounted for this negative response. Study 2 extended this investigation cross-culturally by comparing responses from Western (UK) and Eastern (Asian) participants. The results demonstrated a significant interaction between culture and AI-labelling: while Western participants exhibited lower evaluations of AI-labelled products, Eastern participants did not differentiate between AI-labelled and standard products. Study 3 tested the moderated mediation model, confirming that culture moderates both the relationship between AI- labelling and perceived autonomy, and the downstream effect of autonomy on consumer responses. Study 4 examined whether framing AI-labelled appliances as consumer autonomy-enhancing mitigates the negative reactions of Western consumers. Results showed that consumer autonomy-enhanced framing improved responses among Westerners but had a limited effect on Eastern consumers. Lastly, Study 5 analysed real-world consumer discourse by conducting a text analysis of YouTube video subtitles and comments in the USA and Korea. The findings echoed earlier results: references to autonomy were more frequently associated with anxiety in Western content than in Eastern content, providing ecological validity for the interaction between autonomy and culture observed in the prior experiments. Collectively, these five studies provide robust empirical evidence that autonomy concerns – shaped by cultural values – play a central role in how consumers evaluate AI-labelled products. Figure 1 presents the conceptual framework and overview of the studies.

Study 1 compared consumers’ reactions to AI-labelled (vs standard) home appliances. Participants were from the USA (Study 1a) and the UK (Study 1b), both countries scoring high on Hofstede’s individualism index scale (Hofstede et al., 2010). For generalisability, we used two appliance types (washing machine vs refrigerator), varied within-subjects.

Study 1a used a 2 (label type: AI vs standard; between-subjects) × 2 (product type: refrigerator vs washing machine; within-subjects) mixed design. In total, 275 participants from the USA (68.4% female, M_age = 46.32, SD = 15.13) were recruited via Prolific and were randomly assigned to the label conditions.

Participants viewed appliance advertisements (washing machine and refrigerator) in a counterbalanced order. The AI condition highlighted AI capabilities (e.g. “AI-powered washing machine […] autonomously adapts the washing cycle”.), whereas the standard appliance featured manual control (e.g. “Enhanced with specialized washing cycle […], (you) just select the menu”; Supplementary material). Participants evaluated each appliance on likelihood to use (single item), attitude (three items, α = 0.92; bad/negative/dislike, – good/positive/like), willingness to purchase (a single item) and perceived autonomy (five items, α = 0.96; Pavey and Sparks, 2009). All items used seven-point scales (see Supplementary material for all dependent measures).

A separate US sample (n = 50, 74.0% female, M_age = 36.14, SD = 11.88) evaluated perceived AI-ness (e.g. “This product is equipped with Artificial Intelligence (AI)”, washing machine: α = 0.91, refrigerator: α = 0.92). Because the “AI” label may also shift premium-ness-related perceptions (e.g. signalling sophistication or exclusivity), we measured perceived premium-ness (e.g. “This product seems luxurious”, washing machine: α = 0.87, refrigerator: α = 0.92) to rule out a potential alternative explanation for the results. AI-labelled appliances were perceived as significantly more AI-driven (washing machine: M_AI = 6.24, SD = 0.96 vs M_Standard = 4.37, SD = 1.92, t(48) = −4.29, p  < 0.001, d = 1.22; refrigerator: M_AI = 6.14, SD = 1.18 vs M_Standard = 4.42, SD = 1.81, t(48) = −3.93, p  < 0.001, d = 1.11) and more premium than the standard ones (washing machine: M_AI = 6.33, SD = 0.70 vs M_Standard = 5.76, SD = 1.00, t(48) = −2.34, p  = 0.020, d = 0.66; refrigerator: M_AI = 6.42, SD = 0.78 vs M_Standard = 5.87, SD = 0.94, t(48) = −2.22, p  = 0.031, d = 0.63). Importantly, the label-type effect on perceived AI-ness remained significant when controlling for premium-ness in a repeated-measures ANCOVA (M_AI = 6.19, SD = 0.98 vs M_Standard = 4.40, SD = 1.75; F(1, 46) = 14.29, p  < 0.001, $η p 2$ = 0.24), and the conclusion was unchanged when the model was estimated without the covariate (Supplementary material).

