Surrendering Consumption to Artificial Intelligence: Consumer Agency and Responsible Consumption in AI-Mediated Markets

1. Introduction

Goals 11 and 12 of the United Nations’ Sustainable Development Goals, respectively Sustainable Cities and Communities and Responsible Consumption and Production, have relevance to business planners [1]. At the same time, Artificial Intelligence (AI) shapes all domains of human activities, consumption and production [2]. In turn, the move towards responsible consumption and production shapes social practices [3] for sustainability, making consumption and production more sustainable. AI presents a unique opportunity to reshape consumption and production practices to enhance resource efficiency and sustainability. A practical example is the use of smart meters to optimize the energy efficiency of electro domestic appliances.

The literature on responsible consumption and production has consistently proposed Practice Theory (PT) as a lens to study how human subjects shape their activities (or practices) to reduce the impact of their consumption cf. [4]. A question that concerns research is about the extent to which consumers – as human subjects – have real agency on their consumption activities, as some consumption is invisible and out of human’s control. This is the case with consumption of resources such as heating and energy, of which consumers have little control of who provides them and how [5]. AI can enhance the resource efficiency of human subjects’ activities, and at the same time remove the margins of error of consumer practices, as well as their agency. However, research has recently highlighted that AI can be prone to errors and hallucinations, which could potentially bring about harm to consumers and the environment due to AI’s errors.

The socio-technical development of AI systems vastly outpaces the research into how practices can be monitored and ensure they are “best practices” for the protection of consumers and AI harm mitigation [3]. Nogueira, Lopes [2] characterize the transition to AI’s role in economy and society as an “explosion”, with consequent challenges for policymakers and practitioners to monitor and address relevant issues. There is scant knowledge of what AI means for human behaviour and what AI means for practices performed, or not, by consumers. This is such a crucial issue that the European Union, first in the world, drafted regulations in 2024, with implementation in 2026, in the form of the EU AI Act 2024. This regulation addresses the multi-fold risks of cognitive behavioural manipulation, applications of facial recognition and biometric inferences of social and psychological consumer attributes AI offer consumers and human subjects in general. Furthermore, the act names domains where AI can be particularly consequential for consumers, such as education and healthcare [6]. In the UK, the Digital Markets, Competition and Consumers Act 2024 (DMCC) prohibits "dark patterns" or manipulative designs that use AI to manipulate consumers into unwanted purchases or subscriptions [7]. In contrast, contemporary literature positions consumers as active yet vulnerable participants within AI-mediated consumption systems. Consumers benefit from convenience, personalisation, and decision-support capabilities offered by AI-enabled platforms, while simultaneously being exposed to algorithmic influence, behavioural steering, and psychologically mediated interactions [8,9]. Recent studies suggest that AI systems shape consumers’ preferences, emotions, and sustainable consumption behaviours through engagement mechanisms, digital nudging, and para-social interaction processes between consumers and AI-driven platforms or virtual agents [10]. Therefore, consumers are positioned ambiguously within AI-enabled socio-technical systems, acting both as beneficiaries of enhanced efficiency and as subjects vulnerable to manipulation, reduced autonomy, and digital exclusion [2,11]. One important question is over how consumers can be carriers of resource efficiency practices and how their role in diffusing and reproducing these practices will be challenged if there are no human carriers. Current research says very little on how managers can embed AI in employee practices [12] and, we argue, even to inform consumer practices and policy makers’ oversights for the protection of consumers as it has the potential to disrupt existing practices (Ibid.)

There is then risk that the implementation of AI in consumption practices create disparities and exclusion [2], for example when some types of consumers have little digital confidence, known as the “digital divide” [11].

We answered the call by Sloane and Zakrzewski [3] to refine the conceptual framework of social practice theory to fit the role of AI and Bonetti, Montecchi [12] to investigate how consumers can be engaged in collaboration with AI. The research question that this paper addresses is,

How does AI shape the practices of responsible consumption and production and involve consumers in collaboration between providers and consumers, with the involvement of AI?

The contribution this paper aims to generate is to develop knowledge of how AI can shape responsible consumption and production practices and to propose guidelines to discipline the use of AI in such context and directions for research. The paper is structured as follows: Section 2 revisits the literature on AI and consumption, Section 3 describes our research perspective and Section 4 our methods, Section 5 presents our findings followed by discussion and conclusions.

2. AI and Its Role in Consumption

2.1. AI as a Practice-Shaping Agent

Artificial intelligence (AI) plays a significant role in consumption, by supporting efficient resource management, decision-making and automated service delivery [13]. AI operates through SMART applications, which are AI-powered digital tools that improve the sustainability, accessibility, and efficiency of products and services by using data-driven insights, personalised recommendations and predictive analytics to support sustainable consumption [14].

In domains such as energy and infrastructure, AI systems use real-time data and demand forecasting to coordinate resource consumption with limited direct user intervention, shifting aspects of resource management to automated infrastructures [15]. Within smart grids, these capabilities strengthen demand-response management, balance energy supply and demand and support the integration of renewable energy sources [16]. The transition towards sustainable consumption requires systemic changes in how resources are used and how everyday practices are organised [17]. Therefore, AI is a practice-shaping force embedding competences within algorithms, reconfiguring material arrangements and shifting decision-making from human actors to automated systems.

