Engaging consumers through artificially intelligent technologies: Systematic review, conceptual model, and further research

While consumer engagement (CE) in the context of artificially intelligent (AI ‐ based) technologies (e.g., chatbots, smart products, voice assistants, or autonomous cars) is gaining traction, the themes characterizing this emerging, interdisciplinary corpus of work remain indeterminate, exposing an important literature ‐ based gap. Addressing this gap, we conduct a systematic review of 89 studies using the Preferred Reporting Items for Systematic reviews and Meta ‐ Analyses (PRISMA) approach to synthesize the AI ‐ based CE literature. Our review yields three major themes of AI ‐ based CE, including (i) Increasingly accurate service provision through AI ‐ based CE ; (ii) Capacity of AI ‐ based CE to (co)create consumer ‐ perceived value , and (iii) AI ‐ based CE's reduced consumer effort in their task execution . We also develop a conceptual model that proposes the AI ‐ based CE antecedents of personal, technological, interactional, social, and situational factors, and the AI ‐ based CE consequences of consumer ‐ based, firm ‐ based, and human ‐ AI collaboration outcomes. We conclude by offering pertinent implications for theory development (e.g., by offering future research questions derived from the proposed themes of AI ‐ based CE) and practice (e.g., by reducing consumer ‐ perceived costs of their brand/firm interactions).

Prior research has examined consumers' technology-facilitated brand engagement in the context of technologies, including social media, virtual-, augmented-, or mixed reality applications, and automated or artificially intelligent technologies, among others (e.g., Puntoni et al., 2021), which have been demonstrated to boost corporate performance (Huang & Rust, 2021).Here, consumers' engagement with artificially intelligent technologies-in particular, those able to perform tasks without human intervention (e.g., generative or predictive artificial intelligence [AI]; Dwivedi et al., 2023) -is receiving widespread attention in the consumer psychology and marketing literatures (Sampson, 2021;Wu & Monfort, 2023).Among these, the AI subclasses of machine learning (i.e., an AI subset, which uses algorithms that flexibly adapt to data constellations to improve task performance) and deep learning (i.e., a machine learning subset that involves computing multilayer neural networks) technologies are able to offer improved decisions or predictions over time based on the deployed (e.g., big) data (Hollebeek et al., 2021;Pradeep et al., 2019), thus offering consumers increasingly accurate (e.g., product) solutions (Leung et al., 2018), provided high-quality training data and state-of-the-art algorithms are used.For example, while Google's machine learning-based predictive text (predictive AI) improves (learns) over time, Amazon's Echo or Apple's Siri likewise, provide increasingly customized solutions to their users (Pitardi et al., 2023).
However, despite the importance of fostering consumers' engagement with or through AI-based technologies, authors have adopted myriad theoretical perspectives and methods to investigate this multidisciplinary topic, yielding theoretical inconsistencies and fragmentation.That is, the adoption of different theoretical lenses and approaches to explore AI-based CE has yielded potentially incompatible findings, generating an important literature-based tension.For example, some authors suggest that AI technology (e.g., chatbot)-based social presence, "the extent to which [an AI technology] make[s] consumers feel …they are in the company of another social entity" (Van Doorn et al., 2017, p. 44), acts as key driver of users' engagement (e.g., McLean et al., 2021;Schuetzler et al., 2020).However, other researchers, like Tsai et al. (2021), identify opposing effects in that a chatbot's high (vs.low) social presence-based communication was not found to significantly affect user engagement.
