Algorithmic Sustainability or Algorithmic Excess? How Generative AI Nudging Shapes Sustainable Consumer Behaviour

1. Introduction

Artificial Intelligence (AI) is no longer a peripheral tool in contemporary markets; it has become a structuring force that shapes how consumers perceive, evaluate, and ultimately engage with products and services (Bock, Wolter & Ferrell, 2020 ; Du & Xie, 2021 ; Moleka, 2024a-c). Across digital platforms, AI systems curate information environments, personalize recommendations, and increasingly mediate the very conditions under which choices are made. At the same time, sustainable consumption has emerged as a central societal and policy concern, driven by escalating environmental crises, regulatory pressures, and shifting consumer expectations (Silalahi, 2025). Despite the growing importance of both domains, their intersection remains insufficiently theorized within consumer behaviour research.

Existing studies suggest that AI can facilitate more sustainable consumption by reducing information asymmetries, optimizing decision processes, and guiding consumers toward environmentally friendly alternatives (Sargin, 2024 ; Pentina, Xie, Hancock & Bailey, 2023; Satornino, Du & Grewal, 2024). For instance, AI-powered recommendation systems can highlight lower-impact products, estimate carbon footprints, and align suggestions with consumers’ stated values. From this perspective, AI appears as a powerful enabler of sustainability transitions, capable of operationalizing behavioural change at scale. However, a growing body of research also points to more ambivalent effects. By optimizing convenience, reducing friction, and amplifying personalization, AI may inadvertently intensify consumption, reinforce habitual purchasing patterns, and obscure the broader consequences of individual choices (Puntoni, Reczek, Giesler & Botti, 2021).

This duality reflects a deeper tension that has not yet been fully conceptualized in the literature. Much of the existing research treats AI as an instrumental variable—something that can be designed and deployed to produce desired outcomes. Yet this perspective underestimates the extent to which AI systems actively reshape decision environments and co-evolve with consumer behaviour. In parallel, research on sustainable consumption has emphasized behavioural interventions such as nudging and social marketing (Peattie & Peattie, 2009), but has largely assumed that these interventions operate in relatively stable contexts. The emergence of AI challenges this assumption by introducing dynamic, adaptive, and often opaque mechanisms of influence.

Against this backdrop, the present research addresses the following central question: how does AI reshape sustainable consumer behaviour, and under what conditions does it simultaneously enable and undermine sustainability outcomes? To answer this question, we develop a novel theoretical framework that conceptualizes AI not merely as a tool, but as a generative infrastructure of consumption. We introduce the concept of Generative AI Nudging (GAIN) and propose the Algorithmic Sustainability Paradox (ASP), which captures the simultaneous and interdependent positive and negative effects of AI on sustainable behaviour.

Empirically, we test this framework across two complementary studies. Study 1 examines how AI-driven nudging and personalization influence immediate consumer choices, perceived autonomy, and trust. Study 2 extends this analysis over time, revealing how initial gains in sustainable behaviour may give way to rebound effects driven by moral licensing and behavioural adaptation. Together, these studies provide a dynamic and nuanced understanding of AI’s role in shaping sustainable consumption.

By integrating insights from consumer behaviour theory, human–AI interaction research, and sustainability studies, this paper contributes to advancing a more realistic and theoretically grounded understanding of AI-driven markets. Rather than assuming that technological sophistication necessarily leads to better outcomes, we highlight the conditions under which AI may both support and undermine sustainability goals, thereby offering a more balanced and actionable perspective for scholars, practitioners, and policymakers.

2. Theoretical Background and Hypotheses

2.1. AI as a Transformative Force in Consumer Behaviour

The integration of AI into consumer-facing systems has fundamentally altered the architecture of decision-making. Traditional models of consumer behaviour often assume that individuals operate within relatively stable environments, where information is accessed, processed, and evaluated through cognitive and affective mechanisms (Ledro, Nosella, Vinelli, Dalla Pozza & Souverain, 2025 ; Wang, Chen & Kuang, 2025). However, AI systems disrupt this model by continuously reshaping the informational and choice environment in which consumers operate. Recommendation engines, predictive analytics, and adaptive interfaces do not merely assist decision-making; they actively configure the set of available options and the salience of particular attributes (Santiago, Febiansyah & Dinarwati, 2024 ; Gangadharan, Purandaran, Malathi, Subramanian, Jeyaraj & Jung, 2025).