Digital technology, particularly AI-driven mobile applications, is transforming responsible consumption and production by optimising supply chains, improving waste management systems, and increasing transparency in production processes. Beyond these operational improvements, AI reconfigures businesses towards more sustainable practices [18], while AI-powered gamification encourages consumers to adopt more sustainable behaviours and engage more actively in environmental protection activities.

For example, AI enables transport networks to respond more dynamically to changing patterns of demand through route optimisation, traffic coordination, and multimodal travel planning. This reduces unnecessary resource consumption and environmental impacts [19]. One application is Mobility as a Service (MaaS), a digital platform that integrates transport services to meet users’ mobility needs [20]. AI-enhanced MaaS supports sustainable mobility by offering users a single, optimised interface across transport modes, encouraging a shift from private vehicle ownership towards shared and integrated mobility services [21].

AI can support consumption and production within circular business models, which minimise material use and maximise resource recovery [22]. Examples include Product-as-a-Service (PaaS) and platform-based models, where consumers access products through digital platforms rather than ownership [23,24]. For firms, AI also supports recurring revenue models that extract continued value from products through modular product design that enables reuse [25]. Within these business models, AI uses data analytics and real-time tracking to monitor product use, optimise resource efficiency and enable predictive maintenance by anticipating product failures [26]. In addition, AI supports end-of-life value recovery by sorting and classifying waste for refurbishment, reuse and recycling processes (Ibid.)

2.2. AI in Consumption and Production Practices

Bibliometric evidence identifies personalised interactions, AI-enabled recommender systems, conversational AI, predictive analytics, and virtual assistants as the dominant research themes underpinning this transformation [2]. Recommender systems use AI to learn from consumers’ previous behaviours and preferences to generate personalised recommendations that reduce information overload and enhance decision quality [27]. Within e-commerce, these AI-enabled technologies support consumer decision-making by reducing search effort, improving product relevance, and creating more responsive shopping experiences [28,29].

One of the principal mechanisms through which AI supports consumption is through personalised recommendations and predictive decision support. By analysing consumers’ behaviours, preferences, and previous interactions, AI systems generate tailored recommendations that improve the relevance of products, services, and marketing communications [14,30]. Similarly, AI-personalised recommender systems enhance consumers’ perceived relevance, inspiration, and decision efficiency.

These capabilities rely on extensive behavioural, transactional, and contextual consumer data to generate personalised recommendations and improve decision support, making data collection a foundational component of AI-enabled consumption. While these capabilities increase relevance and convenience, they also depend on consumers’ willingness to share personal information and accept data-driven recommendation processes [31], raising increasingly urgent questions on privacy and trust as AI pervades consumer decision-making [32].

Whilst AI delivers personalised decision support to consumers, firms use the same data and predictive capabilities to optimise marketing strategy and operational decision-making. On the production side, Huang and Rust [33] provide a strategic framework for understanding how AI assist providers, arguing that AI supports marketing research, strategy, and implementation through three complementary forms of intelligence: mechanical AI, which standardises repetitive activities; thinking AI, which supports analytical and strategic decision-making; and feeling AI, which facilitates customer understanding and relationship management. Within this framework, AI strengthens segmentation, targeting, positioning, customer understanding, and personalised marketing actions, while more recent research demonstrates that AI-driven clustering, profiling, and predictive models further enable firms to identify consumer groups more accurately and deliver tailored marketing campaigns [34,35,36].

AI no longer simply helps organisations understand consumer behaviour retrospectively; it enables real-time forecasting and adaptive decision-making. Big data analytics and AI support a wide range of marketing activities, including personalised marketing, customer relationship management, dynamic pricing, product development, and fraud detection, allowing firms to continuously adapt products, prices, and communications to predicted consumer behaviour [37]. As a result, marketing decisions are shifting from periodic, human-led analyses towards continuously optimised, algorithmically supported processes. From a responsible consumption and production perspective, this shift is significant: dynamic pricing, predictive targeting, and continuously optimised marketing are typically designed to maximise conversion and repeat purchasing, objectives that are not inherently aligned with reduced or more sustainable consumption [38,39]. The efficiency gains AI provides to firms therefore do not automatically translate into efficiency gains for the consumption system.

Beyond supporting consumers and firms independently, AI mediates the interaction between them. Conversational AI technologies, including chatbots and virtual assistants, have become integral components of the customer journey by facilitating information search, product evaluation, engagement, and purchasing decisions [40]. Similarly, voice assistants and smart speakers influence shopping behaviour by enabling consumers to complete purchases through natural language interactions [41]. Rather than interacting directly with organisations, consumers engage with AI-enabled interfaces that filter information, recommend alternatives, and shape purchasing decisions, positioning AI as an active intermediary within the consumer journey.