To clarify these conflicting literature-based findings, we systematically review the AI-based CE literature.By elucidating the key themes characterizing this multidisciplinary literature stream, we uncover AI-based CE's hallmarks and dynamics.Following prior systematic reviews (e.g., Ameen et al., 2022), we also develop a conceptual model of AI-based CE and its nomological network (i.e., key antecedents and consequences;MacInnis, 2011).The development of new insight in this rapidly growing area is important, given the myriad applications of AI-based CE that are forecast to continue redefining consumer behavior and marketing alike (McKinsey, 2023), thus serving as a springboard for further AI-based CE research.We view AI as "a system's ability to interpret external data…, to learn from such data, and to use those learnings to achieve specific goals and tasks through flexible adaptation" (Haenlein & Kaplan, 2019, p. 5), and engagement as a consumer's resource investment in their brand-related (e.g., AI) interactions (e.g., Hollebeek et al., 2019;Kumar et al., 2019).Overall, by collating and assessing the corpus of AI-based CE research, our analyses pave the way for this crossdisciplinary area's further development (Page et al., 2021;Pollock & Berge, 2018).This review paper makes the following contributions to the CE, AI, and the broader consumer psychology/behavior-and marketing literatures.First, adopting the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) approach (e.g., Page et al., 2021;Shobhit et al., 2023), we obtain a sample of 89 AIbased CE studies that are analyzed to deduce their theoretical hallmarks (e.g., deployed theories or methods; Rehman et al., 2020) and to uncover their main themes (Creswell & Creswell, 2018).We also develop a model of AI-based CE, reflecting MacInnis' (2011) notion of relating and, thus, unlocking acumen of the concept's position vis-à-vis its antecedents and consequences, a widelyadopted approach in prior PRISMA-based studies (e.g., Ameen et al., 2022;Rehman et al., 2020).These analyses matter, given AI's unique capabilities that have shaped, and which are expected to continue shaping, CE in important ways (Hollebeek et al., 2021).
Moreover, though the scope of AI-based CE has already expanded in recent years (e.g., through the launch of ever-evolving new technologies, like generative AI/ChatGPT), it is also forecast to continue growing in the years to come (McKinsey, 2023), warranting its strategic importance.
Second, following prior authors (e.g., Page et al., 2021;Verma et al., 2021), we derive an agenda for further AI-based CE research from our findings.This is also important, given the continued growth of, and the predicted ongoing (e.g., innovative) developments in, AIbased CE, providing rich opportunities for further exploration.
Specifically, given AI's relative newness in marketing, coupled with its potentially unique and evolving effects on CE, we expect our findings to hold value for researchers (e.g., by serving as a foundation for further exploration), warranting AI-based CE's importance in the coming years.
We next review AI-based CE literature (Section 2), followed by an outline of the deployed methodology (Section 3).We, then, present our main findings (Section 4), followed by an overview of key implications that emerge from our work (Section 5).
However, CE's definition is contested.For example, though Brodie et al. (2011, p. 260) define CE as "a motivational state that occurs by virtue of interactive co-creative, [consumer] experiences with a focal agent/object," Hollebeek et al. (2019, p. 166) conceptualize it as a consumer's "investment of …operant resources [e.g., cognitive/ behavioral knowledge/skills], and operand resources [e.g., equipment] in [their] brand interactions."Notwithstanding these differences, we distil the following generic CE traits.
First, CE is an interactive concept that centers on the consumer's interactions with a brand or (a) specific brand-related object(s) (e.g., brand-related AI technologies; Hollebeek et al., 2023;Perez-Vega et al., 2021).Here, interaction reflects "mutual or reciprocal action or influence" (Vargo & Lusch, 2016, p. 9) between the engagement subject (e.g., consumer) and the engagement object (e.g., an AI technology; Hoffman & Novak, 2018;Hollebeek, 2011;Sung et al., 2021).When consumers interact with AI technologies, they may be aware, or unaware, of their interaction with a machine, as gauged by the Turing Test.

| Artificial intelligence
AI, "a system's ability to interpret external data…, to learn from such data, and to use those learnings to achieve specific goals and tasks through flexible adaptation" (Haenlein & Kaplan, 2019, p. 5), has, likewise gained momentum in consumer psychology/behavior, and marketing, research (Mehta et al., 2022).For example, while Apple's Siri allows consumers to seamlessly execute tasks through voice commands, Google Home permits users to remotely perform tasks at home (e.g., monitoring the alarm; Xiao & Kumar, 2021).Through their ability to perform specific tasks increasingly accurately over time, AI applications that incorporate machine-or deep learning, or generative AI, technology may help consumers complete their tasks more efficiently or effectively (Huang & Rust, 2021;Xie et al., 2022).For example, chatbots tend to offer consumers more accurate, comprehensive, or personalized product recommendations or solutions over time (Dwivedi et al., 2023).
