Artificial intelligence and trust in business-to-business interactions: from decision tools to generative and reasoning AI Open Access

The actor-resource-activity (ARA) model (Håkansson and Snehota, 1995) is a cornerstone for understanding interactions in business-to-business (B2B) contexts according to the industrial marketing and purchasing (IMP) perspective. It is grounded in the ties, links, and bonds that create lasting yet dynamic connections between firms (Gadde and Mattsson, 1987). The model outlines the dimensions that bind firms together: resource ties describe how a firm’s tangible and intangible resources are interconnected and deployed towards counterparts with adaptation, investment and continuity in mind (Baraldi et al., 2012). Activity links refer to the co-creative processes and the extent to which these activities are integrated across firm boundaries. They raise considerations about value creation beyond the dyadic business relationship (Dubois, 1998). Actor bonds describe how representatives of firms link companies together.

By focusing specifically on the actors as the social dimension of firm-to-firm connections (cf. Bondeli et al., 2018), we understand these bonds as interwoven with trusting, liking and preferring. These are the non-economic rationales that underpin the establishment of lasting connections between firms (Ford, 1980).

Trust is a concept extensively elaborated across various literatures (e.g. Mayer et al., 1995; Rousseau et al., 1998; Huemer, 2004). When trust is related to actor bonds within the IMP perspective, it is typically grounded in human-to-human interaction, where it serves to keep companies together, enables deeper engagement in repeated exchanges and supports joint development projects (Ford and Håkansson, 2006a, 2006b). Researchers such as Havila and Wilkinson (2002) and Huemer (1998) have shown how trust allows episodes of exchanges and long-term relationship interruption to be tied together over time through human memory and social continuation.

With artificial intelligence (AI) being increasingly discussed and adopted by businesses in relation to, for instance, customer relationship management (CRM) (e.g. Dwivedi and Wang, 2022; Gaczek et al., 2023), and with its rapid technological development (Öberg et al., 2024; Öberg, 2026), it is imperative to understand how AI affects business exchanges. This is so not only for AI as a decision tool operated by humans (Goto, 2022), but also when exchanges occur between humans and machines, are intermediated by machines or are entirely taken over by machines (Paschen et al., 2020; Geissinger et al., 2021a, 2021b; Wu et al., 2023).

If we understand trust as the belief a human places in another person or organisation, trust without human actors would either result in situations of absent trust where exchanges are driven purely by economic considerations, or require mechanisms such as transparency to serve as substitutes for trust (e.g. Hawlitschek et al., 2018). However, we know very little about this, raising questions about the nature and role of trust in AI-mediated exchanges. Formulated differently: what becomes of trust when AI replaces human actors on one or both sides of B2B relationships? This paper investigates how AI alters trust when machines substitute actors within B2B relationships.

AI encompasses a wide spectrum of applications. This paper focuses on three types of AI: AI as a digital decision-support system, generative AI and reasoning AI (Dwivedi et al., 2023; Kokkinakis and Drikakis, 2025; Nvidia, 2025). These types are selected because they reflect the past, present and anticipated future development of AI. Historically, what has been described as AI has largely involved computers processing large volumes of data to support human decision-making. More recently, generative AI has fundamentally reshaped the notion of AI, while reasoning AI has begun to emerge on the horizon.

Through juxtaposing these three types of AI with human-to-machine, machine-to-machine and machine-intermediated interactions, this paper poses the questions:

This conceptual paper is grounded in an integrative literature review (Torraco, 2016), which synthesises arguments across multiple bodies of literature to generate new insights in a forward-oriented discussion (Berthon et al., 2003; MacInnis, 2011; Jaakkola, 2020). It adopts a typology-like approach in its conceptualisation (Jaakkola, 2020) through juxtaposing across types of AI and dimensions of machine mediation.

This paper demonstrates how much existing B2B research continues to focus primarily on AI as a decision-support system (cf. Chen et al., 2022). Through its integrative and forward-oriented conceptualisation, the paper broadens the scope of discussion by incorporating generative and reasoning AI into the analysis of machine mediations in B2B interaction. The paper thereby critically assesses the current state of research and integrates it across disciplinary boundaries to extend present understandings (Berthon et al., 2003).

The paper’s practical and academic significance lies in addressing a contemporary and future-oriented phenomenon, raising questions not only about how emerging business practices fit into established models, but also about how they challenge and transform foundational understandings of B2B interaction (Ford and Håkansson, 2006a, 2006b). The paper thereby contributes to prior research by linking trust in B2B interactions with emerging technological contexts, emphasising the actor dimension of the ARA model and the shifting role of trust among humans, machines and their various modes of interaction. It offers an early and thought-provoking conceptual contribution to the integration of machine learning, machine interaction and AI within the B2B field, and through its typology-orientation helps to structure different notions of AI and machine involvement related to trust.

Following this introduction, the paper briefly outlines the actor bond dimension in B2B interaction, with particular attention to trust. It then focuses on the role of AI in B2B interaction to position its contribution within the existing literature. The method section describes how the conceptual argument is developed through the integrative literature review and logic-based reasoning across disciplinary fields. The paper concludes with findings and a future-oriented discussion, followed by final remarks addressing theoretical and practical implications, along with suggestions for future research.

The ARA model thus illustrates that interactions between firms are not solely driven by short-term profit maximisation. Instead, B2B interactions are characterised by long-term orientation, mutuality, interdependence and forward-oriented investment (Håkansson, 1982; Runfola and Monteverde, 2023). The model conceptualises firm interactions as comprising actor bonds, resource ties and activity links (Håkansson and Snehota, 1989; Gadde et al., 2003). Resource ties thus describe how firms’ resources are interconnected with adaptation, investment and continuity in mind (Baraldi et al., 2012). Activity links refer to the co-creative processes across firm boundaries (Dubois, 1998), while actor bonds capture how human representatives link companies together and constitute the social, non-economic dimension of exchanges.

