GenAI personalization: antecedents, outcomes, mediators, and moderators

The infusion of AI into hospitality has evolved from isolated digital solutions to ecosystem-level transformations. Tools such as GenAI chatbots, sentiment-aware virtual concierges and dynamic recommendation engines are redefining what it means to “personalize” service (Nanu, 2025; Florido-Benítez, 2025; Florido-Benítez and del Alcázar Martínez, 2025). As these systems increasingly emulate empathy and adaptivity, hospitality firms face both promise and peril, balancing efficiency with emotional authenticity (Zhang et al., 2025; Yi et al., 2024; Semwal et al., 2024; Chotisarn and Phuthong, 2025). This evolution has prompted a surge in academic interest (Loureiro et al., 2021). Many studies explore how AI-enabled personalization affects consumer satisfaction (Anguera-Torrell and Nicolau, 2025), service quality (Gupta and Jaiswal, 2024), emotional engagement (Park et al., 2025) and brand loyalty (Wang, 2025). Widely used frameworks such as the Technology Acceptance Model (TAM) (Mahmoud et al., 2024; Chotisarn and Phuthong, 2025; Kumar et al., 2025), Task-Technology Fit (TTF) (Zhang et al., 2025) and SERVQUAL (Mensah et al., 2025) provide foundational insights into technology adoption and service perceptions. While useful, these models overlook the emotional and relational complexities intrinsic to AI-mediated interactions. Emotional resonance, cultural intelligence and ethical transparency remain undertheorized, creating conceptual silos that hinder integrative understanding. At the same time, scholars have introduced new constructs, such as human–AI similarity, brand resonance and emotional value, to address the affective and experiential dimensions of personalization (Park et al., 2025; Lin et al., 2025; Piriyakul et al., 2024). However, despite this growing body of research, theoretical models are frequently applied without being recalibrated to account for the unique behavioral dynamics that AI introduces into hospitality service environments (Gursoy, 2025). There is also a lack of critical synthesis across empirical domains, which limits the development of robust, interdisciplinary insights (Gursoy and Cai, 2025). In addition, the methodological landscape is evolving faster than it is being understood. While studies have adopted advanced tools such as fuzzy-set qualitative comparative analysis (fsQCA), topic modeling, deep learning and transformer-based sentiment analysis (Kirilenko and Stepchenkova, 2025; Guidotti et al., 2025), few papers critically assess the strengths, weaknesses and contextual appropriateness of these approaches. This results in an unclear picture of what personalization means in practice, beyond automation and how it can be ethically and effectively implemented across hospitality settings. Given these gaps, this paper offers a critical review and conceptual synthesis of recent literature on AI-driven personalization in hospitality. It aims to:

The PRISMA-guided process was performed, followed by the Theory-Context-Characteristics-Methodology (TCCM) framework to conduct the integrative review (see Appendix 1 in the supplementary materials for the selection of the studies). To structure the analysis, the review was conducted across three focal domains:

1. theoretical shifts and contextual gaps in AI personalization;

2. methodological approaches and data sources; and

3. practical and conceptual gaps in human-centric implementation.

Unlike meta-analytic techniques, the critical review approach (Agapito and Sigala, 2024), combined with a causal loop diagram (Gupta et al., 2024), enabled the identification of emerging constructs (e.g. emotional granularity, algorithmic fairness and cultural personalization) and conceptual blind spots, including the lack of cross-cultural validation, emotional modeling and adaptive service logic. By adopting a reflective and theory-led lens (Gupta et al., 2024), this methodology (see Appendices 2 and 3 in the supplementary materials) offers a conceptual map to inform future academic inquiry and practice-oriented design of GenAI personalization systems in hospitality.

The research team maintained ethical integrity and adhered to best practices for responsible data use and privacy protection throughout the study.

The rise of AI-driven personalization in hospitality has prompted researchers to draw upon diverse theoretical frameworks (Table 1). While these signals indicate growing academic interest, the current landscape is marked by fragmentation, limited adaptation and theoretical inertia. Many dominant models, originating from technology adoption, service quality or consumer behavior domains, struggle to address the emotional, ethical and relational complexities introduced by AI. Despite the use of multiple theories, they are often applied in silos, without adequate integration or contextualization for AI-mediated hospitality experiences. The Technology Acceptance Model (TAM) remains the most frequently used framework (Mahmoud et al., 2024; Chotisarn and Phuthong, 2025). Although researchers have expanded TAM to include constructs like perceived trust, entertainment and intelligence, the model retains a rationalist bias and overlooks affective and cultural factors central to personalization. Moreover, it assumes technology use is a binary outcome, adopt or reject, while personalization is inherently graded, subjective and situational. The TTF model offers a performance-driven view of AI personalization (Zhang et al., 2025). It evaluates whether AI-enabled tools, such as automated review responses, suit the customer feedback’s informational and emotional tone. However, TTF reduces personalization to task efficiency, without addressing human–AI relationship dynamics, emotional authenticity or long-term brand engagement factors critical in hospitality. The SERVQUAL model (Mensah et al., 2025) has also been widely applied, but it was designed for human-mediated services and tends to mismatch with AI capabilities. For instance, how do constructs like empathy or assurance apply when a chatbot is the first point of contact? Scholars often stretch SERVQUAL to accommodate AI scenarios, leading to theoretical dilution and measurement inconsistencies. The Expectancy-Disconfirmation Theory (EDT), used to understand satisfaction gaps (Mensah et al., 2025), assumes expectations are shaped independently of the service system. However, personalization algorithms actively shape, nudge and manipulate expectations in AI contexts, introducing a feedback loop that traditional EDT is ill-equipped to model. Furthermore, EDT doesn’t account for algorithmic opacity or user miscalibration of system capabilities, both of which affect satisfaction in AI-mediated personalization. Emerging models, such as the Humanness-Value-Loyalty (HVL) framework, seek to explain how perceptions of AI “humanness” affect perceived value and brand loyalty. However, HVL remains underdeveloped in empirical studies and lacks integration with broader theories of emotional design or customer-brand relationships.