Scholars classify AI in different ways.For example, Huang and Rust (2018) propose a tri-partite typology comprising mechanical AI (i.e., able to perform repetitive or routine tasks), thinking AI (i.e., able to learn from data to make decisions), and feeling AI (i.e., able to display empathy; Hollebeek et al., 2021).Another pertinent AI classification is (a) generative AI, which is used to create novel, original, or creative (e.g., textual/image-based) content (e.g., ChatGPT/Copy.ai;Dwivedi et al., 2023), and (b) predictive AI, which uncovers patterns in historical data to predict or forecast specific future outcomes, behaviors, or events (e.g., predictive SMS; Hollebeek et al., 2021).By helping consumers execute specific tasks more efficiently or effectively (e.g., through predictive text), AI technologies can boost their engagement, whether with the technology or with the brand/firm (Hyun et al., 2022;Kull et al., 2021).For example, personalized AI-generated solutions can help raise consumers' (e.g., monetary/referral-based) resource investments in their brand interactions (Barnes & de Ruyter, 2022), boosting their engagement (Bertrandias et al., 2021).AI adoption can, thus, help foster value-laden customer relationships (Singh et al., 2021), demonstrating AI-based CE's strategic value.

| METHODOLOGY
To synthesize the corpus of AI-based CE literature, we conducted a systematic review to identify and assess published work in this multidisciplinary area (Petticrew & Roberts, 2006;Siddaway et al., 2019).To perform the review, we followed a three-step search process that comprised initial and secondary searches, followed by a snowball search, to ensure all relevant studies were identified and included in our article sample (see Figure 1; Giang Barrera & Shah, 2023;Swaminathan & Venkitasubramony, 2024).In our initial search, we focused on articles published in Marketing and Business journals, which we later broadened to include relevant articles published in non-Marketing and non-Business journals (e.g., Arts/ Humanities, Psychology, and Decision Sciences) in our secondary search.In the snowball search, we further scanned the literature, including the reference lists of the studied articles, to ensure no relevant studies were missed.
We, then, developed a protocol specifying the inclusion criteria for our articles (Hollebeek et al., 2023).Specifically, relevant articles addressing AI-based CE that were published in English, peerreviewed Scopus-indexed journals with an impact factor of ≥3 were considered eligible for inclusion in our review (Giang Barrera & Shah, 2023;Rehman et al., 2020).We did not restrict our search to any particular start date, and instead considered any eligible articles published up until October 8, 2023 (Le et al., 2022).
To guide our analysis, we adopted the widely-used PRISMA protocol (e.g., Hollebeek et al., 2023;Moher, 2009), which comprises three main phases, including Identification, Screening, and Inclusion (Page et al., 2021).First, in the Identification phase, we searched the titles, abstracts, and keywords of eligible Scopus-indexed journals for the following keywords (Shobhit et al., 2023): (engagement AND ("artificial intelligence" OR AI OR "machine learning" OR "deep learning" OR automated OR automation OR self-driving OR autonomous OR chatbot* OR "voice assistant*" OR robot* OR "digital assistant*" OR "virtual assistant*" OR "recommender system*" OR "recommendation agent*" OR "supervised learning" OR "unsupervised learning" OR smart)).We focused on original research, thus excluding prior (e.g., systematic or bibliometric) reviews (e.g., Lim et al., 2022;Mehta et al., 2022) from our article sample (Clarke, 2011).
The search conducted on October 8, 2023, which covered all eligible articles published up until this date, yielded a total of 16,698 records.However, of these, we only considered those published in English-language, Scopus-indexed Marketing and Business journals (Hollebeek et al., 2023).Therefore, studies published in other disciplines (e.g., Neuroscience/Microbiology, n = 15,386), in other languages (n = 7), or those in other outlets (e.g., conference proceedings, books/trade journals; n = 414), were excluded (Zorzela et al., 2016), yielding a sample of 891 publications eligible for further assessment.
In the Screening phase, we undertook title-abstract-keyword screening of the 891 articles (Page et al., 2021;Rehman et al., 2020), limiting our analysis to those articles exploring the interface of specific AI technologies and CE.For example, we excluded Acar and Toker (2019), who address sharing economy-based AI adoption vis-àvis consumers' personality traits (vs.AI-based CE).This part of the Screening phase yielded a total of 115 articles eligible for full-text review (Moriuchi, 2019;Perez-Vega et al., 2021).From these, we removed another 59 articles, either given their lack of relevance to our objective, or because the relevant journal's impact factor was <3.
For example, we removed Jain and Gandhi's (2021) work that is focused on AI and impulse buying behavior (vs.AI-based CE).
Therefore, in the Inclusion phase, we retained a total of 56 articles from our initial literature search (see Figure 1).