Recent research has begun to conceptualize AI as an interaction partner rather than a passive tool, emphasizing its role in shaping consumer experiences and perceptions (Peltier, Dahl & Schibrowsky, 2024 ; Alabed, Javornik, Gregory-Smith & Casey, 2024). From this perspective, consumers do not simply respond to AI-generated outputs; they develop beliefs, expectations, and emotional responses toward the system itself. These relational dynamics are critical for understanding how AI influences trust, autonomy, and ultimately behaviour. At the same time, AI systems are designed to optimize specific objectives—often related to engagement, conversion, or efficiency—which may not align with broader societal goals such as sustainability (Moleka, 2025a-d ; Moleka, 2026a-c).

This tension is particularly relevant in the context of consumption, where AI-driven optimization can lead to increased convenience and reduced effort, but also to higher levels of consumption. By lowering cognitive and transactional barriers, AI systems may encourage more frequent purchasing, thereby amplifying environmental impacts even when individual choices appear more sustainable.

2.2. Sustainable Consumption and Behavioural Interventions

Research on sustainable consumption has long emphasized the gap between attitudes and behaviour, often referred to as the “attitude–behaviour gap.” While many consumers express concern for environmental issues, this concern does not consistently translate into sustainable purchasing decisions. To address this gap, scholars and practitioners have increasingly turned to behavioural interventions, including nudges, defaults, and social influence strategies (Park & Lin, 2020 ; Peattie & Peattie, 2009).

These interventions are typically designed to steer behaviour without restricting choice, leveraging cognitive biases and heuristics to promote desirable outcomes. However, most existing approaches assume that interventions are static and that their effects are relatively stable over time. This assumption becomes problematic in AI-mediated environments, where interventions are continuously adapted based on user data and behavioural feedback (Aghajari, Baumer, Hohenstein, Jung & DiFranzo, 2023 ; Mihai, Aleca & Iordache, 2025).

Moreover, the effectiveness of behavioural interventions depends not only on their design but also on how they are perceived by consumers. Perceived autonomy, trust, and fairness play crucial roles in determining whether individuals accept or resist influence attempts. When interventions are perceived as manipulative or overly intrusive, they may trigger reactance or reduce trust, thereby undermining their effectiveness.

2.3. Generative AI Nudging (GAIN)

Building on these insights, we introduce the concept of Generative AI Nudging (GAIN), defined as adaptive, data-driven behavioural interventions that are dynamically generated and optimized by AI systems in real time. Unlike traditional nudges, which are typically pre-designed and static, GAIN operates through continuous learning and feedback loops. It adjusts not only the content of recommendations but also their timing, framing, and level of personalization.

This dynamic nature makes GAIN particularly powerful, as it allows for highly tailored interventions that align closely with individual preferences and behaviours (Hekler, Klasnja, Riley, Buman, Huberty, Rivera & Martin, 2016 ; Kudapa, 2024). However, it also raises important ethical and psychological questions. As personalization increases, consumers may experience a reduction in perceived autonomy, particularly if they feel that their choices are being guided or constrained by unseen algorithms (Dubazana, 2024 ; Gal, 2018). At the same time, the opacity of AI systems may limit consumers’ ability to understand and evaluate the basis of recommendations, further complicating their responses (Grochowski, Jablonowska, Lagioia & Sartor, 2021).

2.4. The Algorithmic Sustainability Paradox (ASP)

The central theoretical contribution of this paper is the Algorithmic Sustainability Paradox (ASP), which posits that AI-driven systems can simultaneously promote and undermine sustainable consumption. On the one hand, AI enables more informed and efficient decision-making, reduces search costs, and facilitates access to sustainable alternatives. On the other hand, it can reinforce consumption habits, create rebound effects, and obscure the broader implications of individual choices.

This paradox reflects the interaction between technological capabilities and human psychology. For example, while AI may encourage consumers to choose environmentally friendly products, these choices may also lead to moral licensing, whereby individuals feel justified in engaging in less sustainable behaviour subsequently (Pawar, Chavan, Vhatkar, Khang & Gawankar, 2025). Similarly, increased efficiency may reduce the perceived cost of consumption, leading to higher overall consumption levels.

2.5. Hypotheses Development

Based on the theoretical framework outlined above, we propose the following hypotheses:

H1: 

AI-driven nudging increases the likelihood of choosing sustainable products, as it enhances the salience and accessibility of environmentally friendly options (Pentina, Xie, Hancock & Bailey, 2023).