Through these AI-enabled capabilities, consumption is becoming more data-driven, adaptive, and algorithmically supported. While these developments improve consumer experiences and organisational decision-making, they also embed algorithmic systems more deeply into everyday purchasing behaviour—systems whose objectives are commercial rather than sustainability-oriented [38,39]. This growing reliance raises important questions regarding consumer autonomy, trust, transparency, and, ultimately, whether AI-mediated consumption can be reconciled with the goals of responsible consumption and production. These issues provide the foundation for the following discussion of AI, recommender systems, influencers, and consumer agency.

2.3. AI, Influencers/Recommender Systems and Consumer Agency

Building on AI-enabled decision support, modern AI recommender systems have evolved from tools designed to simplify consumer choice into technologies that actively shape how consumers evaluate alternatives and form purchasing decisions. They do so by generating product suggestions that reduce search effort, minimise cognitive load, and improve the relevance of available options [42,43]. AI therefore no longer merely assists consumers in navigating complex marketplaces but plays a growing role in determining which products, services, and combinations thereof consumers consider and how they evaluate them.

Beyond simplifying decision-making, recommender systems function as persuasive mechanisms that shape consumer preferences and purchasing behaviour. They operate as subtle digital nudges, influencing consumers through interfaces, ranking algorithms, and recommendation strategies [44]. Ibid. show that recommender systems can influence consumers’ evaluations through mechanisms such as attribute-level anchoring, whereby even numerical information associated with recommended products affects subsequent product assessments. Similarly, the persuasive effects of AI recommendations depend on how they are presented: collaborative filtering approaches, for example, may encourage information cascade (bandwagon) effects, whereby consumers infer product quality from the apparent choices of others, whereas transparent content-based recommendations appear more effective for products that require subjective evaluation [45]. These findings suggest that recommender systems do not simply reflect existing consumer preferences but participate more actively in shaping them. However, although recommender systems reduce information overload and improve shopping convenience, they may also reduce consumers’ perceived control and raise concerns regarding privacy, transparency, and behavioural manipulation when they become opaque or excessively personalised [46,47].

The effectiveness of AI recommendations depends on consumers’ trust in algorithmic decision-making. Cabrera-Sánchez, Ramos-de-Luna [48] argue that consumers are more likely to adopt and rely on AI recommendations when they perceive the system as trustworthy and useful in helping them accomplish their tasks more efficiently. However, increasing reliance on AI recommendations is not universally beneficial. Indeed, AI recommendations shift consumer decision-making from autonomous information search towards algorithmically guided choice.

In addition to recommender systems, AI influences consumer decision-making through AI-generated influencers, which represent a more direct form of algorithmically mediated persuasion. Unlike recommendation algorithms that shape decisions by filtering and prioritising information, AI-generated influencers interact with consumers through social media content, product endorsements, and ongoing digital engagement, enabling AI to influence not only product choice but also consumers’ perceptions of brands [49,50]. Therefore, AI assumes a more active communicative role within the consumer decision-making process, extending beyond recommendation, towards relationship-building and persuasive interaction.

The effectiveness of AI-generated influencers depends on consumers’ trust and perceptions of authenticity. Although virtual influencers can successfully generate engagement and influence consumer decisions, their ability to shape purchasing behaviour is linked to whether they are perceived as credible, authentic, and consistent in their virtual identity [49,51]. Research further suggests that AI-generated influencers are more persuasive when they maintain a coherent and transparent digital persona that consumers perceive as authentic [52].

There is conflicting evidence of whether consumers trust AI-generated influencers to the same extent as human influencers, despite their achieving similar levels of consumer engagement [53]. In contrast, other research suggests that virtual influencers may be viewed as more objective, credible, and knowledgeable in certain contexts, thereby increasing consumers’ willingness to act upon their recommendations [54].

Nevertheless, trust appears to emerge from a combination of perceived expertise, authenticity, attractiveness, similarity and the congruence between the influencer, the endorsed product, and consumers’ own values [55]. Consumers tend to respond positively to AI-driven personalisation when it feels accurate and beneficial, but this positive response is conditional on consumers’ perception that their autonomy is respected rather than overridden [56]. In digital marketing contexts, perceived intrusiveness can increase scepticism and reduce engagement [57], while privacy concerns and psychological reactance are similarly associated with advertising avoidance [58]. This sensitivity to autonomy is closely tied to broader concerns regarding data handling and fairness. When consumers perceive that their personal data may be misused through deceptive marketing practices, their trust in AI-driven marketing may decline [59].

AI reshapes consumers’ information search, alternatives evaluation and decisions. Although algorithmic assistance can reduce cognitive effort and improve decision efficiency, its implications for consumer agency are not uniformly positive. Consumers may be willing to give up some control in exchange for convenience, particularly when purchasing routine, low-involvement products [60]. However, algorithms may also narrow the alternatives consumers encounter, reinforce existing preferences, and risk weakening their capacity for independent judgement [61]. Importantly, the relationship between AI autonomy and consumer responses appears to be non-linear: moderate algorithmic involvement may support consumer self-efficacy, whereas excessive decisional control can undermine it [62]. Therefore, AI-mediated consumption involves tension between empowerment and replacement, as well as between personalisation and exploitation [63]. Its effects depend on the nature of the decision, how much responsibility is delegated to technology, and whether consumers retain meaningful control over the choices presented to them. This is particularly important for responsible consumption and production, as consumer agency may influence whether AI encourages deliberate, sustainability-oriented choices or further reinforces convenience-driven patterns of consumption.