We, then, conducted a secondary search to ensure that all eligible articles were, indeed, included in our sample (Page et al., 2021; see Figure 1).Given AI-based CE's multidisciplinary nature (Sung et al., 2021) "virtual assistant*" OR "recommender system*" OR "recommendation agent*" OR "supervised learning" OR "unsupervised learning" OR smart)), we identified a total of 5,199 records (see Figure 1).
We, again, excluded articles featuring limited relevance, including those in Physics/Astronomy (n = 3,651), those published in non-peerreviewed journals (n = 607), and those in non-English outlets (n = 15), leading us to retain 926 articles for inclusion in title-abstract-keyword screening.Of these, a further 683 were removed due to lacking relevance, and another 182 owing to duplication, yielding a total of 61 articles for full-text review.For example, we removed Chen et al. ( 2019), given its focus on the interface of consumer-perceived value, past experience, and behavioral intention (vs.AI-based CE).Of the 61 remaining articles, a further 30 were removed (e.g., Lee et al., 2019), as they, upon closer inspection, did not address AI-based CE or were published in journals with an impact factor of <3, yielding 28 additional articles for further analysis (e.g., Jiang et al., 2022;Wen et al., 2022).
Finally, on October 11, 2023, we further verified that all eligible articles were included in our review by scanning the reference lists of the articles in our initial (n = 56) and secondary (n = 28) searches, uncovering five additional studies (e.g., Kumar et al., 2021).Overall, our sample contains 89 articles addressing AI-based CE (see Table 1 and Supporting Information: Appendix 1).

| AI-based CE themes
We next content-analyzed the articles to uncover the main themes of  behavior, online recommender systems are able to provide increasingly personalized solutions to their users (Maslowska et al., 2022).
AI's capacity to learn will help boost consumers' in-and/or extrarole performance (e.g., by enhancing the efficiency or effectiveness of the customer journey; Heller et al., 2021;Hollebeek et al., 2023).
For example, while voice assistants (e.g., Amazon's Echo) allow their users to multitask, AI-based Google Ads' capacity to provide multiple offerings in a matter of seconds can facilitate or accelerate their purchase decision-making (Kumar et al., 2016).Overall, AI's ability to learn, whether through thinking/feeling, machine/deep learning, or generative/predictive AI, permits firms to progressively pinpoint, estimate, or pre-empt those offerings that consumers are interested in, the communications they are likely to respond to, their responses to specific promotions, and so on (Lin et al., 2021) given" (Zeithaml, 1988, p. 14).For example, AI technology may be applied to mow the lawn or clean the house (e.g., robotic vacuum cleaners/lawn mowers).In these processes, consumers may cocreate value with the technology (Gao et al., 2022;Wen et al., 2022), where cocreation refers to a consumer's "perceived value arising from joint, interactive, collaborative, or personalized brand-related activities" (Hollebeek et al., 2019, p. 167).When consumers perceive their AI interactions to be of value, they will tend to derive positive perceived (cocreated) value from their interactions with these, and vice versa (Fang et al., 2022;Prentice et al., 2020), typically fueling their desire to continue engaging with these (Lalicic & Weismayer, 2021).
Customer-perceived AI value is likely to differ across AI subtypes and/or contexts.For example, while mechanical AI is able to automate routine tasks (e.g., a company's automated phone menu), it-unlike machine or deep learning, thinking or feeling, or generative or predictive AI technologies (e.g., conversational agents)-is not designed to learn or improve its performance over time (Hari et al., 2022).Likewise, while students may see more value in specific (e.g., essay) content being created for them through generative AI (e.g., ChatGPT/Google Bard), on holiday, they may wish to primarily interact with predictive AI (e.g., Google Ads to facilitate their purchase decision-making).Overall, our analyses suggest that as technologies increasingly mimic human thinking or feeling processes, their customer-perceived (cocreated) value is likely to rise (e.g., given their capacity to personalize service or to display empathy; Liu-Thompkins et al., 2022;Van Doorn et al., 2017).The consumerperceived value of generative (vs.predictive) AI may also differ (e.g., based on users' unique needs).