Branding theories such as Keller’s customer-based brand equity (CBBE) model (Piriyakul et al., 2024) and Aaker’s brand equity model have been adopted in AI-personalization research to assess how AI affects brand perception. These models capture value judgments and emotional attachment. Still, their application rarely distinguishes between AI-generated and human-generated brand experiences, missing an opportunity to theorize how digital touchpoints influence brand meaning. Meanwhile, affective and emotional theories such as appraisal, social appraisal and discrete emotion theory have been sparsely used (Yu et al., 2024; Li et al., 2024). These frameworks offer powerful tools for explaining emotional granularity, such as why AI-generated surprises may elicit delight in some contexts and discomfort in others. Yet, their integration into hospitality personalization studies is minimal, often limited to post-hoc sentiment coding rather than being part of a core theoretical foundation. Furthermore, Communication Theory and Social Communication Theory, though relevant to personalized message framing, are largely absent from the AI-related hospitality literature. These theories could help explain how AI systems craft socially resonant, culturally attuned or persuasive interactions, an area where personalization plays a vital role.

Several critical theoretical gaps emerge from the existing literature on AI-driven personalization in hospitality. First, there is a clear overextension of legacy models such as SERVQUAL and EDT, which are often applied without proper calibration for AI-specific service contexts. Initially designed for human-mediated environments, these frameworks fail to capture the nuanced, machine-mediated interactions that define AI-enabled personalization. Second, despite the inherently affective nature of personalization, emotional modeling remains underdeveloped. Most studies focus on cognitive or functional factors, with limited theorization of emotional triggers, responses or affective fit. Third, key relational elements, such as trust, empathy and transparency, which are crucial in human–AI interaction, are insufficiently theorized. Fourth, there is minimal integration between technology adoption models and theories of affective or cognitive behavior. Furthermore, the literature demonstrates an absence of culturally sensitive and ethically grounded frameworks, which is concerning given the global application of AI personalization technologies and their potential to reproduce bias or cultural insensitivity. To advance the field, future research should prioritize developing hybrid theoretical models that bridge performance metrics to the emotional, relational and ethical dimensions. For instance, integrating Appraisal Theory with TTF could provide a more holistic understanding of how users evaluate AI interactions in both functional and emotional terms. Similarly, combining Keller’s CBBE model with Discrete Emotion Theory could offer insights into how AI-generated personalization shapes emotional attachment and long-term brand loyalty. Such theoretical integration is essential for capturing the full complexity of personalization in AI-mediated hospitality experiences.

Significant contextual and thematic gaps remain despite the breadth of studies applying AI in hospitality personalization (Table 2). Most research is geographically siloed, focusing on Asia, Africa or Europe, which prevents comparative, cross-cultural validation of AI-driven personalization outcomes. Furthermore, while emotional reflection is analyzed in some studies, emotion-aware AI personalization remains undertheorized and inconsistently operationalized across guest journeys. Although fsQCA and AI techniques such as zero-shot learning and transformer-based models are frequently used, few studies integrate them with real-time personalization (Helmefalk et al., 2025) or guest-facing decision support systems, especially in complex operational environments. Moreover, despite the growing emphasis on the Sustainable Development Goals (SDGs), sustainability-linked personalization (e.g. ethical tourism preferences or green nudges) is scarcely addressed. Therefore, the current contexts disproportionately emphasize static, review-based post-hoc analytics over dynamic, adaptive in-service personalization frameworks. Future research should examine the efficacy of real-time personalization across diverse hospitality settings (e.g. budget vs luxury, rural vs urban), integrate multimodal guest data (e.g. visual, voice and sensor) and evaluate the cultural, emotional and ethical resonance in AI-guest interactions. Bridging these gaps will help AI personalization move from content analysis to context-aware experience orchestration.