2.4. Challenges and Implications for Sustainable Consumption

AI can support SDG 12 by making production and resource management more precise. In manufacturing, AI can identify process inefficiencies, optimise machine settings and support predictive maintenance, reducing material losses, energy use and avoidable downtime [64]. These capabilities can also support sustainable business models by improving knowledge management and incorporating resource considerations into organisational decisions [65]. At a broader level, Wang, Li [66] find that AI development is associated with smaller ecological footprints and carbon emissions and with progress in the energy transition. However, operational improvements do not make AI inherently sustainable. Hasan and Ojala [67] argue that AI contributes to responsible production and consumption only when supported by appropriate management strategies, policies and institutions. AI’s contribution to sustainability depends on both what makes it more efficient and the objectives of the wider system in which that efficiency is pursued.

AI-driven personalisation also has the potential to intensify consumption by lowering the informational and cognitive effort associated with purchasing decisions. Evidence from the functional-food sector, for example, indicates that AI-generated recommendations can strengthen purchase intention [36]. The wider concern arises when personalisation is optimised primarily for engagement, conversion and sales. Under these conditions, AI can move beyond matching existing consumer needs and contribute to the creation of new demand [68,69]. Digital services that reduce time, effort or monetary cost of consumption may encourage greater use of the same service or allow savings to be spent on other resource-intensive activities [70]. Economy-wide evidence similarly shows that improvements in energy efficiency are often partly offset by behavioural and economic responses that increase overall demand [71]. Therefore, improvements in the efficiency of recommendation, production, and distribution do not necessarily translate into lower environmental impacts, if they simultaneously stimulate greater volumes or frequencies of consumption. At the same time, these developments embed algorithmic systems whose objectives are typically commercial rather than sustainability-oriented into everyday purchasing behaviour [38,39].

As AI becomes increasingly embedded in consumer decision-making, the associated ethical, technical, and regulatory risks also become more significant. The EU AI Act [6] suggests that delegating consumer agency to AI in purchasing decisions presents risks for both consumers and suppliers. The AI Act reflects these concerns by classifying AI applications into minimal, limited, high, and unacceptable risk categories. These risks include reduced transparency, diminished consumer autonomy, commercially driven manipulation, and technical vulnerabilities. One example of the latter is prompt injection, where attackers feed malicious instructions to AI Large Language Models (LLM) that fall outside the control of its original developer, potentially leaving consumers on the receiving end of damaging actions [72]. Businesses nonetheless remain responsible for the actions of AI systems they deploy, whether through employees, chatbots, pricing algorithms, or recommendation engines [73]. This highlights the need for clearer guidance on what constitutes high and unacceptable risk in practice. More broadly, the growing reliance on data analytics and algorithms raises important questions regarding consumer autonomy, trust, and transparency. Therefore, whether AI-mediated consumption can be reconciled with the goals of responsible consumption and production requires further in-depth investigation. The next section conceptualises practices within AI-enabled consumption.

3. Conceptual Framework: Practice Theory

Practice Theory (PT) is a cultural theory [74], which focuses on social practices as the main unit of analysis [75]. A practice is a “routinized type of behaviour which consists of several elements, interconnected to one another: forms of bodily and mental activities, “things” and their use, a background knowledge in the form of understanding, know how, states of emotions and emotional knowledge” [74].

To simplify with Shove, Pantzar [75], constituent elements of practices include materials, used to perform practices, competences, know-how human possess which enable human subjects to operate on materials and meanings, the value human subjects attribute to consumed products or services. With AI being involved in consumption, we propose a framework adapted by us from Shove, Pantzar [75]’s:

• Materials – hardware such as laptops, smartphones and apps, interfaces with human subjects (e.g. computer and smartphone screens), vehicles and more.
• Competences – algorithms, recommender systems, new digital literacies and confidence, reliance patterns, delegation to AI, all these competences may reside in AI instead of in human subjects.
• Meanings – trust in automation, expectations of personalization, norms of convenience, simplified life, saved effort and time.

In this AI practice framework, what is called materials can be digital, e.g. apps and other software, and it may be hosted online, e.g. in the cloud. As for competences, situadness, i.e. the location of the competences, counts. In traditional societies, competences can be stored in the mind of human subjects or committed to printed or digital media. In the case of AI shaped practices, the competences – the “know how” – may be stored in AI algorithms because of delegation to AI. In conventional PT, human subjects are the “carriers” of practices. Therefore, practices are performed and reproduced through humans. In the case of AI driven practices, where consumption practices rely on human-human interaction, such as when a salesman advises a client, conversational agents establish the conversation and interact (e.g. advise) clients. AI agents become the carrier of practices, which reproduce through AI agency. For example, a consumer can order food online, with AI agents making decisions on the delivery of the food and managing the transaction and payment.