Because likelihood to use, attitude and willingness to purchase are conceptually distinct yet potentially correlated outcomes, we assessed their discriminant validity using two complementary approaches. Firstly, we reported AVE/CR estimates and the Fornell–Larcker matrix (Fornell and Larcker, 1981) in Supplementary material. Secondly, following Anderson and Gerbing (1988), we compared the freely estimated three-factor CFA model with nested models constraining each pairwise latent correlation to 1. For all outcome pairs, the constrained models fit significantly worse than the freely estimated model (Δχ²(1) = 24.10–98.73, p s < 0.001), indicating that the three outcomes are statistically distinguishable rather than empirically identical. Full details are reported in Supplementary material.

Repeated-measures ANOVAs on the key dependent variables (likelihood to use, attitudes and willingness to purchase) with label (AI vs standard) as a between-subject factor and product type (washing machine vs refrigerator) as a within-subjects factor revealed significant main effects of label, indicating consistently more negative evaluations of AI-labelled appliances compared to standard ones. Specifically, participants reported lower likelihood to use (M_AI = 4.67, SD = 1.70 vs M_Standard = 5.36, SD = 1.22; F(1, 273) = 14.91, p  < 0.001; $η p 2$ = 0.05), less favourable attitudes (M_AI = 5.37, SD = 1.43 vs M_Standard = 5.94, SD = 0.97; F(1, 273) = 14.72, p  < 0.001; $η p 2$ = 0.05) and lower willingness to purchase the AI-labelled versus standard appliances (M_AI = 4.76, SD = 1.79 vs M_Standard = 5.19, SD = 1.31; F(1, 273) = 5.17, p  = 0.024; $η p 2$ = 0.02). Perceived autonomy was also significantly lower in the AI condition (M_AI = 4.12, SD = 1.60 vs M_Standard = 4.87, SD = 1.27; F(1, 273) = 18.80, p  < 0.001, $η p 2$ = 0.06). Product type had a significant main effect only on willingness to purchase (M_{Washing machine} = 5.09, SD = 1.74 vs M_Refrigerator = 4.85, SD = 1.99; F(1, 273) = 3.98, p  = 0.047, η²_p = 0.01), but no significant interactions between product type and label emerged (all p s > 0.1), confirming consistent effects across both appliance types. Observed effect sizes and post hoc power analyses for this study are reported in Supplementary material.

PROCESS Model 4 (Hayes, 2022) with 5,000 bootstrap samples revealed perceived autonomy mediated the relationship between label type and responses. There were no significant direct effects on responses. However, all indirect effects via perceived autonomy were significant (likelihood to use: washing machine: b = −0.40, 95% CI [−0.66, −0.20]; refrigerator: b = −0.45, 95% CI [−0.75, −0.19]; attitude: washing machine: b = −0.28, 95% CI [−0.48, −0.13]; refrigerator: b = −0.45, 95% CI [−0.58, −0.16]; willingness to purchase: washing machine: b = −0.41, 95% CI [−0.67, −0.19]; refrigerator: b = −0.44, 95% CI [−0.73, −0.20]), indicating that AI-labelling negatively influenced these outcomes primarily by undermining perceived autonomy.

Study 1a supported H1 and H2, showing reduced autonomy as the mechanism underlying negative responses to AI-labelled appliances. Study 1b aimed to replicate these findings using a different sample and products, ensuring effects were driven by AI-labelling rather than product functionality differences. It also tested whether merely including the term “AI” in the product label is sufficient to reduce consumer responses among individualistic consumers, thereby replicating and extending Study 1a.

Study 1b used a 2 (label type: AI vs standard; between-subjects) × 2 (product type: robot vacuum cleaner vs smart bulb; within-subjects) mixed design. In total, 156 UK participants (60.3% female, M_age = 41.35, SD = 13.05) recruited via Prolific were randomly assigned to the label conditions.

Participants viewed advertisements for a robot vacuum cleaner and a smart bulb in a counterbalanced order. Advertisements were identical across conditions except for the presence of the “AI” label in the AI condition (e.g. “the [AI-powered] robot vacuum chooses the most optimal cleaning program”; Supplementary material). Participants evaluated products using the same measures as those in Study 1a: likelihood to use (single item), attitude (three items, α = 0.94), willingness to purchase (single item) and perceived autonomy (five items, α = 0.95).

A separate UK sample (n = 55, 49.1% female, M_age = 43.29, SD = 12.81) confirmed the AI-labelled products were perceived as significantly more AI-driven than standard products (robot vacuum cleaner: M_AI = 6.51, SD = 0.69 vs M_Standard = 5.64, SD = 1.23; t(53) = −3.17, p = 0.003, d = 0.86; smart bulb: M_AI = 5.83, SD = 1.27 vs M_Standard = 4.95, SD = 1.59; t(53) = −2.25, p  = 0.029, d = 0.61). Premium-ness did not differ by label type (p s > 0.100). The label-type effect on perceived AI-ness remained significant both with and without controlling for premium-ness (see Supplementary material).