IMP research has extensively studied these aspects, either by focusing on individual dimensions or by emphasising the ARA model as a framework for understanding their interconnections (Harrison and Prenkert, 2009; Raskovic, 2015; Fortezza et al., 2021; Brettmo and Browne, 2024; Baptista and Nunes, 2025). Among these, actor bonds remain the least explored (Huemer, 2012; Bondeli et al., 2018), and are often associated with managerial cognition (Öberg et al., 2007; Corsaro et al., 2011; Geissinger et al., 2021a, 2021b). Yet, they also thus highlight how firms sustain relationships that extend beyond economically rational considerations (cf. Granovetter, 1985). Human touchpoints not only maintain these relationships but also enable adaptation and long-term value creation that would be unattainable if constrained by short-term profitability. Consequently, the social, actor-bond dimension of B2B relationships holds substantial significance.

Trust is interwoven across the actor bonds, resource ties and activity links through investment and future-oriented intentions (Lenney and Easton, 2007). It primarily manifests through human sensemaking regarding the counterpart (Fulmer and Gelfand, 2012; van Zeeland-van der Holst and Henseler, 2018; Krueger and Meyer-Lindenberg, 2019), and thereby foremost via the actor bonds (Huemer, 2004; Finch et al., 2010). Huemer (2004) proposed that trust may be mobilised either through established norms or by accepting an extended degree of freedom on the other party. Once established, trust can largely replace formal contracts, as long as it remains intact (Roxenhall and Ghauri, 2004). The causal claim can be articulated as follows: actors place trust in others, which leads to investments across resources and activities (cf. Meqdadi et al., 2017), while trust in a counterpart may, in turn, depend on that party’s related investments. Individuals may extend trust to objects and actions (Mayer et al., 1995), but objects cannot reciprocate; they require human guidance for trust creation, maintenance and development. Trust, liking and preferring are therefore closely linked to the human and non-economic facets of exchange, making the replacement of humans by machines a delicate matter.

Prior research has examined the emergence of trust in developing B2B relationships, including pre-existing trust, signalling mechanisms and restoring damaged relationships (Denize and Young, 2007; Tähtinen and Blois, 2011; Mandják et al., 2015). Less is however known about what occurs when trust-upholding actors, who provide reassurance, knowledge and investment decisions, are replaced by non-human actors.

Emphasising the notion of “intelligence”, AI conceptually refers to machines’ ability to mimic human cognitive functions (Russell and Norvig, 2021). It has also been defined more pragmatically as the application of data-driven power to support decision-making and analysis (e.g. Beynon et al., 2000; Aytug et al., 2020; Guercini, 2023). As AI has entered public discourse and undergone rapid development, it has come to embody a wide array of meanings, which in the present paper are discussed as three types: AI as decision tool, generative and reasoning AI, following a chronological development while each type is still or emerging in use.

Starting with AI as decision-making, it involves how machines process vast volumes of data for analytical purposes and for managerial execution (Sahoo et al., 2024). In marketing, this includes CRM systems and more interactively chatbots that respond based on pre-set answers (Lin et al., 2022; Rusthollkarhu et al., 2022). The “intelligence” of such tools has long been debated, particularly in relation to their ability or lack of ability to project futures beyond the patterns of the past. This has raised concerns about machines’ capabilities to merely replicate existing knowledge (Öberg et al., 2021; Geissinger et al., 2024).

Contemporary developments suggest that AI – as in generative AI – now emulates a wider range of human cognitive functions possibly approaching the threshold of singularity (Upchurch, 2018). The emergence of large language models, tools such as ChatGPT, MidJourney, Sora and DALL·E 2 are reshaping the conception of machine capabilities (Cillo and Rubera, 2025; Grewal et al., 2025). AI is thereby increasingly associated with creative capacities previously thought to be uniquely human.

In this evolving landscape, tools have redirected attention towards implications for the labour market, professional identities and broader questions around the boundaries between human and machine creativity, affecting or potentially affecting, B2B exchanges in terms of decreased need for external professionals, for instance (cf. Golab-Andrzejak, 2023; Boussioux et al., 2024; Hui et al., 2024).

At the time of writing, discussions have also begun to centre on reasoning AI (Alam et al., 2025; Nvidia, 2025; Park and Choi, 2025), which once again ascribes to machines abilities historically reserved for human cognition, including critical thinking. Whereas generative AI has been criticised for its lack of emotional capacity (Rhue, 2024), reasoning AI implies a more advanced capability. It extends beyond language reproduction to encompass analytical and interpretive understanding of complex issues. More precisely, reasoning AI is understood to use logic and inference in complex problem-solving, to adopt critical perspectives and thereby to move beyond the pattern-matching capabilities of generative AI. Thus, human directorship over AI, insofar as it involves the evaluation of generated solutions, would be incorporated into AI itself. This enables AI to compare different outcomes, criticise them and argue for the superiority of one alternative over others, as well as to produce chains of thoughts and elaborate on the reasons for its choices.

AI can thus be regarded along three types: decision-making, generative AI and reasoning AI. Moreover, and as seen from the literature review in this present paper, AI can also be understood as the human it replaces, allowing configurations across human-to-machine, machine-to-machine and machine intermediated interactions (e.g. Paschen et al., 2020; Wei and Geiger, 2024). Combined the types of AI and the dimensions of machine involvement creates a grid (see Figure 1).

This grid guides the literature review and future-oriented outlooks in this paper, where the complete or partial replacement of humans puts trust at the centre of the discussion. This is underpinned by how trusting would normally require humans and apply to the social, actor bond dimension of relationships (Huemer, 1998) important for continued interactions and thereby with implications for B2B interactions as such. Relative the grid, the paper addresses the outlined questions: How is trust in B2B relationships shaped when human actors are replaced by machines? How do specific applications of AI – such as decision-support systems, generative AI and reasoning AI – mediate and transform this process?