The methodological landscape in AI-driven personalization research within hospitality is marked by innovation and inconsistency. Recent studies (Table 3) have embraced various techniques, from classical quantitative models to advanced AI and Natural Language Processing (NLP) methods. While this evolution signals a maturing field, it also reveals essential limitations in alignment, transparency and contextual appropriateness. One major trend is the adoption of hybrid computational approaches, including topic modeling (e.g. LDA, BERTopic), sentiment analysis fsQCA. For example, studies by Guidotti et al. (2025) and Kirilenko and Stepchenkova (2025) use transformer-based architectures and zero-shot learning to classify user sentiment and extract review themes. Similarly, Yi et al. (2024) and Tan et al. (2025) use fsQCA to model causal configurations in service perception and pricing fairness. While these techniques offer granular insights, they often lack interpretability and present a “black box” problem, making it difficult to link outcomes to theoretical constructs or managerial actions. This challenge is particularly acute when AI personalization research is divorced from human-centric interpretive frameworks. Another trend is the increased reliance on user-generated content (UGC), such as online reviews, comments and social media posts (Mahmoud et al., 2024; Mensah et al., 2025; Félix et al., 2025). While UGC offers rich and spontaneous data, it introduces selection bias, lacks demographic depth and rarely captures contextual factors such as customer expectations, service conditions or prior experiences. Moreover, studies that use scraping tools often fail to address data ethics and consent, raising concerns about the responsible use of digital traces in hospitality research. Much of the literature focuses on exploratory sentiment analysis, without moving toward hypothesis-driven models or longitudinal validation. For instance, while models such as VADER or TextBlob are used to label review polarity (Mensah et al., 2025), they are rarely triangulated with survey or interview data, which limits the ability to generalize findings or to confirm emotional depth. Even when emotion theories are cited (Yu et al., 2024; Li et al., 2024), the methodology often remains at the level of keyword extraction rather than providing a robust emotional mapping aligned with discrete emotion theory or appraisal theory. More complex methods, such as BERT-based cosine similarity, LGBM classification or zero-shot LLMs (Zhang et al., 2025; Félix et al., 2025), demonstrate methodological sophistication but often overshadow the hospitality context they aim to explore. These studies frequently lack domain-specific model calibration, and results are discussed in technical rather than experiential terms, creating a disconnect between method and meaning. This tension underscores the need for methodological reflexivity. AI tools should be evaluated for their predictive power, explanatory clarity and relevance to hospitality phenomena.

Furthermore, the current literature shows limited methodological integration across qualitative and quantitative paradigms. Very few studies apply mixed methods that combine computational analysis with human interpretation (e.g. interviews, netnography or observational fieldwork). This is a missed opportunity, mainly since personalization is deeply rooted in subjective perception, emotional resonance and social interaction. Furthermore, sampling strategies in AI-personalization studies are often opaque or opportunistic. While purposive sampling from online platforms is common (Zhang et al., 2025; Guidotti et al., 2025), little discussion is given to sampling biases, representativeness or demographic variation. Studies also tend to focus on English-language data or single-country contexts, limiting their cross-cultural generalizability and applicability to global hospitality operations.

Figure 1 presents a structured framework comprising aggregate dimensions derived from existing literature to explore the disconnect between technological advancement and human-centered service delivery. These dimensions highlight where AI-personalization in hospitality falls short in capturing cultural nuance, emotional depth and experiential alignment, thereby revealing key areas for practical enhancement and conceptual refinement.

In AI-enabled hospitality, cultural personalization is underdeveloped and often limited to superficial cues, such as generic destination guides, pre-set language options or templated itineraries. This approach overlooks the evolving expectations of modern travelers, particularly cultural tourists and digital nomads, who increasingly seek deeper, more meaningful interactions with the local heritage, people and values of the places they visit. Guests today are no longer content to be passive consumers of local attractions; instead, they seek authentic experiences that reflect cultural depth, emotional connection and participatory engagement. For instance, a system may recommend a well-known castle or museum, yet fail to align this with a guest’s stated interest in indigenous folklore or artisanal craft traditions. Similarly, a vegetarian guest from South Asia might receive restaurant suggestions centered on meat-heavy cuisine without considering religious or cultural dietary preferences. Such misalignments compromise guest comfort and create a perception of cultural insensitivity, ultimately eroding trust and diminishing brand loyalty. Moreover, many personalization systems fail to account for intercultural diversity, assuming uniform preferences among guests from the same country. A Chinese guest from Guangdong may favor different historical narratives or aesthetic experiences than one from Beijing. Likewise, visitors to Italy might prefer Renaissance art, agritourism in the countryside or culinary immersion, but systems tend to default to top-10 travel lists that lack nuance and relevance. To address these limitations, personalization in hospitality must be informed by cultural intelligence frameworks that discern where the guest is located and how they relate cognitively and emotionally to that place. One effective strategy is integrating dynamic cultural calendaring, enabling AI systems to recommend time-sensitive, culturally rich experiences tied to festivals, holidays or spiritual observances. For example, a guest’s itinerary can be extended to coincide with Gion Matsuri in Kyoto or Diwali in Jaipur. Similarly, origin-destination cultural profiling can refine recommendations using demographic, regional and linguistic data to distinguish between intra-national cultural groups and tailor suggestions accordingly. The semantic tagging of cultural assets, labeling destinations or activities with attributes such as “heritage value,” “local storytelling” or “intangible cultural significance,” can enrich the retrieval of content aligned with deeper guest values. AI personalization can also benefit from analyzing guest-generated content, such as reviews, photos and search queries. These inputs often contain subtle indicators of cultural and emotional preference, revealing a desire for experiences described as “spiritual,” “handcrafted” or “authentic.” Moreover, personalization should extend to the way information is presented, both visually and verbally. Design preferences, narrative style and tone should be calibrated for cultural resonance, for example, minimalist layouts and neutral color schemes for Korean travelers versus vibrant visuals and expressive text for Brazilian guests. By embedding this kind of cultural depth into AI-driven personalization logic, hospitality businesses can move beyond simple localization toward experiences that are not only customized but humanized. Such efforts are essential for enhancing guest satisfaction and emotional engagement and advancing sustainable and respectful tourism that fosters meaningful connections between travelers and host communities.