Social Practices are collective (social in nature), recursive (they are performed multiple times) and have scale [76]. In AI shaped practices, these characteristics are due to the diffusion of AI agents supporting consumption, e.g. a travel booking agent. Further, practices are connected and dependent on each other, for example, practices linked with work and the practice of shopping depend on mobility practices [77]. Indeed, practices are aggregated in practice constellations [78], a term first used by Schatzki [79 170]. Constellations are larger interdependent bundles of practices and material arrangements (Ibid.), including distributed activities, ends, purposes (meanings in Shove, Pantzar [75], terms), rules and material arrangements, including infrastructure, which can be encompassed in the term sociotechnical landscape. In line with Shove, Pantzar [75] and Watson [77], we borrow the latter term from Geels [80], who defines it as external structures of the context of society, e.g., material and spatial arrangements of cities, factories, highways, electricity infrastructures and heterogeneous factors such as economic growth, wars, demographics, political coalitions, cultural values and environmental problems, which shape actors’ interactions [81]. To wit, practices aggregate in systems [75] or “bundles” of interdependent practices [82]. For example, the practice of conducting management meetings online depends on the practice of using digital technologies. Finally, practices are dynamic (they change in time), and their constitutive elements may be tightly or loosely coupled [83].

There is little research on how AI payment apps shape consumption practices. Bai [84] claimed that AI apps facilitate online purchases by consumers and encourage consumption (but only focusing on the acquisition stage and not use, which goes against Shove, Pantzar [75]. One conclusion could be that AI payment apps encourage conspicuous consumption. The next step presents the methods used.

4. Materials and Methods

This research drew on a case study strategy to investigate how AI shapes responsible production and consumption processes. Following Thomas [85], a multiple case study approach was adopted, to aid the robustness of the methods and to explore two different exemplars, one where the practices performed by consumers are shaped by AI and the other where AI performs the consumption practices directly. Two case studies, Mobility as a Service (MaaS) and Ant Forest (AF), are described in section 5. Case study research enables study of phenomena within a given context [86] and encompasses diverse methods, including secondary data and collection of primary qualitative and quantitative data. This study is supported by secondary data.

The MaaS case study draws on secondary evidence and 34 interviews of users of shared mobility offerings akin to MaaS. Since this data was collected by one of the authors for another purpose, for a study that was then published, the authors took the view that this should be considered secondary data. Participants in that study consisted of a convenience sample of Hertfordshire (UK) based students and staff at a UK university and parents of members of a scout club, augmented by snowball recruitment. Table 1 summarises the characteristics of the interview participants. The “P” column is the progressive number of participants, used to identify them in the findings.

The original interview guide featured questions about virtual materials (e.g. apps), and the meanings, routines and associations the users connected with their transport behaviour. This makes the data suitable for this study.

Documents as type of the secondary data source include popular culture sources (e.g. magazines), personal sources (e.g. blogs) and public sources (e.g. government reports) (Largan and Morris, 2019). The Ant Forest case study draws on secondary data from various sources such as fora on Zhihu and LinkedIn, News (e.g. BBC, Xinhua News), comments from social media (Webo, Rednote), and reports (Ant Group), with a strategy similar to netnography, where users’ comments accessible through fora are accessed and analysed [87]. Examples of secondary data accessed can be seen in Table 2. This enabled the researchers to access comments by multiple people (some in Chinese), who commented on the competences they (or AI) use, the materials and what all means to them, when they perform consumption practices. The authors benefited from one of them being of Chinese ancestry, which was empowering in this project as AF is a feature of Alipay, a Chinese payment app [88], part of the Ant Group, which belongs to the Chinese Alibaba Group [89]. Alipay is used by many consumers of Chinese extraction, and their utterances are often in Chinese. The user comments were translated where necessary and pasted into Word files, then thematic analysis was conducted to develop themes and identify patterns to generate new insights and understandings. To ensure credibility, this study applied the RIPES model (Reflexivity, Interpretation, Procedural consistency, Evaluation and Situatedness) [90]. The interview transcripts and the diverse qualitative data were coded in NVivo, a qualitative analysis software package [91], based on a flexible template [92]. Table 2 summarized the data sources used for Ant Forest.

5. Case Studies

To illustrate how AI shapes consumption, we use the two case studies mentioned in section 4. Ant Forest (AF) is a pro-environmental feature within Alipay app launched in 2016 by Ant Financial Services Group, an Alibaba affiliate in China, using AI technology to enhance the public engagement to sustainable behaviour. It has over 690 million users and results in 475 million trees being planted and millions of tons of carbon emission reduced [93]. The other case study is Mobility as a Service (MaaS) increasingly governed by AI, which shapes mobility practices and consumption of transportation means.

5.1. Ant Forest

Gamification is “the process of making activities more game-like”, involving users through competitions and rewards such as points, badges and leaderboards to shape their actions to support overall value creation [94]. AF gamifies sustainable behaviour to the public through AI technology stimuli, which trigger significant changes in social practices [95]. A comment from LinkedIn post noted:

"Alipay is used by tens of millions of people in China, and through this Ant Forest, a great thing can involve millions of people, which is both ordinary and great!"