This paper is based on an integrative literature review (Torraco, 2016). While it synthesises existing research, it also seeks to integrate emerging perspectives to offer a forward-looking agenda for the topics under consideration. More specifically, as demonstrated in the paper, research on trust in the context of AI within B2B interactions remains notably limited (Hengstler et al., 2016; Shneiderman, 2020) as is research on generative and reasoning AI (Alam et al., 2025; Cillo and Rubera, 2025; Grewal et al., 2025; Park and Choi, 2025) related to B2B interaction.

To address these gaps, the paper draws on broader notions of trust beyond the specific B2B setting and integrates it with research on the three types of AI introduced earlier: AI as a decision-making tool, generative AI and reasoning AI (Rosenthal-Sabroux and Zarate, 1997; Hui et al., 2024). It also addresses gaps concerning the three dimensions of machine intervention in B2B exchanges – human-to-machine, machine-to-machine and machine-intermediated exchanges. Focusing on trust, the actor dimension of the ARA model serves as the analytical framework for the discussion, thereby positioning the paper firmly within the literature on B2B exchanges. In this context, resource ties and activity links are explored as potential mechanisms for compensating the loss of human touchpoints (cf. Koponen and Rytsy, 2020; Kot and Leszczynski, 2020; Wei and Pardo, 2022).

In the review of existing research, various search combinations were tested and evaluated. The Web of Science database was selected due to its coverage of peer-reviewed articles in high-quality journals and its capacity for structured combinations of search terms. These were essential features for the scope of this study. The review began with integrated searches combining trust, B2B interactions and AI, reflecting IMP in the searches, then progressed to two-by-two combinations of terms, broadened the searches to AI in B2B research also beyond IMP. For each of the searches, a range of synonyms was used to ensure that conceptual meanings, rather than exact phrasings, were captured in the search results. Table 1 presents the search results across various combinations, including exact search terms and the time span of the literature retrieved.

The initial search results yielded several irrelevant articles, for example, instances where “ARA” referred to something other than the intended model, or where the content related to medical research. Consequently, all searches were limited to journals classified under Business and Management in the database and were individually assessed for relevance in terms of describing the ARA model or IMP as based on their meanings in B2B research. Given the limited number of IMP-related articles and the rapid development of AI, IMP conference papers were also reviewed. These were identified by searching the IMP website (impgroup.org) for conference contributions that included references to AI.

In analysing the resulting literature, the articles were firstly coded based on contents. This was summarised as short synopses for each article, outlining purpose, data/method and findings Initial codes were reduced across comparisons among the articles (Pratt, 2009) into themes to capture key contents in the literature and is presented in the section “Trust and artificial intelligence in business-to-business relationships – results and analysis of reviews”.

Next, the article contents were categorised into the three-by-three grid (see Figure 1), based on the three types of AI – AI as a decision tool, generative AI and reasoning AI – and the three dimensions of machine interaction in B2B exchanges: machine-to-human, machine-to-machine and machine-intermediated interactions. This framework enabled the identification of gaps in the literature and revealed how AI was approached in different research contexts. It revealed, as elaborated on in the paper, how much of the literature focused on AI as decision tools related to the seller/supplier side of B2B interaction in individual episodes of exchange.

To fill the gaps, the broader literature on AI was prospectively integrated using logic combining with the B2B interaction understanding of trust and the ARA model so as to determine what the lacking combinations would mean in the present or near future in relation to B2B interaction and specifically its social, non-economic dimension of actors (Torraco, 2016; Scully-Russ and Torraco, 2020). Searches were conducted in the broader marketing literature, but also beyond that, combining “trust” and “generative AI” and “reasoning AI”, searching “generative AI” in the Business and Management literature and searching “Reasoning AI” without any further delimitations. These searches were conducted as consequences of limited results when several terms were combined and to figure out the state-of-the-art literature on AI.

Since this part of the paper is prospective, it importantly was discussed with other researchers to figure out alternative interpretations, while both as gaps and based on the integrative approach being important for future research and practice. Figure 2 illustrates the steps of the analysis.

This section of the paper is organised into two parts:

1. an overview of IMP research on AI and trust; and

2. a description of AI within the broader B2B marketing literature (see the upper and lower part of Table 1, respectively, for sources).

Based on existing IMP studies (see the first two searches in Table 1), three distinct orientations towards AI can be identified:

1. AI as a complement to human actors;

2. AI as a replacement for human actors; and

3. AI as an outcome of interactions.

As seen in Table 1, the number of articles and conference papers remains limited, and across the three orientations, the literature appears fragmented and conceptually diffuse.

AI as a complement to human actors is, for instance, captured by Pardo et al. (2022), who highlight how the Internet of Things (IoT) creates separate networks from those typically described in the ARA model. Meanwhile, Paschen et al. (2021) explore how AI reshapes human resources and activities, noting, that at the time of writing, its full implications were still uncertain.

AI as a replacement of actors is represented by studies such as Keegan et al. (2023), who argue that AI blurs the boundaries between activities, resources and actors, primarily within the context of human-to-machine interaction. Similar orientations are found in research on digital agents by Kot and Leszczynski (2022) and Kot and Leszczynski (2020). The former examines AI agents representing suppliers in customer interactions, while the latter conceptualises virtual agents as either actors or boundary-spanning objects in B2B interaction (see also Biggemann and Soto, 2018; Leszczyński et al., 2019; Keegan et al., 2023).

AI as an outcome of interaction is exemplified by Sabatini et al. (2023), who focus on customer involvement in the development of smart solutions, thereby portraying AI as the result of co-created innovation rather than a pre-existing actor or resource.