As digital interfaces continue to reshape the hospitality landscape through kiosks, mobile apps, chatbots and voice-enabled assistants, the promise of seamless personalization remains largely unfulfilled. While these technologies often excel in delivering functional efficiency, such as quick check-ins, service requests or booking confirmations, they frequently fall short in offering emotionally resonant or cognitively adaptive guest experiences. Personalization in this context is commonly reduced to algorithmic matching of past behaviors or pre-programmed workflows, disregarding user intent, mood or contextual nuance. The result is an interaction that, while efficient, feels impersonal, mechanical and ironically misaligned with the ethos of hospitality. For example, a chatbot may deliver instant responses to queries but fail to detect tone, urgency or sentiment in guest messages. A visually polished mobile app may offer intuitive navigation, yet be overwhelming for older adults unfamiliar with digital interfaces or linguistically inaccessible to nonnative speakers. In high-touch settings like luxury hotels or resorts, overly rigid automation may alienate rather than support. Emotional fatigue, frustration or perceived neglect can result when digital tools do not recognize subtle cues such as hesitations, repeated commands or rapid switching between service tabs, which may signal confusion or dissatisfaction. Moreover, hospitality technology often lacks adaptability across cultures and guest personas. A minimal, icon-based layout may appeal to some markets, while others might expect more detailed, text-rich interfaces. Similarly, business travelers may prefer fast, transactional interactions. In contrast, leisure travelers may respond more positively to storytelling-based prompts, concierge-style dialogue or rich media showcasing property features and local experiences. Without adaptive personalization based on guest purpose, preferences or behavior, digital systems risk offering one-size-fits-all experiences that dilute service differentiation.

To address these challenges, personalization engines must evolve toward more context-aware interaction design. Systems should dynamically adjust interaction parameters, such as response tone, language complexity, pacing and content depth, based on the user’s device, location, time of day, behavioral signals and prior engagement history. For instance, a guest interacting late at night might be served with minimal, quiet-mode prompts, whereas the same guest in the morning could receive full itinerary suggestions and notifications. Response interfaces should be capable of detecting emotional tone and cognitive load, and shifting their behavior accordingly, slowing down, simplifying or rephrasing based on guest needs. In addition, integrating multimodal feedback such as touch behavior, voice inflection or usage rhythm can enable digital tools to interpret not just what the guest wants but also how they want it delivered. Personalization should also extend to visual and linguistic preferences, offering culturally responsive templates, fonts and color palettes that enhance usability and comfort. Hospitality brands that implement such adaptive interfaces would improve operational efficiency and reinforce their brand promise of care, attentiveness and personalized service, even when that service is mediated by technology. Moreover, digital interaction in hospitality must shift from transactional automation to relational augmentation, using technology not as a substitute for human empathy, but as a bridge that enhances and extends it. This requires a reimagining of personalization, not as a static set of features, but as a living, learning interface that reflects and respects every guest’s journey’s diversity, pace and emotion.