Using the same procedures as in Study 1a, we again assessed discriminant validity for likelihood to use, attitude and willingness to purchase (see Supplementary material). The Anderson and Gerbing (1988) chi-square difference tests indicated that fixing any pairwise latent correlation to 1 significantly worsened model fit relative to the freely estimated model (Δχ²(1) = 19.38–97.74, p s < 0.001). Accordingly, we continue to report the three outcomes separately in the subsequent studies.

Repeated-measures ANOVAs revealed significant main effects of label type. AI-labelled products elicited lower likelihood to use (M_AI = 4.20, SD = 1.67 vs M_Standard = 4.97, SD = 1.33; F(1, 154) = 10.14, p  = 0.002, η²_p = 0.06), less favourable attitude (M_AI = 5.09, SD = 1.29 vs M_Standard = 5.62, SD = 1.17; F(1, 154) = 7.19, p  = 0.008, η²_p = 0.05), lower willingness to purchase (M_AI = 4.40, SD = 1.72 vs M_Standard = 5.25, SD = 1.39; F(1, 154) = 11.42, p  = 0.001, η²_p =0.07) and lower perceived autonomy (M_AI = 4.36, SD = 1.59 vs M_Standard = 5.14, SD = 1.28; F(1, 154) = 11.53, p  = 0.001, η²_p = 0.07) than standard products. Product type significantly affected only perceived autonomy (M_{Vacuum cleaner} = 4.52, SD = 1.71 vs M_{Smart bulb} = 4.97, SD = 1.56; F(1, 154) = 17.43, p  < 0.001; η²_p = 0.10); no significant interactions emerged (all p s > 0.100). Observed effect sizes and post hoc power analyses for this study are reported in Supplementary material.

PROCESS Model 4 (Hayes, 2022) with 5,000 bootstrap samples showed perceived autonomy mediated the relationship between label type and responses. The direct effects of label type were non-significant; however, all indirect effects via perceived autonomy were significant (likelihood to use: robot vacuum cleaner: b = −0.48, 95% CI [−0.77, −0.20]; smart bulb: b = −0.42, 95% CI [−0.76, −0.16]); attitude: vacuum cleaner: b = −0.38, 95% CI [−0.68, −0.10]; smart bulb: b = −0.24, 95% CI [−0.48, −0.05]; willingness to purchase: vacuum cleaner: b = −0.53, 95% CI [−0.90, −0.21]; smart bulb: b = −0.46, 95% CI [−0.80, −0.16]). Thus, AI-labelling negatively impacted consumer responses primarily by reducing perceived autonomy.

Study 1b confirmed that merely labelling products as “AI” negatively impacted evaluations through diminished autonomy perceptions. Next, we examined whether these findings generalise beyond Western, individualistic contexts.

Study 2 examined if the negative effect of AI-labelling on appliance evaluations is moderated by culture, hypothesising that Eastern (vs Western) consumers are less negatively impacted (H3). Participants represented Western (high individualism) and Eastern (low individualism) cultures based on Hofstede’s individualism index (Hofstede et al., 2010).

Study 2 used a 2 (label type: AI vs standard) × 2 (culture: Eastern vs Western) between-subjects design. Initially, 315 participants (58.4% female; M_age = 38.97, SD = 13.47) were recruited via Prolific and randomly assigned to the label conditions. After excluding 27 participants for incomplete or conflicting nationality data, the final sample included 288 participants (M_age = 39.19, SD = 13.73, 59.4% female). The Western group (n = 155) consisted of UK participants, and the Eastern group (n = 133) was primarily from China (25.6%), South Korea (13.6%), Pakistan (9.8%) (Supplementary material).

Participants viewed the refrigerator advertisement from Study 1a (Supplementary material) and evaluated the refrigerator on likelihood to use, attitude (α = 0.95) and willingness to purchase.