(Post ID #2)

Consumers can access AF on Alipay through smartphones and tablets. AF is user-friendly and facilitates seamless operation. AF encourages consumers to use public transportation such as walking and cycling, reuse and recycle goods, paying utility fees and purchase online, going paperless, buying sustainable products, renting shared bicycles and interacting with friends via the app. The instances of sustainable behaviour are accumulated and calculated into green energy points to nurture sapling, accumulate and swap energy for planting real trees. In contrast, the users fund the collection of point via AF is time-consuming as they must log-in in a limited time window. The AI assistant "Ah Bao" in Alipay Apps helps to collect points, which reduce the amount of work users need for using AF. Additionally, the feature allowing friends to help to harvest the energy makes AF convenient and functionable. As this quote shows,

"I would rely on the AI assistant to collect green energy so I could use the saved time to wash up when getting up in the morning."

(Post ID #26)

"You can help your network with collecting the energies if they did not visit Ant Forest for a while (...). Also, it will notify your friend about your help. It’s a very clever decision to keep people involved in it."

(Post ID #1)

Competences Acquisition Through AF

Challenges constraining public engagement with sustainability include cognitive bias and lack of knowledge [96]. Different low-carbon behaviour will generate different virtual green energy points which can be transformed into various species of trees. AF offers individualized experience via recommender systems, enabling users to decide when and where to plant which tree species. During the process, users acquire relevant sustainability knowledge (e.g. low-carbon behaviour and footprints, desertification, reforestation, and ecological restoration). In addition, through the Amazing Species project, users discover rare new species. A Weibo post states:

"I have participated to AF for 3,400 days and I chose to plant the Hedysarum scoparium in Alxa in Inner Mongolia. It has the functions of sand fixation and soil improvement".

(Post ID #35)

A comment from RedNote shows,

"In the past, few people knew or cared about the Aral Sea, and few people knew about the Prosopis Juliflora tree. Now, the situation is (…) different using AF."

(Post ID #25)

Collaborating with Jifei technology, AF allows users to gain sustainable consumption competences situated in algorithms which monitor forest growth rate, calculate carbon footprint, and adjust green energy needed for trees [27]. Users reported,

"With technologies such as artificial intelligence (…), users can see directly the real impact of their actions. To see sandy plains transforming into forests." (Post ID #1) and “When I see the tree I planted via the satellite map, I realise that there will be big difference from changing low-carbon lifestyle". (Post ID #16)

Acquisition of Awareness by Consumers

AF has been awarded a Champions of the Earth award, the UN’s highest environmental honour, recognising its contributions to plant around 122 million trees and highlighting how AF leverages technology to inspire consumers to behave in a more sustainable manner [97]. As this comment illustrates,

"What wakes me up every day is not dreams, but Ant Forest! Every day at 7 o’clock, I set the alarm and get up on time to collect energy, steal energy... I now have 20 environmental certificates."

(Post ID #10)

"I walk and bike more now. The game’s incentives make walking feel more purposeful… now, I assess how many green points I can get wherever I go"

(Post ID #14).

Controversies of AF

Despite the positives, there are concerns with authenticity and trust of whether AF encourages adoption of sustainable practices. Users complain that they spend much time in playing various games advertised by Alipay when using AF. As Zhihu shows,

"With the popularity of the project, there have been many behaviours of walking and swiping orders online and offline, which (…) violates the original intention of the (…) design and affects the confidence of the participants in the transparency and fairness of the project".

(Post ID #17)

Users can buy green energy from black market such as Xianyu which is China’s largest consumer-to-consumer community and marketplace for idle goods. Although AF has responded to this by blocking 26,000 accounts that violated their terms [98], its fairness and integrity are suspicious, as illustrated by the quote,

"Can you believe that the gifts awarded to the top 5 in the leaderboard are being sold on Xianyu for 188 RMB?"

(Post ID #18)

Consumers can also generate points by watching ads and scrolling the Alipay’s shopping pages without adopting low-carbon behaviour. Additionally, AF held player killing competition to foster social impact. Some users find that this is entertaining, but others feel unfair for loss of green energy if they fail the game. Examples of RedNote post:

"The AF should stop doing P.K. competitions and it’s too time-consuming." (Post ID #33) and "I feel too anxious about the P.K. competitions and will not use them. AF makes the competition too fierce." (Post ID #24)

Despite these controversies, users of AF are motivated by themes such as sustainability, altruism, enjoyment, entertainment, social ranking, personal achievement and satisfaction. This means that, assuming that AF is trustworthy, consumers retain their agency and decision autonomy.

5.2. Mobility as a Service

Within a full MaaS, AI can reduce the amount of work users need to perform to use MaaS apps, which is a significant barrier to adoption. Durand, Harms [99] suggest that use of basic MaaS challenges users with the amount of work and level of skill needed to use several types of shared mobility. Participants in the study reported a strain from cumulative work needed to acquire competences to manage mobility apps which encompass booking, ticketing and journey planning. Even the management of apps themselves is a challenge as this citation illustrates,

“… you tend to go back to the apps you’ve already got and (make them) work rather than (download another). (…). You know, spend time working out with the [app] you can connect to the (…) mobile data and [must input new registration information (…) to make it work”

(P26).