H2: 

High levels of personalization reduce perceived autonomy, as consumers may feel that their choices are being guided or constrained by algorithmic systems (Puntoni, Reczek, Giesler & Botti, 2021).

H3: 

Perceived autonomy mediates the relationship between AI nudging and trust, such that lower autonomy leads to reduced trust in the system (Hou, Yang, Xiong & Pavlou, 2025).

H4: 

AI nudging leads to rebound effects over time, as initial sustainable behaviour may trigger moral licensing and increased subsequent consumption (Satornino, Du & Grewal, 2024).

3. Study 1: Immediate Effects of AI Nudging and Personalization on Sustainable Choice

Study 1 was designed to provide a rigorous examination of the immediate behavioural and psychological consequences of AI-driven nudging within a controlled decision-making context. While prior literature has established that algorithmic systems can influence consumer preferences and guide decision-making processes (Pentina, Xie, Hancock & Bailey, 2023), relatively little attention has been paid to the experiential dimension of such influence, particularly in relation to sustainability-oriented choices. This study therefore seeks to bridge this gap by investigating not only whether AI nudges are effective in promoting sustainable consumption, but also how they shape consumers’ perceptions of autonomy and trust—two constructs that are central to the legitimacy and long-term effectiveness of behavioural interventions.

To address these objectives, we employed a 2 (AI nudging: present vs. absent) × 2 (personalization: low vs. high) between-subjects experimental design. This design allows for a nuanced analysis of both the direct impact of AI-generated recommendations and the moderating role of personalization intensity. The inclusion of personalization as a key variable reflects the increasing sophistication of AI systems, which are not merely capable of recommending products but can do so in ways that are highly tailored to individual users. Such personalization has been widely celebrated as a means of enhancing relevance and user engagement; however, it may also introduce unintended psychological consequences by altering how consumers perceive the origin and ownership of their decisions.

Participants were immersed in a simulated online shopping environment designed to replicate common digital consumption experiences. They were presented with a choice between two products within a given category: one characterized by environmentally friendly attributes (e.g., eco-certification, lower environmental impact) and another representing a conventional alternative with a lower price point. In the AI nudging condition, participants were exposed to a recommendation explicitly framed as generated by an intelligent system, thereby activating the perception of algorithmic guidance. In contrast, participants in the control condition made their choices without such intervention. Personalization was manipulated through the framing of the recommendation message, with high-personalization conditions referencing individual preferences and behavioural history, while low-personalization conditions employed more generic and impersonal language.

The results of Study 1 provide robust support for the hypothesized effects and offer important insights into the mechanisms underlying AI-driven influence. Consistent with H1, the presence of AI nudging significantly increased the likelihood of selecting the sustainable product. This finding aligns with prior research suggesting that algorithmic recommendations can enhance the salience of specific options and reduce the cognitive effort associated with decision-making (Pentina, Xie, Hancock & Bailey, 2023). By foregrounding environmentally friendly alternatives and implicitly endorsing them, AI systems appear capable of nudging consumers toward more responsible choices without restricting their freedom.

However, the analysis also reveals a critical tension associated with personalization. In line with H2, high levels of personalization were associated with a significant decrease in perceived autonomy. Participants exposed to highly personalized recommendations reported feeling less in control of their decisions, suggesting that personalization may blur the boundary between assistance and influence. This finding resonates with emerging concerns in human–AI interaction research, which highlight the potential for algorithmic systems to undermine users’ sense of agency when their operations are perceived as intrusive or overly directive (Puntoni, Reczek, Giesler & Botti, 2021).

Importantly, the relationship between AI nudging and trust was found to be mediated by perceived autonomy, providing support for H3. When consumers felt that their autonomy was compromised, their trust in the system decreased, even if the recommendation itself was perceived as useful or accurate. This result underscores the complex interplay between effectiveness and legitimacy in AI-driven interventions. While AI can guide behaviour in desired directions, its long-term success depends on maintaining a balance between influence and autonomy.

Taken together, the findings of Study 1 highlight the dual nature of AI-driven nudging. On the one hand, it is an effective tool for promoting sustainable choices by simplifying decision-making and enhancing the visibility of desirable options. On the other hand, its effectiveness is contingent upon how it is experienced by consumers, particularly in terms of autonomy and control. These results provide empirical grounding for the Algorithmic Sustainability Paradox by demonstrating that the mechanisms that drive behavioural change may also generate psychological resistance.