A separate group of 100 participants (47% female, M_age = 36.12, SD = 10.21) from Western (UK and US) and Eastern (Asian) countries completed the manipulation check task. Consistent with studies 1a and 1b, we measured perceived premium-ness to rule out a potential confound. A 2 (label type: AI vs standard) × 2 (culture: Western vs Eastern) between-subjects ANCOVA (controlling for perceived premium-ness) on perceived AI-ness revealed a significant main effect of label type (M_AI = 6.56, SD = 0.66 vs M_Standard = 4.30, SD = 1.88; F(1, 95) = 61.77, p  < 0.001, η²_p = 0.40), whereas culture and the interaction effect were non-significant (p s > 0.100). The label-type effect remained significant when the model was estimated without controlling for perceived premium-ness (see Supplementary material)

An independent-samples t-test on the Hofstede scores confirmed the cultural grouping: Western participants scored significantly higher on individualism than Eastern participants (M_Western = 89.00, SD = 0.00 vs M_Eastern = 22.57, SD = 11.86, t(286) = 69.74, p  < 0.001, d = 8.06).

Western participants reported a significantly lower likelihood to use the AI-labelled refrigerator (vs the standard refrigerator) (M_AI = 4.31, SD = 1.90 vs M_Standard = 5.37, SD = 1.43; F(1, 284) = 16.64, p  < 0.001, η²p = 0.06), less favourable attitudes (M_AI = 5.31, SD = 1.51 vs M_Standard = 5.99, SD = 1.03; F(1, 284) = 11.24, p  = 0.001, η²p = 0.04) and reduced willingness to purchase (M_AI = 4.26, SD = 2.05 vs M_Standard = 5.13, SD = 1.58; F(1, 284) = 9.68, p  = 0.002, η²p = 0.03). By contrast, Eastern participants showed no significant differences across measures (p s > 0.740). Cultural differences appeared only in the AI-labelled condition, with Western participants rating appliances significantly less favourably than Eastern participants (p s < 0.001). No cultural differences were found in the standard condition (p s > 0.590). Observed effect sizes and post hoc power analyses for this study are reported in Supplementary material.

Study 2 confirmed the moderating role of culture proposed in H3: Western consumers had significantly less favourable attitudes toward AI-labelled appliances compared to standard ones, while Eastern consumers did not differ. The following study explored whether these cultural differences are explained by perceived autonomy, identified as a key mediator in Studies 1a and 1b.

Study 3 examined how AI-labelled appliances influence autonomy perceptions among Western and Eastern consumers and how these perceptions, in turn, affect appliance evaluations.

Study 3 used a 2 (label type: AI vs standard) × 2 (culture: Eastern vs Western) between-subjects design. A total of 602 participants (49.0% female; M_age = 33.77, SD = 9.81) recruited via Prolific were randomly assigned to the label conditions. Cultural grouping used nationality: Western (94.3% UK), Eastern (China 31.5%, India 37.1%, Japan 4.3%; Supplementary material).

Participants viewed the washing machine advertisement from Study 1a (Supplementary material) and evaluated the washing machine on likelihood to use, attitude (three items, α = 0.96), willingness to purchase and perceived autonomy (five items, α = 0.94).

A separate group of 100 participants (47% female, M_age = 36.12, SD = 10.21) from Western (UK and US) and Eastern (Asian) countries completed the manipulation check task. A 2 (label type: AI vs standard) × 2 (culture: Western vs Eastern) between-subjects ANCOVA (controlling for perceived premium-ness) on perceived AI-ness revealed a significant main effect of label type (M_AI = 6.27, SD = 1.06 vs M_Standard = 4.13, SD = 1.86; F(1, 95) = 62.19, p  < 0.001, η²_p = 0.40), whereas culture and the interaction effect were non-significant (p s > 0.100). The label-type effect remains significant without controlling for perceived premium-ness (see Supplementary material).

Western participants had significantly higher individualism scores than Eastern participants (M_Western = 75.01, SD = 5.31 vs M_Eastern = 31.36, SD = 14.29, t(600) = 49.60, p  < 0.001, d = 4.04), validating cultural grouping.