Economic personalization in hospitality has traditionally focused on transactional indicators such as booking frequency, average spend or past purchase behavior. While useful for segmentation, this one-dimensional view fails to capture the psychological, emotional and contextual drivers that shape guests’ perceptions of value. In most AI-enabled systems, personalization remains narrowly tethered to price points and reward tiers, overlooking how fairness, transparency and situational context influence economic behavior and brand loyalty. For instance, offering high-priced add-ons to guests who recently opted for a budget-friendly stay, perhaps due to a family emergency or an economic downturn, can come across as tone-deaf and erode trust. Furthermore, dynamic pricing engines and loyalty algorithms often function without adequate interpretability, leaving guests uncertain about why they received a particular offer or discount. This lack of clarity can give rise to perceptions of price manipulation or unfair treatment, particularly when guests compare their experiences with others. In hospitality, where emotional resonance and perceived equity are critical, such breakdowns in perceived fairness can directly affect brand credibility. Guests are more likely to advocate for brands that align not only with their budget but also with their values and life situations, such as those that reward long-term engagement, recognize personal milestones or support ethical tourism. Economic personalization must therefore move beyond reactive pricing to become contextually sensitive, emotionally intelligent and transparently communicated. For example, a returning guest who celebrated an anniversary at a resort might be offered a custom package the following year, not merely based on historical spend, but as recognition of emotional memory. Similarly, a digital nomad who regularly books short stays during off-seasons could be rewarded with discounts and curated value-added services, such as co-working access or local community events, that enhance their perceived return on investment. Personalization systems should also consider travel motivation, life stage, loyalty sentiment and variation in booking behavior. A young solo traveler may be more responsive to cost-effective bundled experiences, while a frequent family traveler might value upgrades in flexibility or dining credits more than minor savings. Hospitality brands should also recognize and reinforce nonmonetary loyalty signals, such as social media engagement, positive reviews or participation in brand storytelling. Integrating these forms of soft engagement into personalization logic can humanize offers and build reciprocal relationships that extend beyond pricing incentives. Personalization should align economic recommendations with the brand’s core values to support brand advocacy. Offers that reflect a commitment to sustainability, local partnerships, wellness or cultural preservation can deepen guest connection, especially among value-conscious travelers who prioritize ethical and responsible consumption. For instance, a resort might personalize dining or excursion suggestions that highlight local artisans or farm-to-table experiences, reinforcing economic value and the guest’s identity as a conscious traveler. Therefore, economic personalization in hospitality must evolve from algorithmic optimization to narrative and emotional contextualization. By combining behavioral economics, personalization ethics and brand storytelling, hospitality providers can offer guests price-based incentives and emotionally resonant, context-aware experiences that cultivate long-term trust and brand advocacy.

While the operational backbone of hospitality, such as room readiness, service response times and resource deployment, remains fundamental to guest satisfaction, personalization in this area often lags behind more visible touchpoints like digital interfaces or pricing offers. Despite advances in AI-driven optimization behind the scenes, most operational personalization fails to meaningfully or perceptibly reach the guest. This disconnect between back-end efficiency and front-end experience becomes especially noticeable during peak seasons, delays or unanticipated service interruptions. Guests may perceive slow service, lack of coordination or generic delivery, even when systems operate at maximum throughput, because operational intelligence is not translated into personalized, empathetic guest interactions. For example, a guest returning to their room after a long day may find housekeeping still in progress, without being notified or offered an alternative solution. Similarly, guests waiting for in-room dining may receive no feedback on estimated delivery time, even when systems internally track queue lengths and kitchen workload. In such cases, the absence of proactive, contextual communication diminishes the guest’s perception of control, care and responsiveness. More critically, guests often interpret operational silence as indifference, not realizing that service optimization is happening silently in the background. Hospitality experiences are not judged solely by outcomes, but by how well those outcomes align with individual expectations and momentary needs. Two guests receiving the same service, say, a 20-min room cleaning or a late check-in, may perceive it differently based on their context, emotional state or the degree of transparency and personalization in service delivery. This underscores the need to bridge operational excellence with guest-facing personalization to enhance the quality of the real-time experience (Helmefalk et al., 2025).

To achieve this, personalization must incorporate real-time operational data into communication flows (Helmefalk et al., 2025). Guests should be notified of service progress, estimated wait times or disruptions in a tone and format suited to their profile through push notifications, voice updates or personalized screen displays in smart rooms. For instance, informing a guest that their room will be ready in 15 min and offering a complimentary beverage in the lounge creates a sense of anticipation rather than frustration. Similarly, integrating guest preferences for cleaning schedules, dining hours or ambient settings (temperature, lighting and entertainment) can significantly improve how services are perceived and received. Another improvement lies in offering time-of-day adaptive services. If a guest consistently requests extra towels in the morning or prefers late-night dining, systems should learn and anticipate these behaviors, queuing services accordingly. Smart room technologies can be used for operational convenience and to reflect known preferences, adjusting room ambiance before check-in, configuring entertainment systems or even preloading welcome messages or curated playlists. These touches, though subtle, create a profound sense of being “expected,” which is a hallmark of luxury and attentive hospitality. Therefore, translating operational excellence into visible, guest-facing value reinforces the feeling that the brand is responsive and invested in the individual’s well-being. It elevates service from merely functional to emotionally resonant, ensuring that operational effort is not lost in invisibility but is integrated into the guest journey. By designing systems communicating operational insights in contextually relevant and personalized ways, hospitality providers can convert routine logistics into moments of trust, delight and service distinction.