The number of smartphone apps required to use MaaS can be overwhelming and induce apps fatigue, where users suffer from "overapp" [100]. The entry of additional personal information in various databases is a deterrent, because of privacy and long-term commitment concerns. Means of transport such as car clubs and other vehicle-sharing apps are perceived as complicated as this comment illustrates,

“… there’s a load of admin that comes with it and a lot of associated cost.”

(P7)

Therefore, the labour-intensive nature of MaaS and the complexity of using MaaS apps may be a deterrent [101]. Participants said that journey planning is cumbersome, as they noted,

"The only thing that would put me off is if, it was too complicated, (…) a mishmash of train booking and scheduling, club cars [and other steps]".

(P30)

All this suggests that consuming MaaS requires novel competences compared to just using one’s own car. It is no longer simply a matter of getting into a car and driving, many ICT and administration skills are involved, requiring elevated levels of digital confidence.

This is where AI may support users in consuming MaaS. As seen in sections 2 and 3, competences which are part of consumption can be stored in algorithms and recommender systems. This means that AI may integrate competences consumers need to use MaaS apps and can facilitate MaaS consumption.

A key contribution of AI may lie in enhancing users’ perception of safety. AI has potential to enhance users’ confidence and sense of control by enabling them to access information that helps them feel safe. Users can navigate the landscape and track and to be tracked by associates, check where vehicles are and how crowded they are.

AI embodies competences to match the provision available and the landscape of an area, as well as the identity and professional details of human drivers of manned vehicles. Through smartphone apps AI grants users’ access to ID, service-quality rankings and other quality assurance information that certifies drivers’ trustworthiness. Services such as Uber may be tailored:

“… the whole point is that [AI may] send you a picture of John, you know, in his Toyota Prius 12345, […] give you the registration number, (…) (his) picture, so you know the driver’s name and even their phone number”.

(P11)

Participants claimed they used MaaS apps to manage their costs. A participant reported:

“… even Google Maps now flashes [up] how much an Uber would be, not that I trust the price, (…) I’d always go and double-check.”

(P23)

So, AI also helps users manage the costs of a journey, acting like a butler. AI is connected to safety hardware, so apps enhance perceived safety as this citation illustrates,

“Most of the stations have got CCTV, which I think’s good. Some trains have got it.”

(P9)

Cameras and other security hardware can be placed in vehicles and in the landscape, for example inside and around trains and docking stations. AI provides feelings of control through information,

"...it gives me a real sense of security that I know I can just pull out my phone and get Google Maps up or TfL (Transport for London) to see when the next or the last bus or train (…) is going to take me home"; “… you know that the app will at least give you a route or a way to (…) get home and that’s (…) comforting,

Specialist AI also helps: one participant stated that she had "Find my Friend” on her phone. AI can keep users informed about the correct stops, routes and transfers, and is empowering. Participants mentioned that they used apps to see what areas they would need to travel through and whether these were dangerous areas, thus enhancing safety.

Despite these potential benefits, participants identified several factors that may constraint trust in an AI-enabled MaaS. Our data suggests that human subjects prefer being supported by other humans rather than AI and apps. Some participants said that they found the presence of driver and service people on board, such as the provider’s staff, reassuring – especially if these staff members are safety trained and vetted.

Some participants reported that they did not see enough security staff on services or at stations, and they would like there to be more. This suggests that AI cannot replace human service staff. Some participants claimed that they would rather walk than use MaaS. AI driven MaaS apps can be misleading, because they might give users,

“a route that goes through a dark alley or a park instead of a well-lit road simply because it is the shortest route…” (P26) and “most annoying wasn’t like, don’t go this way on the path. There was a nice path along the side of the A road that just literally came to a stop…” (P26).

i.e. there can be a lack of correspondence between the maps on the system and the hard landscape. This error could be made by AI as it would draw on the same data that basic MaaS apps use and users could attribute this risk to AI, such as the well-known MaaS hallucinations, which can affect trust.

Importantly, AI depends on energy, as it would have to interface with users through smartphones, which can run out of battery. Some users said they had “battery charge anxiety,” and feared the failure of a MaaS app could leave them stranded in a remote area. To manage travel, AI would also need to interface with consumers’ bank details, which could cause glitches. As illustration, a participant shared that during use, the MaaS app would not connect with her bank details and the payment failed, which meant she could risk a fine for not having a ticket on board a service. Most participants cited connections between modes, such as switching from a train to a bicycle, as the least safe legs of their journey. This would shape users’ expectations and trust in AI in acting as an integrator of legs of journeys managed by different providers.

AI-enabled smartphone applications may also be perceived as intrusive, raising concerns about privacy and surveillance. A participant illustrated this by mentioning her suspicions and said that she worried about being tracked. In summary, using an AI driven MaaS would involve confidence, reliance patterns and delegation to AI.