Two-way ANOVAs (label type: AI vs standard × culture: Western vs Eastern) revealed significant interactions for all key outcomes and for perceived autonomy (p s < 0.004). Western participants responded more negatively to the AI-labelled washing machines than to standard ones: lower likelihood to use (M_AI = 4.22, SD = 1.91 vs M_Standard = 5.50, SD = 1.10, F(1, 598) = 53.55, p  < 0.001; η²_p = 0.08) (see Figure 2), less favourable attitudes (M_AI = 4.89, SD = 1.76 vs M_Standard = 5.84, SD = 1.02; F(1, 598) = 38.04, p  < 0.001, η²_p = 0.060), reduced willingness to purchase (M_AI = 4.33, SD = 2.11 vs M_Standard = 5.27, SD = 1.34; F(1, 598) = 25.33, p  < 0.001, η²_p = 0.04) and reduced perceived autonomy (M_AI = 3.87, SD = 1.62 vs M_Standard = 5.16, SD = 1.18); F(1, 598) = 63.93, p  < 0.001; η²_p = 0.10). By contrast, Eastern participants showed no significant differences in likelihood to use, attitude or willingness to purchase (all ps > 0.740) (see Figure 2), although perceived autonomy was lower in the AI-labelled condition than in the standard condition (M_AI = 4.72, SD = 1.51 vs M_Standard = 5.36, SD = 1.19; F(1, 598) = 15.84, p  < 0.001; η²_p = 0.03), with a markedly smaller effect than among Western participants. Cultural differences appeared only in the AI-labelled condition, with Western participants evaluating the AI-labelled washing machine less favourably than Eastern participants across measures (all p s < 0.001), with no differences in the standard condition (p s > 0.210). These interaction patterns are consistent with H3 and motivate the moderated mediation test reported next. Observed effect sizes and post hoc power analyses for this study are reported in Supplementary material.

To test H3, we estimated PROCESS Model 58 (Hayes, 2022), which specifies a single mediator (perceived autonomy) and allows culture (Western vs Eastern) to moderate both the effect of label type (AI vs standard) on autonomy (Path a) and the effect of autonomy on consumer responses (Path b). We estimated the model separately for each outcome (likelihood to use, attitude and willingness to purchase).

The autonomy-reducing effect of AI-labelling was stronger among Western than Eastern participants (β_Western = −1.28, p  < 0.001 vs β_Eastern = −0.64, p < 0.001).

Autonomy effects on responses were stronger among Western participants than Eastern participants for likelihood to use (β_Western = 0.68 vs β_Eastern = 0.51; interaction β = −0.16, p  = 0.022) and willingness to purchase (β_Western = 0.73 vs β_Eastern = 0.52; interaction β = −0.21, p  = 0.007), with a directionally consistent pattern for attitude (β_Western = 0.59 vs β_Eastern = 0.48; interaction β = −0.10, p  = 0.09).

Consistent with these dual-stage moderation effects, the indirect effects of label type on responses via perceived autonomy were stronger for Western participants (likelihood to use: b = −0.87, 95% CI [−1.16, −0.59]; attitude: b = −0.75, 95% CI [−1.02, −0.51]; willingness to purchase: b = −0.93, 95% CI [−1.24, −0.65]) than for Eastern participants (likelihood to use: b = −0.33, 95% CI [−0.52, −0.16]; attitude: b = −0.31, 95% CI [−0.49, −0.15]; willingness to purchase: b = −0.33, 95% CI [−0.53, −0.16]). The index of moderated mediation was significant for likelihood to use (0.54, 95% CI [0.21, 0.88]), attitude (0.45, 95% CI [0.14, 0.76]) and willingness to purchase (0.60, 95% CI [0.26, 0.95]), supporting H3.

Study 3 robustly supported H3, showing a dual moderating role of culture. Western participants perceived greater autonomy threats from AI-labelled products, leading to stronger negative reactions than Eastern participants. These findings align with prior evidence suggesting interdependent cultures more readily accept externally guided decision-making (Barnes and Shavitt, 2024a,b). The next study investigated whether strategic message framing can mitigate autonomy concerns.

Study 4 tested whether consumer autonomy-enhanced advertising mitigates the negative impact of AI-labelling for Western consumers.

Study 4 used a 2 (culture: Western vs Eastern; between-subjects) × 3 (advertisement type: automatic AI vs consumer autonomy-enhanced AI vs standard; within-subjects) × 2 (product: oven vs washing machine, within-subjects) mixed design. A total of 604 participants (48.5% female; M_age = 38.35, SD = 12.80) were recruited via Prolific and culturally grouped by nationality. The Western group consisted mainly of UK participants (89.4%), whereas the Eastern group included participants from India (47.0%), China (12.9%) and Japan (6.3%; Supplementary material).

Participants viewed three advertisements in a counterbalanced order:

1. automatic AI-labelled;

2. consumer autonomy-enhanced AI labelled; and

3. standard.

For half of participants, the advertisements featuring ovens were presented first, followed by the washing machine advertisements; for the other half, the order was reversed. (See Supplementary material for the oven and washing machine ads). Participants evaluated each advertisement on likelihood to use (a single item).