| AI-based CE's reduced consumer effort in their task execution
To perform their in-role activities (e.g., by researching, evaluating, or purchasing goods; Piercy, 2006) and extra-role activities (e.g., by providing brand-related word-of-mouth; Karaosmanoglu et al., 2016), consumers are, traditionally, required to invest specific cognitive, emotional, and/or behavioral resources, reflecting their engagement (Hollebeek et al., 2019).Consumers may perceive their resource investments to vary in terms of their perceived difficulty, exposing differing levels of role-related effort (Sweeney et al., 2015).While AI technologies do not necessarily remove consumers' required resource investments in their in-or extra-role activities (or rolerelated effort) altogether, these technologies may, indeed, reduce their required resource investments or effort (e.g., by performing specific tasks for them; Sampson, 2021).For example, video-tag recommender systems suggest specific tags to be added to online videos (Yang & Lin, 2022), reducing the user's necessary (e.g., cognitive) resource investment (e.g., in determining suitable tags) and boosting the technology's perceived ease of use, as proposed in the Technology Acceptance Model (Davis, 1989), while lowering their technological effort expectancy, as professed in the Unified Theory of Acceptance and Use of Technology (Venkatesh et al., 2003).Likewise, companies like Amazon or McDonald's are increasingly delivering their orders through land-based or airborne robots (e.g., drones), removing consumers' need to collect their order and enabling faster delivery.
In turn, consumers' technology adoption and continued usage are expected to rise.
The AI-induced reduction in consumers' required resource invest-  Hollebeek et al., 2021).Relatedly, rather than reducing consumers' required resource investments per se, their AI interactions may also shift the nature or composition of their engagement.For example, the use of AI-generated (vs.human-generated) product recommendations may lower their cognitive resource investment or effort.Personal factors address the consumer's individual characteristics, perceptions, and expectations, which we further subdivide into consumers' expected AI benefits and perceived AI congruency/identification. First, expected AI benefits refer to consumers' anticipated benefit of (using) specific AI technologies (Bertrandias et al., 2021;Kumar et al., 2021).Key constructs examined in this area include
Second, consumer-perceived AI congruency/identification denotes the extent to which consumers perceive an AI technology to be congruent with (i.e., fit or match) their actual or desired self, and the degree to which they identify with it (Loureiro et al., 2022;Yin et al., 2023).These include brand-self distance (Kull et al., 2021), selfcongruity (Yin et al., 2023), community identification (Wen et al., 2022), lifestyle congruency (Loureiro et al., 2022), and chatbot identification (Loureiro et al., 2022; Figure 2), among others.For example, Shumanov and Johnson (2021) identify the role of congruent consumer-chatbot personality as a key driver of AIbased CE.Overall, personal factors comprise consumers' functional or instrumental motives for engaging with AI technologies, alongside their more emotively driven (e.g., actual or ideal self-based) motives (Loureiro et al., 2022;Yin et al., 2023).The greater consumers' anticipated functional and emotive AI benefits, the greater the predicted initiation (e.g., uptake) and continuation (e.g., repeated use) of their AI engagement.
Technological factors refer to the characteristics of specific AI technologies, which are likely to differ across (e.g., mechanical, thinking, or feeling, or generative/predictive) AI (e.g., Huang & Rust, 2021).Based on our review, we further subdivide this category into technological design/appearance and technological capabilities.
First, technological design/appearance captures the AI's embodiment and presentation (e.g., its anthropomorphic shape or perceived coolness; Ashfaq et al., 2021;Wirtz et al., 2018).Here, the uncanny valley suggests that an AI's more anthropomorphic (human-like) traits tend to yield users' more favorable evaluations, and engagement, up to a point, which, however, decrease post-this optimum (e.g., as users start to perceive highly human-like robots as creepy; Belanche et al., 2021).
Interactional factors reflect the dynamics characterizing consumers' interactions with specific AI technologies (Fang et al., 2022), exposing AI-based CE's bilateral nature that may be instigated or maintained by the consumer or the AI (Hollebeek et al., 2021).
While some social factors may enable or accelerate AI-based CE (e.g., the ability of friendship AI, like Replika, to foster social bonding; Marriott & Pitardi, 2023), others might reduce it (e.g., AI-hesitant social norms or values).