Across these examples, AI is primarily associated with individual episodes of exchange (Håkansson, 1982) rather than with the development or maintenance of ongoing B2B relationships. Except where speculative future scenarios are considered (Geissinger et al., 2020), AI is conceptualised as smart or automated devices potentially capable of operating independently of human involvement but typically restricted to executing predefined tasks (Biggemann and Soto, 2018). This reflects a focus on AI as a decision-support tool relying on large data volumes to support or automate decision-making rather than on generative or reasoning AI capable of adaptive sensemaking or learning. Moreover, the focus largely remains on machine-to-human interaction, with AI typically positioned on the supplier side of exchanges, whether framed as an opportunity for marketers (Mero and Keranen, 2019), a managerial challenge when introducing new technologies across business networks (Geissinger et al., 2020), or a functional representation through digital agents (Kot and Leszczynski, 2020; Kot and Leszczynski, 2022).

Within this framing, the ARA model as applied to AI can be understood as illustrating how AI blurs the traditional boundaries between activities, resources and actors (Keegan et al., 2023) through the dual processes of complementing and replacing human agency (cf. Koponen and Rytsy, 2020; Kot and Leszczynski, 2020; Wei and Pardo, 2022, cf. Liu et al., 2024). This boundary blurring raises conceptual challenges for the ARA model itself, which presupposes social agency and mutual interdependence – features that are not easily replicated by algorithmic entities.

Notably, the concept of trust is explicitly mentioned in only one article (Kot and Leszczynski, 2020), where it is briefly noted that customers may not trust algorithms. Hence, discussions on the social glue traditionally provided by human actors which links discrete episodes into enduring B2B relationships, is largely absent in the current AI-related IMP literature. This underscores a theoretical gap: although AI is increasingly embedded in interaction processes, the mechanisms through which trust is created, transferred or substituted remain largely unexplored. Because AI is primarily positioned at the operational level of episodic exchanges rather than in the ongoing relational or strategic layers of interaction, trust is treated as a functional concern (e.g. trusting the accuracy of a chatbot or decision tool) rather than as a relational one.

As will be further elaborated, generative and reasoning AI along with machine-intermediated and machine-to-machine exchanges, remain underexplored domains within IMP research. These emerging forms of AI raise new questions about whether and how trust can be constructed in the absence of human actors, how machines might mediate or even generate trust and how such shifts might reshape the very foundation of actor bonds in B2B networks.

Expanding the review beyond studies grounded in the IMP perspective, but still within the scope of B2B marketing research (see the lower part of Table 1), a similar pattern emerges, namely, a focus on AI as a decision-support tool, typically positioned on the supplier or marketer side and associated primarily with episodes of exchange rather than with ongoing business relationships.

Across the 152 reviewed articles (see Figure 3 for their distribution by year), five main themes were identified:

1. AI in CRM and as a decision tool used by suppliers, with the supplier remaining represented by a human actor;

2. chatbot research, where the supplier is replaced or complemented by AI in human-to-machine interaction episodes;

3. algorithmic AI applications that coordinate exchanges through digital intermediation;

4. interactional perspectives where AI contributes to innovation and value creation predominantly from the supplier’s standpoint; and

5. AI-driven multi-party coordination across supply chains.

The dominant focus remains on the first category: AI as a decision tool for suppliers within human–machine interactions in supplier organisations, particularly in the areas of CRM and sales (Syam and Sharma, 2018; Chen et al., 2022; Dwivedi and Wang, 2022; Gaczek et al., 2023; Rahman et al., 2023; Rodriguez and Peterson, 2024; Fehrenbach et al., 2025; Hautamäki and Heikinheimo, 2025). To exemplify, Mikalef et al. (2023) highlight the role of AI in gaining customer insights, thereby reinforcing AI’s function for processing large volumes of data. Similarly, Paschen et al. (2019) and Mikalef et al. (2021) discuss AI in relation to marketing knowledge, while Rusthollkarhu et al. (2022) and Moradi and Dass (2022) expand this into the customer journey and Ameen et al. (2025) focus on service recovery.

These studies thereby describe AI related to tools used by the supplier rather than in exchanges. Along that vein of research, Bag et al. (2021) discuss how AI-driven decisions are rationalised through data and machine-based processes. While being far from discussing trust, this implies that non-economic reasons are ripped from the exchanges. A bulk of the B2B marketing AI literature also remains rooted in a performance orientation, focusing on how AI enhances supplier capabilities (Baabdullah et al., 2021), thereby emphasising the economic outcome of exchanges.

As part of the literature on chatbots (e.g. Kushwaha et al., 2021; Lin et al., 2022; Fotheringham and Wiles, 2023), Kushwaha et al. (2021) address the tension between human replacement and augmentation of touchpoints, while van Esch (2024) describes robots executing human tasks, highlighting human-to-machine interaction but still within episodic exchanges and while stilled fuelled by the supplier and with the human also being represented in exchanges.

Machine-intermediated exchanges are addressed by Bag et al. (2022) in the context of the sharing economy, while other studies on machine intermediation touch on platforms in retail or supplier networks. These platforms are usually treated as operational structures rather than as transformations of exchange relationships, where the broader literature on the sharing economy introduces trust cues (see the next section), but which is not extensively explored in the B2B marketing literature. Both chatbot and platform studies suggest focusing on the episodes of exchange, leaving broader aspects and consequences underexplored, including relational trust.

While technologies such as chatbots, IoT and AI-driven decision-making may affect interactions, explicit interactional analyses are rare. Exceptions include Osmonbekov et al. (2024) who speculate on how organisational buying may change due to AI, offering conceptual insights without empirical grounding.