The emotional-cognitive personalization dimension focuses on understanding and responding to a guest’s evolving emotional and cognitive state across touchpoints throughout the service journey. Traditional personalization mechanisms, primarily driven by static data points such as booking history, demographic information or preference tags, fall short because they cannot adapt in real time to emotional transitions. Guests may experience a cascade of emotions even within a single interaction, from excitement at the prospect of a vacation to anxiety during check-in delays, frustration over unexpected charges or relief after a problem is resolved. When the personalization tone fails to reflect these emotional shifts, it creates a sense of detachment and impairs the overall service experience, even if the functional delivery is flawless. This disconnect becomes especially pronounced in high-stress or emotionally sensitive contexts. For example, a guest dealing with a lost reservation or billing error might receive templated reassurance that fails to match the intensity or urgency of their concern. Conversely, a guest expressing enthusiasm in a pre-arrival chat may receive a generic response, squandering an opportunity for deeper engagement. In such scenarios, a failure to align communication tone with emotional state can inadvertently signal indifference, reducing satisfaction and weakening emotional attachment to the brand. Moreover, most hospitality systems cannot interpret affective cues from guest inputs, such as tone of voice, sentence structure, typing speed or linguistic sentiment, leaving a critical gap in real-time emotional adaptation (Meng et al., 2025) in emotionally charged industries like hospitality, where outcomes and perceived empathy shape service quality judgments, such gaps can undermine trust and loyalty.

To address these challenges, emotional-cognitive personalization must be proactive, nuanced and situationally adaptive. Systems should be designed to detect and interpret sentiment and intent embedded in user-generated inputs via live chats, voice queries, feedback forms or facial recognition at kiosks. For instance, a guest expressing disappointment in a message should trigger a calm, empathetic response accompanied by compensatory gestures, while a guest enthusiastically planning a celebration stay might benefit from upbeat, personalized prompts about premium packages or local events. Moreover, personalization must go beyond adjusting content and consider delivery tone, pace and style. A distressed guest may prefer slow-paced, concise and reassuring communication, while a relaxed leisure traveler might respond positively to humorous, energetic prompts. This requires hospitality AI systems to dynamically shift their interaction strategies, adapting not just what is offered, but how it is offered. Furthermore, emotional-cognitive personalization can be enhanced by integrating memory and learning over time. Recognizing repeat guests’ past emotional pain points or high points and adjusting future interactions accordingly adds depth to personalization and fosters a sense of being genuinely understood. For example, a guest who previously expressed anxiety about travel delays could receive proactive notifications and flexible transportation or check-in options, signaling care beyond the transaction. Therefore, emotional-cognitive personalization allows hospitality providers to transform AI from a functional tool into a responsive companion, capable of mirroring human emotional intelligence.

Causal loop diagram (Figure 2) illustrates the dynamic feedback system of emotionally intelligent AI personalization in hospitality settings. It captures how systems that understand emotional states can create immersive, tailored experiences while warning of systemic imbalances caused by fatigue, bias and over-personalization. The system’s heart lies in Reinforcing loop R1, representing a positive feedback cycle between emotionally intelligent AI systems and guest satisfaction. As sentiment detection accuracy improves, LLMs trained on nuanced affective corpora (e.g. GPT-4 with emotionally tagged dialogues or transformers fine-tuned with hospitality-specific data sets) enable a deeper analysis of the guest’s emotional state. This insight allows for deeper personalization of sentiment, such as adjusting a chatbot’s tone from formal to friendly, or providing context-sensitive offers, like a spa treatment suggestion after detecting stress in messages (e.g. “I had a long travel day”). As emotionally adaptive personalization increases, perceived empathy also rises, directly boosting guest satisfaction. Satisfied guests are more likely to provide high-quality feedback through reviews, chats or in-app surveys. This feedback fuels the learning of AI personalization systems, enabling GenAI to fine-tune its outputs further. This self-reinforcing cycle continues to improve, resulting in elevated brand affinity and loyalty.

Balancing loop B1 captures the risks of expectation sensitivity and perceived algorithmic bias. As personalization becomes more emotionally attuned, guests’ expectations rise. However, when GenAI misinterprets context or offers culturally irrelevant suggestions (e.g. recommending a wine-tasting tour to a teetotaler or suggesting a Western breakfast to a South Asian guest), guests may feel algorithmic bias. This reduces perceived empathy, breaks the emotional connection and lowers guest satisfaction. The dissatisfaction then weakens brand trust and emotional loyalty, reducing engagement and ultimately diminishing the availability of high-quality feedback. This loop reveals a critical insight: more personalization does not guarantee greater satisfaction unless it is aligned with cultural intelligence and contextual sensitivity. Balancing loop B2 focuses on the cognitive burden experienced by users interacting with dynamic GenAI interfaces. When personalization interfaces (e.g. app dashboards, chatbots) update in real time or overwhelm users with data (recommendations, offers and mood-based messages), they increase sensitivity to cognitive load. This feeling of mental strain or overwhelm leads to reduced guest engagement and heightens privacy concerns. For example, a guest who receives a perfectly timed message like “We noticed your heart rate dropped. Would you like a relaxation massage?” may find it intrusive unless trust has been established. As privacy sensitivity increases, guests become reluctant to share data, thereby limiting the AI’s ability to learn from personalized experiences. The personalization system weakens, lowering sentiment adaptation quality and disrupting the R1 loop. Balancing loop B3 highlights the problem of repeated prompts and saturation of personalization. If not regulated, LLM-based systems may make repeated offers, such as dining reservations, room upgrades or review requests. While these may appear helpful, they can cause personalization fatigue, where guests feel overwhelmed. As a result, guest engagement declines and feedback loops become sparse, starving the personalization engine of inputs and reducing overall AI effectiveness. This loop is especially relevant in luxury or wellness resorts, where over-prompting can disrupt the desired sense of tranquility. An overly talkative chatbot or emotionally misaligned prompt (“Don’t forget to leave a review!” after a guest complains about poor service) can cause guests to disengage entirely.