Situational factors are transient contextual (e.g., time/locationspecific) variables that may impact AI-based CE (Hand et al., 2009).Our analysis reveals the particular role of situational variables in driving consumers' AI-based page visits or views, and that of discovery, surprise, and perceived relevance in shaping AI-based CE (Wen et al., 2022;Xiao & Kumar, 2021).For example, Maslowska et al. (2022) identify the effect of consumers' webpage visits on their engagement with AI-based recommendation agents.Given their ephemeral nature, situational factors are difficult to control or predict.We, however, expect more adaptable (e.g., thinking/feeling) AI to, generally, be better equipped to handle changing situational characteristics (Huang & Rust, 2021), thus exerting a more positive effect on AI-based CE. ) and/or (e.g., AI) usage intent, and customer retention (e.g., Aluri et al., 2019;Lin & Wu, 2023;Moriuchi, 2019;Rahman et al., 2023).While AI may be used to boost consumers' perceived brand-related outcomes (e.g., recommendations or advocacy), individuals may, likewise, recommend specific AI technologies in their own right (e.g., Loureiro et al.'s (2022) chatbot advocacy).
Second, firm-based consequences are the effects of AI adoption for the firm (Mishra et al., 2022), including the ability to command a price or revenue premium (Jiang et al., 2022) and to exploit customers' lifetime value (Maslowska et al., 2022), among others.
Given the, typically, more challenging task of obtaining (e.g., commercially sensitive) firm data, studies in this subcategory, however, remain limited (vs.those addressing consumer-perceived consequences of AI-based CE).
Third, human/AI collaboration consequences are the outcomes that arise from AI-based CE for users or consumers, specific AI technology, and/or other stakeholders (Chen, Gong, et al., 2022;Hyun et al., 2022).Despite its importance, this subcategory of AIbased CE consequences, likewise, remains modest in size to-date.
Thinking and feeling, or generative and predictive, AI technologies, in particular, are able to improve their performance over time (Dwivedi et al., 2023)

| Managerial implications
Our analyses also raise pertinent managerial implications.First, our initial theme of increasingly accurate service provision through AI-based CE suggests that the ability of AI technologies to learn or to reduce human error benefits service accuracy and -quality (Hollebeek et al., 2021;Huang & Rust, 2021).We, therefore, advise managers to scan their firms for relevant AI implementation opportunities, which we anticipate to raise long-term service quality, while reducing service issues and failure, in turn boosting organizational performance (Chen, Gong, et al., 2022).However, as not all tasks may be equally suited to AI adoption (e.g., in some cases, customers may prefer talking to a real person; Longoni & Cian, 2022), we advise managers to carefully assess those firm priority areas in which to adopt AI technology.
Second, our theme of the capacity of AI-based CE to (co)create consumer-perceived value suggests AI-based CE's ability to (co)create consumer-perceived value (Hollebeek, Sharma, et al., 2022).Specifically, AI technologies may help reduce perceived cost (e.g., by saving consumers time or effort in their task execution), or they may offer more convenient access, communication, or personalization options (Hari et al., 2022;Jiang et al., 2022), boosting consumer-perceived (co)creation (Vargo & Lusch, 2016).We, therefore, advise managers to conduct upfront (e.g., scoping) research with their target audiences to pinpoint those areas in which they would most value engaging with AI technologies.At the same time, we also caution against potential AI-based (e.g., privacy or security) risks (Bertrandias et al., 2021).Third, our final theme of AI-based CE's reduced consumer effort in their task execution proposes that AI-based CE reduces consumers' required effort in executing their role-related activities (e.g., by automating routine tasks), which we expect to, in many cases, raise their service (quality) assessments (Hollebeek et al., 2021;Leung et al., 2018).Correspondingly, we recommend managers to implement consumer-perceived effort-reducing AI technologies, given their predicted beneficial impact on users' service assessments (Sampson, 2021).

| Limitations and further research
Notwithstanding its contribution, this study also has limitations.First, we relied on the Scopus database to identify relevant AI-based CE articles in English, thus excluding articles published in non-Scopus journals.Future researchers could, therefore, consult other or related databases (e.g., Web of Science/Google Scholar) to source their data and include non-English works in their further reviews of AI-based CE.
Second, though we adopted a broad range of search keywords, AI's rapid innovation and evolution may spark new (future) AI-related terminology that is not covered in our analysis.We, therefore, recommend scholars to carefully scrutinize the emerging AI discourse, and assess its potential impact on or implications for CE, as well as for other stakeholders' (e.g., employees' or suppliers') engagement (Hollebeek, Kumar, et al., 2022).In other words, the emergence of new AI (or CE)-based insight may generate a need to revisit, test, validate, or refine the proposed AI-based CE themes.