Value creation is discussed by, for example, Li et al. (2021) conceptualising AI providers as third parties facilitating such value, while Petrescu et al. (2022) explore AI in the context of innovation (Arias-Pérez and Huynh, 2023; Sahoo et al., 2024; Liu et al., 2025). Across these studies, AI is framed as a means by which value is created in the form of new offerings or innovation, thereby resembling AI related to outcomes in the IMP research, where humans would continue to operate on both sides of the exchanges. These studies are generally framed in an opportunity-driven discourse with suppliers in focus, whereby issues such as trust is not explored.

In multiparty settings, AI is described as tools for planning and control in supply chains, repeatedly linked to issues like product responsibility. Here, AI either functions as a planning tool deployed across organisational boundaries or operates under the guidance of an orchestrator, without fully replacing human actors. While Dubey et al. (2021) examine AI analytics in supply chains, much of this literature addresses broader technological and digital developments, not AI specifically (Bodendorf et al., 2023; Sharma et al., 2024; Sutar et al., 2024; Niranjan et al., 2025). Two exceptions are Yan et al. (2022), who explore machine learning in supply chains, and Plangger et al. (2020), who analyse how knowledge flows differ when machines replace human actors, again manifesting the economic side of exchanges, while not reflecting on trust. Moreover, such replacement typically takes place at the level of exchanges rather than at the strategic level. Meanwhile, we can anew understand more rationalistic approaches to the supply chain planning, decreasing the influence of non-economic dimensions.

In terms of trust, Liu et al. (2024) explore the implications of AI for guanxi, pointing to potential disruption. Chang (2022) argues that replacing human salespeople with AI may undermine trust during vulnerable stages of a relationship (i.e. initiation and decline), reinforcing the idea that trust is more readily built through human interactions. Hitti and Ramadan (2025) suggest that accountability mechanisms in AI may help sustain trust, while Grewal et al. (2021) downplay trust as a major issue in B2B settings compared to B2C. Meanwhile, Hadjikhani and Lindh (2021) discuss the dual impact of IT on the related concept of commitment, which may increase cooperation or heighten uncertainty. Shankar et al. (2025) take a different angle, proposing that trust is essential for the introduction of AI itself.

In summary, the broader B2B marketing literature on AI is heavily supplier-focused, oriented towards episodic exchanges rather than relational constructs and conceptualises AI largely as a decision-making tool. Relationships among firms are seldom discussed, likely due to different ontological and theoretical foundations compared with IMP. The literature is also characterised by a reliance on literature reviews rather than empirical studies, which when present, are often dated. While decision-making AI is thereby extensively covered, discussions of generative and reasoning AI are minimal or absent. Notable exceptions include studies addressing managerial intentions and anticipated performance (e.g. Yuan et al., 2025) but these too focus on foresight rather than actual implementation.

Based on the reviews on IMP and the broader B2B marketing literature, there are thus numerous perspectives absent from prior research on AI. This is not least the case when attention shifts towards generative and reasoning AI, as well as relational dynamics. Much of the existing literature has conceptualised AI as complementing human actors in specific episodes, typically from the supplier’s perspective. That is, it is most often the supplier who introduces AI technologies, such as chatbots, decision-support systems or AI-enabled CRM tools. But for chatbots, these technologies are mostly used internally rather than being embedded in interactional processes with customers or partners. Studies including AI in customer interfaces remain positioned as human-to-machine interactions with some other studies focusing on machine-intermediated exchanges.

Table 2 repeats the three-by-three grid that combines types of AI (decision-making, generative and reasoning) with the different dimensions of human replacement in interactional contexts. This section elaborates on prospective outlooks, expanding beyond the episodic and supplier-centric focus of past research, while doing so through integrating AI research and practice from beyond the B2B marketing field (Torraco, 2016).

Before diving into the different combinations of machine-mediated interactions, it is essential to briefly reflect on generative and reasoning AI (Boussioux et al., 2024). In contrast to decision-making AI, which primarily supports rule-based, “if-then” logic and enhances analytical tasks, generative AI introduces more profound implications particularly for creative work (Boussioux et al., 2024; Ooi et al., 2025). Its application has, namely, raised concerns about replacing humans in roles that demand creativity, such as content writing, translation, photography and design (Amankwah-Amoah et al., 2024). This represents a shift from AI as a decision-making assistant to AI as a source of novel content and ideas, potentially challenging the role of the human actor in entirely new ways.

Despite these implications, the interactive dimension of generative AI remains underexplored in B2B marketing literature. As highlighted in the reviews, existing research seldom addresses generative AI in the context of exchanges (cf. Kumar et al., 2025; Li et al., 2025; Yuan et al., 2025), whether episodic or relational. While part of this absence may be attributed to the time lag between emerging technologies and academic publishing cycles, even studies that explore managerial intentions towards AI tend to pay limited attention to generative AI and its use in interaction.

Where generative AI is discussed, it tends to invert traditional AI dynamics: rather than AI aiding managerial decision-making, we see AI generating creative outputs, while managers take on the role of evaluators or curators, guiding the AI towards more desirable or contextually appropriate solutions (Öberg et al., 2024). This flips the conventional narrative of AI supporting human decision-makers and instead positions humans as supporting, directing or refining machine creativity.

Reasoning AI, although absent from current B2B literature, represents yet another conceptual leap. It can be viewed as combining the functions of both decision-making and generative AI by offering critical, contextual and reflective inputs (Kokkinakis and Drikakis, 2025; Park and Choi, 2025; Wang et al., 2025) and empathy (Alam et al., 2025). In this configuration, AI is not simply a tool for execution or generation, but a quasi-analytical partner. As such AI is capable of proposing insights, challenging assumptions and engaging in strategic reflection. This has potentially transformative implications for interactional processes, as it introduces a non-human entity capable of not just supporting but actively shaping the exchange through interpretation, evaluation and indeed empathy (Alam et al., 2025; Kokkinakis and Drikakis, 2025).