For hospitality practitioners and system designers, the causal loop model offers actionable insights for designing emotionally intelligent, context-sensitive AI ecosystems. Emotionally and cognitively enriched personalization systems should be able to interpret subtle affective cues from guests, such as tone, linguistic style, typing rhythm or facial expression and respond in ways that convey empathy, warmth and situational awareness. By leveraging GenAI and LLMs, hospitality organizations can create adaptive concierge bots and service assistants that dynamically adjust their tone, language and emotional intensity to match the guest’s mood and intent. Such GenAI-driven interfaces can transform digital exchanges into emotionally attuned dialogues that strengthen relational trust and humanize technology-mediated service delivery. Embedding feedback loop logic within personalization architectures allows AI systems to continuously monitor engagement patterns and recalibrate interaction intensity, avoiding excessive messaging or over-personalization that may lead to cognitive fatigue or perceived intrusion. Integrating cultural intelligence within GenAI frameworks is equally vital. When trained on culturally nuanced data sets, LLMs can generate contextually appropriate recommendations and communication styles that respect regional preferences, values and sensitivities, thus enhancing inclusivity and cultural resonance in guest interactions. Continuous learning from user sentiment data derived from chats, online reviews, voice tone analysis and even subtle behavioral indicators enables AI systems to refine the quality of personalization over time. Such real-time data loops help transform one-way automation into reflective, self-correcting ecosystems. Allowing guests to actively shape their personalization experiences by choosing communication styles, content themes (such as wellness, efficiency or indulgence) and data-sharing preferences further reinforces trust, transparency and perceived control. Regular auditing of personalization outputs is also essential to identify and correct implicit biases that could compromise fairness or emotional alignment. Consistency in emotional tone, narrative coherence and personalization logic across all touchpoints, hotel websites, mobile apps, in-room assistants and physical concierge systems ensures a unified, brand-consistent experience that feels authentic and emotionally coherent. Human collaboration remains at the heart of this design philosophy. Involving frontline staff in co-developing and supervising GenAI-driven personalization protocols ensures that AI complements rather than replaces human empathy, intuition and adaptability.

While the aggregate dimensions identified through the integrative review provide a comprehensive understanding of the experiential landscape shaped by generative AI and large language models (LLMs), they are not directly amenable to empirical testing using survey-based analysis. Accordingly, this study translates these higher-order experiential dimensions into a mechanism-oriented conceptual framework (Figure 3) comprising empirically testable constructs, mediating processes and moderating conditions. This translation follows a theory-building logic in which broad experiential domains are decomposed into underlying interactional, cognitive and evaluative mechanisms that explain how AI-driven personalization influences guest outcomes. The dimension of Cultural Experiences and Social Dynamics is operationalized through conversational naturalness and human-likeness and co-creation quality, which capture the ability of GenAI-enabled systems to facilitate culturally resonant, socially embedded and participatory interactions. To account for cross-cultural heterogeneity in personalization effects, cultural orientation is incorporated as a moderating variable, allowing cultural experiences to be examined as interactional processes rather than static attributes. The Digital Interaction and Technological Affordance dimension informs the inclusion of generative AI capability, LLM-driven emotional adaptability and personalization fit. Extending beyond traditional technology acceptance perspectives, technological affordance is conceptualized as the system’s capacity to generate context-aware, emotionally adaptive and personalized content, with personalization fit reflecting guests’ perceived congruence between AI-generated outputs and situational needs. The dimension of Economic Dynamics and Brand Advocacy is reflected in the framework’s outcome structure. Economic evaluations such as perceived value and price fairness are embedded within perceived experience quality, which captures guests’ holistic assessments of AI-enabled service encounters and subsequently influences guest satisfaction and brand loyalty. Experiential Quality and Service Delivery is conceptualized as a process-level dimension shaping personalization fit and co-creation quality, rather than as a direct outcome, thereby preserving experiential richness while avoiding construct redundancy. The Emotional–Cognitive Experience dimension is operationalized through emotional engagement and personalization fatigue/cognitive overload, capturing both positive and adverse psychological responses to AI-driven personalization. To reflect individual differences and ethical sensitivities, trust propensity and privacy concern are incorporated as moderating conditions influencing experience evaluations.