Western participants had significantly higher individualism scores than Eastern participants (M_Western = 75.99, SD = 0.17 vs M_Eastern = 29.66, SD = 15.21, t (602) = 52.92, p  < 0.001, d = 4.31), confirming the cultural grouping.

A separate group of 100 participants (48% female, M_age = 41.54, SD = 15.04; 50 Western, 50 Eastern) assessed the extent to which each advertisement emphasised consumer autonomy (e.g. this oven/washing machine emphasises that users have freedom of choice and are in control of decision-making). Two separate 3 (advertisement type: Automatic AI vs Consumer Autonomy-enhanced AI vs Standard; within-subjects) × 2 (culture: Western vs Eastern; between-subjects) mixed ANCOVAs, controlling for product enjoyment (cooking/laundry enjoyment) showed significant effects of advertisement type (oven: F(2, 194) = 4.79, p  = 0.009, η²_p = 0.05; washing machine: F(2, 194) = 25.92, p  < 0.001, η²_p = 0.21). The comsumer autonomy-enhanced and the standard ads emphasised perceived autonomy more than the automatic AI ads (oven: M_Automatic = 3.27, SD = 2.16 vs M_{Autonomy-enhanced} = 5.77, SD = 1.12 vs M_Standard = 6.08, SD = 1.32; washing machine: M_Automatic = 3.38, SD = 2.16, M_{Autonomy-enhanced} = 5.77, SD = 1.18, M_Standard = 5.93, SD = 1.41; all p s < 0.001). The consumer autonomy-enhanced and the standard ads did not differ significantly. These results confirmed the manipulation was effective regardless of culture and product enjoyment.

A 3 × 2 × 2 repeated-measures ANOVA on the likelihood to use revealed significant main effects of advertisement type (F(2,1204) = 75.20, p  < 0.001, η²_p = 0.11), culture (F(1,602) = 15.73, p < 0.001, η²_p = 0.03) and product (F(1,602) = 7.44, p  = 0.007, η²_p = 0.01). The three-way interaction was non-significant (p = 0.184). Significant two-way interactions emerged between culture and advertisement type (F(2,1204) = 82.05, p  < 0.001, η²_p = 0.12) and advertisement and product (F(2,1204) = 29.42, p < 0.001, η²_p = 0.05). The interaction between culture and product was non-significant (p  = 0.176).

Western participants showed significantly greater likelihood to use consumer autonomy-enhanced (M = 5.14, SD = 1.61) than automatic AI-labelled appliances (M = 3.97, SD = 1.86; p < 0.001). However, they preferred the standard appliances (M = 5.77, SD = 1.21; p < 0.001) significantly more than both consumer autonomy-enhanced and automatic AI-labelled appliances (all pairwise comparisons, < 0.001). Eastern participants showed significantly higher likelihood to use consumer autonomy-enhanced (M = 5.79, SD = 1.09) than both automatic AI-labelled (M = 5.08, SD = 1.57) and standard appliances (M = 4.87, SD = 1.35; p s < 0.001) (see Figure 3). Additionally, they significantly preferred automatic AI-labelled appliances over standard appliances (p  = 0.029). Western participants preferred standard appliances significantly more than Eastern participants, while they were significantly less likely to use automatic and autonomy-enhanced AI appliances (p s < 0.001).

For ovens, standard (M = 5.39, SD = 1.48) and consumer autonomy-enhanced (M = 5.49, SD = 1.55) were significantly preferred over automatic AI-labelled (M = 4.29, SD = 2.06, p  < 0.001). The difference between standard and consumer autonomy-enhanced ovens was not significant (p  = 0.245). For washing machines, consumer autonomy-enhanced (M = 5.44, SD = 1.54) was preferred over both standard (M = 5.24, SD = 1.53, p  < 0.01) and automatic AI-labelled (M = 4.76, SD = 1.95, p  < 0.001). Standard was preferred over automatic (p  < 0.001).

Study 4 confirmed that consumer autonomy-enhanced advertising effectively increased acceptance of AI-labelled appliances among Western consumers, supporting H4. However, consumer autonomy-enhanced framing did not fully restore autonomy perceptions to levels associated with standard products. Observed effect sizes and post hoc power analyses for this study are reported in Supplementary material. While Studies 1–4 provided robust experimental evidence of cultural moderation, Study 5 explored if these patterns generalise to real-world consumer conversations via content analysis of YouTube video subtitles and comments.