Relatedly, while AI washing-claimed AI deployment when this is not the case (Leffrang & Mueller, 2023)-may aim to raise engagement, consumers learning about this falsehood may, in fact, lower their engagement, thus also meriting further scrutiny.
model of the article selection process.PRISMA, Preferred Reporting Items for Systematic reviews and Meta-Analyses.
, we broadened the search to articles published in Social Sciences, Arts/Humanities, Psychology, Decision Sciences, and Multidisciplinary journals.Using the same keyword combination: ((consumer* OR customer* OR brand* OR user* OR visitor* OR tourist* OR buyer* OR gamer*) AND engagement AND ("artificial intelligence" OR AI OR "machine learning" OR "deep learning" OR automated OR automation OR self-driving OR autonomous OR chatbot* OR "voice assistant*" OR robot* OR "digital assistant*" OR

( 2021 )
suggest that AI may reduce, or (co)destroy, perceived value, including in cases of service failure or unmet expectations (e.g., when the algorithm is still learning).Likewise, while authors, including Hlee et al. (2022) and Hyun et al. (2022), show that elevated AI friendliness, coolness, or competence boost CE, at low levels, these may hamper the development of these variables.
ments, or effort, to execute their in-or extra-role activities exposes an interesting literature-based tension: While the CE literature, conventionally, suggests that raising or optimizing CE will boost firm performance (e.g., Brodie et al., 2011), AI's role in lowering users' required resource investments may engender a need to revise this original CE-based assertion in the AI context (i.e., as AI may reduce their required effort;

Following
prior systematic reviews (e.g.,Ameen et al., 2022;Rehman et al., 2020), we next develop a model that depicts AI-based CE vis-àvis its key antecedents and consequences (Figure2).By synthesizing AI-based CE's nomological network, the model serves as an important resource for further AI-based CE scholarship.For definitions of the model's constituent concepts, please refer to Supporting Information: Appendix 3. 4.3.1 | AI-based CE antecedents Reviewing the corpus of AI-based CE research, we uncovered five categories of AI-based CE antecedents, including personal, technological, interactional, social, and situational factors, as detailed below.

4. 3
.2 | Consequences of AI-based CE We next identify key AI-based CE consequences that emerged from our review, which we classify as consumer-, firm-, and human/AI collaboration-based consequences.First, consumer-based consequences comprise user-perceived outcomes of AI-based CE, including the technology's perceived contribution to their brand experience, actual (vs.expected) perceived value, satisfaction, attitudinal and behavioral brand loyalty, brand equity, positive brand or AI-related word-of-mouth, affective brand or AI commitment, behavioral (e.g., purchase offers significant value for further researchers in this multidisciplinary field.For example, while the attained acumen of AI-based CE's antecedents facilitates assessments of how to cultivate the concept, its identified consequences help warrant its strategic value (e.g., given AI-based CE's demonstrated effect on key firm performance indicators like customer retention/lifetime value;Maslowska et al., 2022).
Third, though the emerging subfield of AI-based CE remains relatively nascent to-date, our sample, nevertheless, contains 89 articles, revealing the vibrant research activity in this area.We, however, expect AI-based CE research to proliferate in the coming years, offering opportunities for further (e.g., bibliometric) reviews of AI-based CE to complement our findings.Finally, based on the observed paucity of prior research in specific AI-based CE subareas (e.g., the effect of AI-based CE on human/AI collaboration), we recommend the development of further insight in these areas.For example, what AI attributes are core (vs.less core) in shaping users' engagement with specific AI technologies or with the brand?What is the effect of AI-based automated social presence (Van Doorn et al., 2017) on consumers' (e.g., brand) engagement?How many current or evolving AI technologies (uniquely) affect CE? Providing answers to these questions is pivotal to better understand the interplay between AI and CE in an increasingly automated world.
To what extent do AI-based mistakes influence consumers' brand engagement?○ How do AI-powered tools, such as chatbots or voice assistants, help consumers reduce the time/ effort required for complete their tasks and how does this impact their brand engagement?○ How does human/AI collaboration enhance the accuracy of AI-base predictions/ recommendations and how does this impact consumers' brand engagement?Which (if any) automation level exerts the greatest impact on consumer-perceived effort reduction and/or convenience?○ What factors motivate consumers to engage with AI technologies, and how will this impact their engagement-based resource investments in their role-related tasks?○ (How) does AI-based CE, and/or its nomological network, develop over time?