Most past research on AI in the IMP and broader B2B marketing research, has thus addressed human-to-machine interaction, with a primary focus on AI as a decision-making tool and as an agent complementing rather than replacing humans in exchange episodes (Bag et al., 2021; Papagiannidis et al., 2023; Gaczek et al., 2025). The complementarity suggests that AI typically supports, rather than substitutes, actor bonds and operates under human oversight.

Chatbot systems and increasingly sophisticated robotic service agents (de Kervenoael et al., 2020; Lin et al., 2022; Mahdi et al., 2022; Gustafsson et al., 2025) illustrate the growing presence of human-to-machine interaction at the customer interface, though still largely situated on the supplier side. These interfaces require a level of responsiveness and adaptability that goes beyond rigid, rule-based programming. Nevertheless, even in these more dynamic implementations, interaction is fundamentally shaped by human design, with AI relying on machine learning drawn from previous encounters (Kot and Leszczynski, 2022).

As shown in previous research, AI as a decision tool enhances the precision and algorithmic nature of decision-making. This precision influences exchange processes by marginalising the social, actor-bond dimension of interaction (Bag et al., 2021). For instance, decision algorithms may deprioritise certain customers without considering relational contexts. When machines replace human agents – such as through the use of chatbots – the customer must either trust the machine, develop trust over time through repeated encounters or adjust to a model in which transparency substitutes for traditional trust (Hawlitschek et al., 2018).

Since machines in these settings operate primarily on the episodic level of interaction, it is generally presumed that the relational level is still maintained by human actors on both sides. However, as shown in research on strategic and operational levels of interaction (e.g. Öberg, 2010), episodic exchanges at the operational level are critical for long-term relationship development. Thus, machine replacement at this level has relational implications. Combined, this means that with more of a focus on economic rational decisions (Bag et al., 2021) and a limitation of the human-to-human interaction in episodes of exchanges, the establishment and meaning of trust is altered. Essentially it is given less weight that may indeed include that B2B relationships are not established or preferred.

As for generative AI, applications thus tend to focus on content production and internal efficiencies (Boussioux et al., 2024; Cillo and Rubera, 2025; Grewal et al., 2025; Ooi et al., 2025), rather than reconfiguring how firms interact or co-create value in B2B relationships. Consequently, rather than reshaping trust dynamics, generative AI tends to insource previously external creative supply. Here, prompting replaces negotiation, and outputs are generated algorithmically rather than co-developed through interpersonal dialogue. This shift points to a transformation in the nature of B2B exchange from dialogical and negotiated processes to machine-mediated prompting and human review (Öberg et al., 2024).

When generative AI is incorporated into customer offerings, it can introduce scepticism, particularly regarding ethics and the authenticity of outputs, but also its quality and creativity (cf. Dwivedi et al., 2023; Golab-Andrzejak, 2023; Cillo and Rubera, 2025). Baek and Kim (2023), relatedly, highlight trust connected to generative AI, then introducing trust in its abilities, rather than trust in an interaction counterpart. This describes how the object of trust, but also its meaning, shifts, with trust not being linked to familiarity but rather to accuracy. In such scenarios, customers may experience doubt, with suppliers needing to compensate through increased quality assurance or a renewed emphasis on human negotiation to regain credibility and, indeed, trust.

Reasoning AI, finally, may carry both a heightened potential for trust and increased scrutiny. As its outputs improve upon those of generative AI by offering more critically reflective and context-aware contributions (Park and Choi, 2025; Wang et al., 2025), it may paradoxically be seen as more trustworthy yet also more suspect (cf. Baek and Kim, 2023), especially regarding authenticity and agency. These developments call for a reconsideration of how humans evaluate authenticity and whether the use of advanced AI tools fosters more mindful human–machine interactions. What furthermore is interesting is how reasoning AI introduces meanings of empathy (Alam et al., 2025), thereby humanising the machine, potentially closing the circle related to (a new kind of) non-economic reasons for repeated exchanges.

Machine-intermediated exchanges describe scenarios where AI or algorithmic systems act as intermediaries between interacting parties (Bag et al., 2022; Wei and Geiger, 2024). This is increasingly evident in both the sharing economy and emerging digital infrastructures such as distributed ledger technologies (Gligor et al., 2021), where platforms and algorithmic mechanisms do not merely support but actively monitor, structure and even govern the terms of exchange. Core features of these systems include algorithmic coordination, the centrality of digital platforms and the promise of complete information transparency.

In blockchain-based systems (Queiroz et al., 2020; Geissinger et al., 2021a, 2021b), intermediation is redefined altogether: human oversight is replaced by cryptographic protocols, smart contracts and distributed consensus. Here, decision-making authority shifts from human agents to machine logic, and the touchpoints of traditional interaction are either displaced or recoded into data points and system parameters. Whether through platforms or decentralised ledgers, the net effect is a transformation towards transactional forms of exchanges, often devoid of repeated interaction and thus also of relational continuity.

This transactional shift does not eliminate the importance of trust but rather reconfigures it. In platform-mediated environments, trust is often constructed through trust cues rather than built through interpersonal familiarity. Trust cues in this context include user photos, item descriptions and third-party ratings (Ert et al., 2016; Pelgander et al., 2022). Sharing economy research repeatedly shows that such cues enhance perceived reliability, enabling interactions even among previously unknown parties (Mao et al., 2020; Pelgander et al., 2022). Meanwhile, such trust cues are partly contrasted by algorithms (Geissinger et al., 2021a, 2021b), thus meaning that decisions on counterpart is either based on preference or algorithms, where AI geared platforms would lean more on algorithms. This indicates how trust in its human interactional meaning is either shielded through third party evaluations and descriptions about the supplier, its resources and past activities or how trust through algorithms are replaced with only economic arguments. In the very essence of the sharing economy lies that transactional character of exchanges.