Study 5 extended the experimental findings (Studies 1–4) by analysing social media data to examine if actual consumer conversations provided evidence for cultural differences in emotional responses to autonomy framing. Specifically, YouTube subtitles and viewer comments from two culturally distinct countries, the USA (Western) and South Korea (Eastern), were analysed.

YouTube videos featuring AI-related content with over 300 views (January to October 2023) were included. The data set comprised 202 videos (101 English, 101 Korean) and 77,674 viewer-generated comments (55,564 from the USA, 22,110 from Korea). Korean subtitles were translated into English using Google Translate and manually reviewed. Subtitle texts were analysed at the video level (i.e. each video’s subtitle text represented a single unit of analysis, n = 202), whereas viewer-generated comments were analysed at the individual-comment level (i.e. each comment represented an independent unit of analysis, n = 77,674).

An autonomy dictionary (86 terms) was created following LIWC2022 guidelines (Boyd et al., 2022) and internally validated (Supplementary material). Word frequencies were normalised per 1,000 words. Using LIWC-22, anxiety (e.g. “worried”, “nervous”) and positive tone (e.g. emoticons: “(-:” and words: “beneficial”, “care”) were measured (Supplementary material). These measures are widely used to assess emotional conversation in consumer research (Boyd et al., 2022; Gray and Wegner, 2012; Mori et al., 2012).

Study 5 provides a field-based, partial test of H3 using naturally occurring YouTube data. Unlike our experiments (e.g. Study 3), the social media data set does not allow us to cleanly classify observations as “AI-labelled” versus “standard-labelled” products, making it impractical to estimate the full dual-stage moderated mediation model in PROCESS Model 58. Instead, we test the central boundary condition implied by H3 – whether culture moderates the association between autonomy-related language and evaluative sentiment – by conducting separate analyses of video subtitles (video-level transcripts) and viewer comments (comment-level text).

For subtitles, the unit of analysis was the video-level transcript (n = 202): we computed the frequency of autonomy-related terms per 1,000 words and examined whether culture moderated the relationship between autonomy language and LIWC-based sentiment in the transcript (anxiety and positive tone). For comments, the unit of analysis was the individual comment (n = 77,674): we tested whether autonomy-related language in a comment predicted comment sentiment differently across cultures. In both data sets, we estimated moderation models of the form: sentiment = autonomy + culture + autonomy × culture (PROCESS Model 1) (Hayes, 2022).

Autonomy-related language marginally interacted with culture in predicting anxiety (β = −0.06, p = 0.080). In Western videos, greater use of autonomy terms was associated with higher anxiety (β = 0.05, 95% CI [0.00, 0.10]), whereas this association was not observed in Eastern videos (β = −0.01, 95% CI [−0.05, 0.03]).

Autonomy significantly interacted with culture in predicting positive tone (β = 0.26, p  < 0.001). Autonomy-related terms were less likely to co-occur with positive tone among Western commenters (β = −0.26, 95% CI [−0.30, −0.22]), whereas no such association emerged among Eastern commenters (β = −0.00, 95% CI [−0.06, 0.06]). Observed effect sizes and post hoc power analyses for this study are reported in Supplementary material.

Together, these patterns suggest that autonomy-related discourse around AI is more negatively valenced in Western than Eastern contexts, consistent with the theorised cultural boundary condition of autonomy observed in our experiments. While text analysis cannot fully disambiguate the intent behind autonomy references, convergence across subtitles (content framing) and comments (audience reactions) strengthens the ecological validity of our conclusions.

The series of studies presented in this research offers a comprehensive understanding of consumer responses to AI-labelled home appliances, highlighting the roles of autonomy perceptions and cultural differences (see Supplementary material for summary of studies). Study 1, conducted in individualistic Western contexts (USA, UK), revealed that AI-labelled home appliances were less favoured than standard appliances, primarily due to perceived reduced autonomy. Although individualistic consumers reported moderately positive attitudes towards AI-labelled appliances, they were less likely to use or purchase these appliances, relative to standard ones. This gap aligns with research showing that behavioural intentions are more conservative and predictive of actual purchases than evaluative attitudes alone (Ajzen, 1991; Morwitz and Fitzsimons, 2004). While AI labels may enhance overall product evaluations through associations with innovation and premium quality, consumers may still be reluctant to adopt them due to autonomy concerns. Studies 2 and 3 expanded the investigation by showing that Western participants consistently displayed less favourable responses toward AI-labelled appliances than Eastern participants, due to reduced perceived autonomy. The influence of autonomy on consumer responses was stronger among Western participants, highlighting cultural variability in reactions to consumner autonomy-enhanced messaging. Study 4 indicated that consumer autonomy-enhanced advertisements, which position AI as a facilitator or supporter of consumer decisions rather than as an autonomous solution provider, effectively mitigated Western consumers’ resistance. Extending beyond controlled experiments, Study 5 analysed real-world YouTube videos and viewer comments and showed that autonomy-related content consistently co-occurred with increased negative emotional responses to AI-labelled appliances (higher anxiety, lower positive sentiment) among Western consumers. This pattern was weak or absent among Eastern consumers.