4. Study 2: Longitudinal Dynamics and the Emergence of Rebound Effects

While Study 1 establishes the immediate impact of AI nudging, sustainable consumption cannot be adequately understood as a series of isolated decisions. Instead, it unfolds as a dynamic process shaped by repeated interactions, evolving motivations, and contextual feedback. Study 2 was therefore designed to capture this temporal dimension by examining how AI-driven interventions influence behaviour over time, with particular attention to the emergence of rebound effects and behavioural adaptation.

Building on insights from sustainability research and complex adaptive systems (Satornino, Du & Grewal, 2024), this study adopts a longitudinal experimental design in which participants engage in a sequence of consumption decisions across multiple rounds. Each round represents a distinct but related decision context, allowing for the observation of behavioural trajectories rather than static outcomes. This approach reflects real-world consumption patterns, where individuals make repeated choices within evolving environments influenced by prior behaviour and external feedback.

Participants were randomly assigned to either an AI nudging condition or a control condition. In the AI condition, recommendations were dynamically updated based on participants’ previous choices, simulating the adaptive nature of contemporary AI systems. This feature is particularly important, as it captures the recursive feedback loops that characterize real-world algorithmic environments. Participants received information about the environmental impact of their choices, along with cumulative feedback reflecting their overall “sustainability performance.” In the control condition, participants made decisions without algorithmic guidance, providing a baseline for comparison.

The findings reveal a complex and temporally structured pattern of behaviour. Consistent with the results of Study 1, AI nudging significantly increased the likelihood of choosing sustainable options in the initial rounds. This confirms that AI-driven recommendations are effective in promoting pro-environmental behaviour, at least in the short term. The system’s ability to adapt to user behaviour and provide tailored feedback appears to reinforce this effect, creating a sense of progress and alignment with sustainability goals.

However, this initial success does not persist uniformly over time. As participants progress through the decision rounds, the proportion of sustainable choices in the AI condition begins to decline. This pattern suggests the presence of behavioural adaptation, whereby the influence of AI nudging diminishes as users become accustomed to its presence. In contrast, behaviour in the control condition remains relatively stable, indicating that the observed decline is not simply due to fatigue but is specifically linked to the dynamics of AI-driven intervention.

The most striking finding of Study 2 is the emergence of a significant rebound effect. Participants who initially engaged in sustainable behaviour were more likely to shift toward less sustainable choices in later rounds. This phenomenon is consistent with the theory of moral licensing, which posits that individuals who perform morally positive actions may subsequently feel justified in relaxing their standards. In the context of AI-driven nudging, this suggests that the system may inadvertently create a psychological buffer that reduces the perceived need for continued pro-environmental behaviour.

Further analysis confirms that moral licensing mediates the relationship between early sustainable choices and later rebound behaviour. Participants who accumulated higher sustainability scores reported a stronger sense of having “fulfilled their responsibility,” which in turn predicted a greater likelihood of selecting non-sustainable options. This mechanism highlights the importance of considering not only direct behavioural effects but also the underlying psychological processes that shape decision-making over time.

In addition to behavioural changes, the study reveals shifts in perceived autonomy and trust. While participants initially expressed relatively high levels of trust in the AI system, this trust declined over time, particularly among those who became more aware of its influence. Similarly, perceived autonomy decreased as participants experienced repeated exposure to algorithmic guidance. These findings suggest that the long-term effectiveness of AI interventions may be constrained by their impact on user experience and perception.

Overall, Study 2 reinforces the central argument of this research by demonstrating that AI-driven nudging produces both enabling and constraining effects over time. While it can successfully initiate sustainable behaviour, it may also generate unintended consequences that undermine long-term outcomes. These findings underscore the need for a dynamic and process-oriented approach to understanding sustainable consumption in AI-mediated environments, where behavioural trajectories are shaped by complex interactions between technology and human psychology.

5. General Discussion

The findings of this research provide a nuanced understanding of the complex interplay between artificial intelligence and sustainable consumer behaviour. Across both studies, AI-driven interventions demonstrate a dual character, capable of simultaneously promoting and undermining sustainability outcomes. This dynamic underscores the importance of moving beyond simplistic narratives that treat AI either as an unqualified enabler of sustainable consumption or as a neutral instrument devoid of behavioural consequences.