Distributed ledger systems, meanwhile, introduce a logic of trust-through-transparency. Information is made openly available to all parties, enabling traceability and provenance (Wang et al., 2019; Kimani et al., 2020) and effectively allowing trust to be placed in the system architecture rather than in the human counterpart. These systems promote a technologically mediated trust model, decoupled from social or relational histories and whereby actor bonds between suppliers and customers beyond episodes of exchange would not be established. In this context, decisions remain human, particularly regarding which exchanges to engage in and which systems to trust, but the execution and enforcement of those exchanges increasingly lie with machines.

Beyond trust cues and distributed ledgers, these machine-intermediated interactions may also create trust or indeed distrust, in the intermediating machine (Pelgander et al., 2022), for instance based on biases and disabilities to create accurate decisions (Akter et al., 2022; Baek and Kim, 2023) not least beyond pure coordination.

As of now, generative and reasoning AI has not been widely explored in the context of machine-intermediated exchange. Yet these developments invite speculation about future forms of machine-facilitated preference articulation and more advanced decision intermediation, but also then risks of hallucinations and wrong decisions (Baek and Kim, 2023) and the impact on trust. Meanwhile, reasoning AI could enable not just data-driven matches but also context-aware selections, embedding a greater degree of nuance and flexibility into the intermediary process.

The implications for trust are thereby significant. As more decision-making tasks shift to machines, the supplier and customer may each begin to form enduring ties to the intermediary itself, whether a platform, a system or a generative AI tool. In these cases, trust is built in and through the machine’s outputs (Baek and Kim, 2023), which are evaluated for authenticity, plausibility and quality. At the same time, human users continue to steer, select and interpret these outputs, resulting in a human-in-the-loop trust dynamic. Human discretion remains critical, but the interaction is increasingly triangulated through machine-generated representations and decisions, marking a profound shift in the relational infrastructure of B2B exchanges.

Prior research on AI in supply chain coordination highlights how automation technologies such as IoT-enabled devices, sensor-based communication and predictive maintenance systems, extend operational control across firm boundaries (cf. Pardo et al., 2022). These systems facilitate machine-to-machine interactions that are primary episodic. Activity links in such contexts are no longer shaped by interpersonal negotiation or social embeddedness, but by data-driven protocols, anomaly detection algorithms and real-time optimisation mechanisms. As such, they introduce non-human activity links that operate autonomously, bypassing traditional human touchpoints, while interestingly including investments in shared systems as an expression of trust on the relational level.

In these machine-to-machine settings, trust is not established through actor bonds, but through the performance reliability of the systems involved (Glikson and Woolley, 2020). Here, interaction is governed by algorithmic and ledger-based mechanisms, which prioritise transactional efficiency and predictive control over relational continuity. While resource ties and activity links may still carry long-term strategic value, for example, in systems that require ongoing interoperability, integration or shared maintenance infrastructure, these do not compensate for the absence of human-based trust dynamics. The relational space is de-socialised, unless deliberately designed otherwise.

In this prospective future, machines would not trust but optimise. Interaction would be grounded in algorithmic negotiation and learning systems, with decision rules evolving through reinforcement or generative feedback loops.

In the scenario of generative and reasoning AI, business interactions would no longer be mediated by human decision-makers but by autonomous creative agents, capable of content generation, problem-solving and negotiation within pre-defined parameters. Albeit the literature beyond B2B marketing elaborates on trust related to AI, it still assumes that there is a human trusting (Baek and Kim, 2023; Saffarizadeh et al., 2024), and also distributed ledgers and trust cues are based on this assumption (Hawlitschek et al., 2018). Machine-to-machine interaction would, if trust were to re-enter such systems, likely be simulated, proxied or programmed. It would be embedded into system design rather than socially constructed through repeated encounters. Here, preferencing, liking and discretion would be quantified and codified, shifting the foundation of B2B relationships from interpersonal processes to machine-logics of mutual optimisation.

This paper reviewed and discussed trust across human-to-machine, machine-intermediated and machine-to-machine interactions, exploring how B2B exchanges are being transformed by the advent of AI in terms of decision-making, generative and reasoning AI. At its core, the study addressed two questions: How is trust in B2B relationships shaped when human actors are replaced by machines? Moreover, how do specific applications of AI – such as decision-support systems, generative AI and reasoning AI – mediate and transform this process?

As indicated, both when trust is discussed and when it is more implicitly understood based on how AI operates, past notions of trust are deemed to change. This is an area that remains insufficiently explored within B2B marketing research. Trust, along with the closely linked concepts of preferring and liking, becomes increasingly abstracted and systematised, entangling the question: Does the focus on episodes of exchanges mean that AI destroys the B2B relationships and more specifically the social, actor dimension as glueing episodes together through trust?

In this evolving context, two conceptual trajectories emerge. One posits that trust cues (Etzioni, 2019) and algorithmic transparency (Hawlitschek et al., 2018) can be embedded within AI systems, enabling users to form alternative modes of trust such as trust in the system’s functionality, consistency and output quality. The opposing view, more sceptical, warns that such systems may erode trust altogether. In this framing, trust becomes decoupled from social norms, affective commitment or mutual moral obligations. What remains might be better described as confidence or reliance, rather than trust in its traditional, interpersonal form and the social, non-economic dimension increasingly becomes replaced by ideals of optimisation (Bag et al., 2021). This transformation is not only about humans learning to trust machines; it is also about the machine’s structural inability to reciprocate or even register social trust.