Our research makes several theoretical contributions. Firstly, we contribute to research on consumer responses to AI-labelled products (Castelo et al., 2019; Chugunova and Sele, 2022; Puntoni et al., 2021). While prior work has highlighted the benefits of AI in domains where efficiency, accuracy or objectivity are valued, our findings demonstrate that even in such domains, AI-labelling can backfire. By showing that consumers evaluate AI-labelled home appliances less favourably than functionally identical non-AI-labelled ones, we reveal a previously underexplored negative bias towards AI in utilitarian product categories. This challenges the prevailing assumption that AI cues universally signal enhanced competence and suggests that marketers should be cautious when using AI branding, even in functional domains. Recent work also shows that AI can dampen perceptions of premium-ness, as service robots often signal standardisation rather than exclusivity (Hoang et al., 2025). This suggests that AI cues can prompt unintended negative inferences and that such inferences may vary across cultures. For global brands, cultural perceptions of AI may therefore influence how luxury or premium cues are received.

Secondly, we contribute to the literature on autonomy and psychological reactance in consumer decision-making (Brehm and Cole, 1966; Deci and Ryan, 2000; Fitzsimons and Lehmann, 2004). We show that the negative response to AI-labelled appliances is driven by consumers’ perceived loss of autonomy – a sense that AI may take over control or decision-making in an unwanted manner. Although autonomy is central in self-determination theory and consumer empowerment research, it has received limited attention in the context of AI-labelling. Our findings thus illuminate autonomy as a key psychological mechanism underlying resistance to AI adoption and suggest that reactance may occur not only when choice is removed but also when consumers perceive that technology is exerting undue influence on their actions or routines.

Thirdly, we add to the growing body of research on culture and consumer technology adoption (Aaker and Williams, 1998; Varnali, 2021; Zhang and Shrum, 2009), by demonstrating that the effect of AI-labelling on product evaluation is moderated by cultural orientation. Specifically, we find that consumers from individualistic cultures react more negatively to AI-labelled appliances than those from collectivistic cultures, consistent with prior work showing that individualists place greater emphasis on autonomy and personal agency. This contributes to cross-cultural marketing research by highlighting how cultural values interact with technological cues – in this case, AI framing – to shape consumer evaluations. It also underscores the importance of tailoring AI-related communication strategies to cultural contexts, a dimension that remains underexplored in existing AI marketing research.

Lastly, by integrating insights from AI adoption, psychological reactance and cross-cultural psychology, our work bridges technology-focused marketing research with broader themes of consumer empowerment and control. This responds to recent calls for more nuanced, psychologically grounded accounts of consumer–AI interactions (Puntoni et al., 2021). Rather than assuming linear effects of technological innovation on consumer preferences, our findings reveal that consumers may experience ambivalence or resistance, especially when AI is perceived as infringing upon core values such as autonomy.

As AI technology becomes increasingly integrated into home appliances, companies focus on promoting the advanced capabilities of AI-labelled appliances. However, our findings indicate a disconnect between this strategy and consumers’ concerns about autonomy infringement. Our finding that the mere addition of “AI” to product labels reduced purchase intentions among Western consumers emphasises the need for culturally sensitive branding and advertising strategies. When marketing AI appliances to Western consumers, marketers would benefit from framing these appliances as tools that enhance, or at least do not diminish, personal autonomy. In Eastern markets, by contrast, AI-labelling should be encouraged as it is unlikely to prompt negative inferences.

The implications of this research extend beyond advertising to the product development stage. Companies could enhance consumer acceptance of AI-labelled appliances in Western markets by adding product features that afford greater user autonomy and flexibility. For example, an AI washing machine could have options that allow the consumer to override the AI-driven settings. Recognising and accommodating cultural differences during the design and marketing stages can significantly improve consumer acceptance and satisfaction with AI-labelled home appliances.

The supplementary material for this article can be found online.