From a PT perspective, users are engaged in a complex network of interlinked practices. Participants described mobility as being embedded within multiple interconnected practices, including shopping, caring responsibilities and leisure activities. Consequently, AI would need to coordinate not only multiple transport providers but also users’ complex daily routines. This presents a considerable technical challenge and raises concerns that increasing reliance on AI could reduce users’ agency in planning and adapting their journeys.

6. Discussion

This paper has explored how Practice Theory can be adapted to integrate AI and how AI can shape responsible consumption and production practices. The analysis shows that AI can indeed be beneficial for sustainability as in the case of the operation of domestic appliances such as washing machines, which smart meters automatically set on 30 degrees wash programs.

AI offers opportunities for consumers to learn and establish routines and practices to aid responsible consumption and production. In “traditional” Practice Theory, human subjects perform practices based on competences and meanings that are rooted in their social context or codified in media. For example, in the study of food practices by Halkier and Jensen [102], cooking practices such as selecting which specific type of cooking oil and in what quantity integrates competences (e.g. type of oil needed to cook a meal) and meanings (e.g. a type of oil may be unhealthy), which are stored in cultural resources human subjects share in or, possibly, these competences and meanings may be codified in media such as cooking books. In contrast, AI stores competences and meanings but as described in the case study, the use of gamification in AF encourages users to transfer competences and meanings in their heads, whilst they develop new routines. So, AI can have positive educational outcomes.

In the case of AF, the app shapes consumer behaviour. Existing studies have confirmed AF’s contributions to customers low-carbon purchasing practices (e.g., Ding et al., 2025; Huang et al., 2023; Wang & Yao, 2020). In contrast, in the case of MaaS, the need to decide is removed. Indeed, AI driven MaaS can facilitate travel by being able to select the most resource efficient routes and means of transport autonomously.

On the negative side, AI is prone to hallucinations and errors, so there are risks that decisions conveyed to consumers are wrong, e.g. outdated maps in the case of MaaS. In the case of AF and other apps based on gamification, there is risk of malevolent content injection [72], where actors with special interests may introduce data in the AI framework which favour products which are not necessarily sustainable. Importantly, in transport, AI hallucinations may be caused by human error as human actors fail to update maps.

A further risk is misbehaviour by human users using gamified apps such as AF, where they use different devices to report their consumption of green energy and state false performance to get rewards. There is also a danger that gamification is seen just as entertainment and so the benefit of achieving sustainable practices fail [94]. Finally, AI driven apps may challenge users when they have little digital confidence and skills referred to as the digital divide [11]. Furthermore, some may be concerned with privacy [100]

A theoretical issue for concern is related to who are the carriers of practices – here the practices that shape consumption – e.g. using a product in the correct way – may stop residing within human subjects. Therefore, humans might not be the carriers of these practices at all. This raises the question of who is responsible for the reproduction of practices – here this task might be housed within AI agents. These practices will be faithfully reproduced, but human subjects may lose these capabilities and the ability to learn them, so if AI fails, this might result in the fading [103] of practices – and service failure. Therefore, a consequence of using AI may be that some competences may fade from human subject’s mind, so the ability to consume may rely on the functioning of AI agents and the hardware to interface users and on users’ digital confidence and trust in AI. For example, using MaaS apps may mean that people lose the ability to use physical maps. Because of all the above reasons, legislation, regulation and professional practice are essential to protect consumers and suppliers alike.

7. Conclusions

This paper’s contribution to theory is to have formulated a version of Practice Theory which is adapted to AI, and we contribute to the literature on responsible consumption and production driven by AI from a perspective integrating AI and PT. With business models such as MaaS and platform-based business models – where consumption practices involve the use of an AI governed platform (e.g. the delivery of print cartridges or food) and Product as a Service, AI can encourage consumers to develop new competences or govern consumption directly. On the other hand, AI will increasingly govern the exchange, and consumers might increasingly lose control.

An implication for policy is that consumers need to be educated about the limitations of AI, and they should be deterred from “cheating.” At the same time, further legislation is needed to protect consumers.

An implication for practice is that gamification apps need to be designed so that they are fair and transparent. AI shaped MaaS apps need to be updated to avoid that users are routed to stations and locations that do not exist. The concerns regarding privacy, connected with consumer autonomy which is affected by individuals having limited visibility and control need to be addressed, and so is transparency on how AI systems collect, process and utilize personal information.

The limitation of our research is that since this is a conceptual paper, we relied on secondary data, which used interview transcripts generated for previous research on MaaS, and publicly available consumer data for AF. Future research should address this limitation by working with live data.

Directions for research are on whether consumers trust AI and how they can be trained to be able to give their informed consent. At the same time, research should investigate how AI can be transparent on what data it uses and how it uses it. Furthermore, the possibility exists that consumers see AI as a soft target for misrepresenting their purchasing intentions. An important direction for research is over the different workings of AI applications in diverse cultural contexts. For example, adaptations AF, a Chinese app, would need to be used in the West.