Moreover, past literature is intensively built on the notion of a human as trusting. What really creates challenges is when the machine is to be the trusting party. This creates research gaps as we move into machine-to-machine interactions, not least in the essence of possible singularity whereby humans no longer guide the process (Upchurch, 2018), where nothing beyond optimisation and a disconnect from past notions of trust emerge. The other side lies in the current development of reasoning AI. It entails aspects of critical thinking and empathy (Kokkinakis and Drikakis, 2025; Park and Choi, 2025; Wang et al., 2025), thereby upholding ideals closely related to the non-economic, social dimension of exchanges. The emergence of reasoning AI along with distrusts following from lack of authenticity and quality of output from machines (Park and Choi, 2025; Wang et al., 2025), indicate how those constructing AI algorithms may still see the importance of trust as their ideal. Across the three types of AI, we can thereby envision a loop: from human trust, over replacing trust for optimisation ideals, into how we want decisions and interactions to be humanised as development ideals of reasoning AI. Whether this loop will be closed is though too early to say.

The main contribution of this paper lies in its integrative literature review on AI, whereby present research on AI in B2B marketing is put against broader literatures on AI and a further extension related to state of the art and emerging types of AI. Its focus on trust draws attention to a profound gap related to relational aspects where B2B marketing, especially within the IMP tradition, would importantly be able to provide answers. The development where ideals move from social out of necessity over optimisation where aspects of non-economic dimensions are ripped from exchanges and into human ideals as part of reasoning AI creates an interesting loop affecting each type of machine mediation in AI. As such, the discussion on trust challenges present models and assumptions about interaction and the trust concept. The paper thereby argues for conceptualising of machine trust not as an extension of interpersonal trust but as a distinct phenomenon, while also considering how displacements of trust may not be the final say related to AI.

The paper’s key contributions to the IMP literature lie in its synthesis and conceptualisation of the current state of AI-mediated exchanges. IMP’s extensive insights into the relational dimensions of exchange are particularly valuable for understanding the effects of AI, as the tradition emphasises the social, non-economic glue that binds exchanges into enduring relationships over time. At the same time, AI introduces new contexts in which critical questions must be addressed – such as the risk of increasingly transactional rather than enduring exchanges, optimisation at the expense of shared value creation and algorithms and control mechanisms potentially displacing trust (cf. Huemer, 2004; Roxenhall and Ghauri, 2004). Concepts such as trusting, liking and preferring therefore to require renewed examination, particularly regarding the value they create and how they can be sustained in AI-mediated exchanges.

Managers should formulate a deliberate AI strategy when integrating machines into B2B interactions. However, based on the types of AI and machine integration complementing or replacing humans on one or both sides of exchanges, there is not a one strategy fits all. That said, the following needs to be considered by managers:

• The balance between AI efficiency and relational consequences. This has implications for whether or not to introduce AI, consciously ensuring that strategic relations are kept within human supervision.

• Design trust mechanisms that fit with the type of AI but also the human-to-machine, machine intermediated and machine-to-machine interactions. Trust mechanisms include transparency and explainability, trust cues and interaction protocols, clearly defined accountability structures and where humans remain involved, support for the interaction party’s sensemaking and confidence building.

• Managers also need to vary their interaction strategies based on AI types to establish and maintain trust: For AI used as a decision-support tool, managers need to clarify decision rights between humans and AI systems, communicate the scope and limits of AI-geared recommendations to exchange partners and prevent decision opacity from undermining partner confidence in machine-intermediated settings. For generative AI, and with a future-oriented perspective, managers should prepare for AI to function as an interactional party and establish clear guidelines for its use in communication with interaction parties, while preventing trust erosion caused by over-automated or inauthentic interactions. For reasoning AI, managers must ensure that reasoning processes remain inspectable and contestable, particularly in machine-intermediated and machine-to-machine exchanges and that interaction partners understand how AI arrives at conclusions that affect joint outcomes.

• Finally, managers should introduce performance metrics that help to monitor AI as part of exchanges and relationships. These include metrics capturing AI performance and indicators of relational effects, such as perceived fairness, predictability and partner confidence.

We are witnessing a transformation from a paradigm of digitally embedded exchange, where AI modifies human interaction, to one of organisationally embedded AI, where AI may become the primary medium of interaction. This transition has profound implications for how we understand B2B relationships, value co-creation and the very ontology of exchange. This paper seeks to raise questions about how AI affects trust as a core element of long-term, committed B2B relationships. Nevertheless, the paper’s design and focus entail several limitations. Accordingly, future research is needed to further examine how AI shapes the organisation and outcomes of B2B relationships, as well as to extend reviews to encompass concepts, types (beyond AI as a decision tool, generative and reasoning AI) and dimensions of AI not covered in the present paper.

Meanwhile, the literature reviewed in this paper reveals several gaps that warrant future research. These include the role of AI at the relational level, rather than its confinement to discrete exchange episodes; the emergence of generative AI in exchanges and activities spanning organisational boundaries; and the ways in which reasoning AI may be incorporated into organisations and B2B relationships. The prospective outlooks summarised in Table 2 thereby suggest promising avenues for future research, particularly regarding the relational dimensions of exchange and the implications of AI replacing human actors. As illustrated by the grid, much, namely, remains to be explored and empirically captured by B2B marketing scholars related to generative and reasoning AI, to machines replacing rather than complementing humans and to machine-to-machine interactions on strategic levels.

A sustained focus on trust, combined with a deeper examination of the actor dimension within exchanges, offers the potential to advance our understanding of emerging modes of interaction and their relational implications in AI-integrated B2B contexts. Future research should therefore re-evaluate core assumptions underlying interaction models, including the nature of trust, the role of human agency and the meaning of relationships in AI-driven business environments. Such studies should ideally be based on empirical real-time research but may also use scenario-based approaches.

In IMP research, examining empirical applications of AI and how they reshape the very notions of B2B relationships, actor bonds and trust is of fundamental importance – both now and in the future. Furthermore, IMP research needs to acknowledge and examine AI beyond its role as a decision-support tool used by suppliers.