5.1. Theoretical Implications

From a theoretical standpoint, this research advances the literature in three key ways:

1. From Tools to Behavioural Architectures

Traditional consumer behaviour models conceptualize technology as an instrumental tool that aids decision-making. Our findings suggest that AI should instead be viewed as a dynamic behavioural architecture. Through Generative AI Nudging (GAIN), algorithmic systems not only present options but actively shape the structure, salience, and perception of choices, thereby participating in the co-construction of consumer behaviour. This perspective shifts attention from outcomes to processes, highlighting the importance of relational and experiential dimensions in understanding sustainable consumption.

2. Dynamic View of Sustainability

Sustainable behaviour is not static but evolves over time. Study 2 demonstrates that initial gains in sustainable consumption can erode due to behavioural adaptation, rebound effects, and moral licensing. Consequently, interventions must be evaluated longitudinally rather than through snapshots of immediate behavioural responses. This insight integrates complexity science with consumer behaviour theory, emphasizing feedback loops, temporal dependencies, and the emergent properties of AI-mediated environments.

3. Paradox as a Core Mechanism

The Algorithmic Sustainability Paradox (ASP) provides a conceptual framework for understanding the simultaneous positive and negative effects of AI. By highlighting the tension between behavioural effectiveness and perceived autonomy, as well as short-term gains and long-term sustainability, ASP serves as a theoretical lens for future research on technology-mediated behavioural interventions. It underscores the need to consider the ethical, psychological, and systemic dimensions of AI deployment.

5.2. Managerial Implications

From a managerial perspective, the findings offer actionable guidance for organizations seeking to leverage AI for sustainability:

• -Avoid excessive personalization: Highly tailored recommendations, while effective in the short term, may reduce perceived autonomy and erode trust. Firms should balance personalization with transparency and user control.
• -Design for long-term behaviour: Interventions should aim to foster sustained engagement rather than immediate conversions, accounting for moral licensing and potential rebound effects.
• -Integrate ethical constraints: Recommendation algorithms should embed sustainability objectives and user welfare considerations, ensuring alignment with societal goals beyond profit maximization.

5.3. Policy Implications

Policymakers can draw several lessons from this research:

• -Regulate algorithmic transparency: Consumers should have access to clear explanations of how AI recommendations are generated.
• -Monitor environmental impacts: AI systems’ indirect influence on consumption patterns should be assessed alongside direct product-level impacts.
• -Protect consumer autonomy: Regulatory frameworks should safeguard individuals from undue algorithmic influence that may undermine agency, even when the outcome appears desirable.

5.4. Limitations and Future Research

Despite its contributions, this research has limitations:

• Experimental settings: Both studies used simulated environments, which may not fully capture the complexity of real-world consumption.
• Sample composition: Participants were primarily recruited from online panels in developed markets, limiting cross-cultural generalizability.
• Behavioural scope: The studies focused on product choices; future research could ex amine broader consumption behaviours, including services and digital goods.
• Long-term dynamics: Although Study 2 extended over multiple decision rounds, longer longitudinal studies are needed to capture habit formation and sustainability trajectories.

Future research should explore cross-cultural validation, real-world field experiments, and the integration of AI interventions with other behavioural and institutional mechanisms. Moreover, examining interactions between AI, social norms, and collective decision-making may illuminate additional pathways through which algorithmic systems influence sustainable outcomes.

6. Conclusion

Artificial Intelligence represents a transformative force in consumer markets, capable of reshaping the way individuals perceive, evaluate, and act on sustainability considerations. This research demonstrates that AI is neither inherently beneficial nor neutral with respect to sustainable consumption. Through Generative AI Nudging (GAIN), AI can effectively guide choices toward environmentally responsible options, but it can also introduce psychological and behavioural trade-offs, including reduced autonomy, erosion of trust, and rebound effects.

Understanding the Algorithmic Sustainability Paradox is therefore critical for designing interventions that maximize the potential of AI while mitigating its unintended consequences. By adopting a dynamic, process-oriented, and ethically informed perspective, scholars, managers, and policymakers can harness AI not only as a tool for efficiency, but as a mechanism for promoting enduring sustainability transitions. Ultimately, the challenge is to leverage AI’s generative capabilities in ways that support both individual agency and collective environmental goals, ensuring that technological progress translates into genuine and lasting behavioural change.