This research contributes theoretically to the emerging literature on GenAI and LLM-driven personalization by moving beyond static, technology-centric explanations toward a dynamic, human-centered understanding of personalization in hospitality service systems. It advances personalization theory by reconceptualizing GenAI-enabled personalization as a multi-layered socio-technical system rather than a unidirectional technological intervention. Existing personalization research primarily emphasizes accuracy, relevance or efficiency of algorithmic outputs. In contrast, the proposed framework (Figure 3) integrates Generative AI capabilities, LLM-driven emotional adaptability, conversational naturalness and algorithmic transparency as foundational antecedents that indirectly shape personalization outcomes. This theoretical shift foregrounds how personalization is enacted through interactional and emotional processes, rather than merely what is delivered by AI systems. The framework extends human–AI interaction and service-dominant logic by introducing personalization fit and co-creation quality as central mediating mechanisms. By theorizing personalization as a co-created process between guests and GenAI systems, this bridges gaps between AI research and service theories that emphasize value co-creation, relational exchange and experiential alignment. This contribution challenges instrumental views of AI and positions LLM-based systems as active relational agents within hospitality service encounters (Dwivedi et al., 2025). This also contributes to affective and cognitive theories of technology use by explicitly modeling emotional engagement and personalization fatigue/cognitive overload as parallel, competing pathways. Prior research often treats positive engagement as the dominant outcome of AI personalization. The inclusion of fatigue and overload introduces a balancing logic, demonstrating that emotionally intelligent GenAI can simultaneously generate value and strain. This dual-pathway perspective advances theoretical understanding of the nonlinear and sometimes paradoxical effects of advanced personalization technologies. By incorporating trust propensity, privacy concern and cultural orientation as moderating conditions, the framework advances context-sensitive and cross-cultural theorization of GenAI personalization. Rather than assuming universal effects, the study shows that personalization effectiveness is contingent upon individual dispositions and cultural norms. This contribution responds directly to calls for culturally grounded AI theories and moves personalization research beyond Western-centric and technologically deterministic assumptions. Furthermore, this research contributes methodologically and conceptually by integrating systems thinking into GenAI personalization research. The framework highlights reinforcing and balancing feedback mechanisms linking experience quality, satisfaction and brand loyalty, thereby extending static causal models commonly used in hospitality research. This systems-based theorization provides a foundation for future longitudinal, configurational and simulation-based studies examining the evolution of GenAI personalization.

GenAI personalization should be implemented as an adaptive rather than excessive service strategy. Hospitality managers are encouraged to calibrate the intensity of personalization by dynamically regulating message frequency, recommendation depth and interaction length in response to real-time guest engagement signals (Helmefalk et al., 2025). Behavioral indicators such as repeated dismissals, delayed responses or declining interaction depth can serve as early warnings of personalization fatigue and cognitive overload. Service interfaces, including virtual concierges, chatbots and in-room assistants, should be designed to exhibit emotional intelligence by adjusting tone, pacing and content based on detected emotional cues. Guests experiencing stress or frustration benefit from concise, reassuring communication, whereas leisure-oriented guests may prefer more exploratory, expressive interactions, making emotion-aware prompt engineering and sentiment detection essential components of LLM deployment. Personalization engines should also incorporate cultural intelligence that extends beyond language translation to reflect cultural norms, values and travel motivations. Training GenAI systems on culturally diverse data sets and enabling contextual adjustments based on guest origin, travel purpose and local customs can enhance perceived authenticity, trust and relevance. Transparency and user control play a critical role in mitigating privacy concerns and strengthening acceptance of GenAI-enabled services. Clearly explaining why recommendations are generated and allowing guests to customize interaction styles, data-sharing preferences and thematic focus can increase perceived autonomy and trust. Operational intelligence used internally for scheduling, housekeeping or resource allocation can also be leveraged to create guest-facing value through proactive, personalized communication, such as notifying guests of service timelines or offering alternatives during disruptions. Aligning GenAI-driven personalization with brand values and long-term relationship-building goals is essential, as emotionally resonant gestures, recognition of guest milestones and context-aware value offerings are more effective in fostering satisfaction and loyalty than purely transactional or price-driven recommendations.

While this research proposes a conceptual framework and causal loop diagram grounded in an extensive literature review and expert insights, several limitations suggest productive directions for future research. Although the framework and CLD articulate dynamic relationships among GenAI capabilities, emotional processes and personalization outcomes, they remain conceptually specified rather than empirically parameterized. Future studies could operationalize the proposed constructs and empirically test the causal relationships using quantitative approaches such as PLS-SEM, longitudinal surveys or system dynamics simulation. In this regard, future research may address questions such as FRQ1: How do reinforcing and balancing feedback loops empirically evolve across repeated guest–AI interactions? And FRQ2: Which feedback mechanisms most strongly predict long-term satisfaction and loyalty?. While the expert interviews informing the framework provide depth and domain relevance, they reflect a limited range of professional perspectives. Extending this work through a multistakeholder inquiry that includes guests, frontline employees, technology vendors and policymakers would enrich the understanding of GenAI personalization in practice. Accordingly, future research could explore FRQ3: How do guests’ lived experiences of GenAI-driven personalization align or diverge from managerial and expert assumptions? and FRQ4: How do frontline employees negotiate emotional labor when working alongside emotionally intelligent AI systems?. In addition, although cultural orientation is theorized as a key moderating condition, the framework has not yet been empirically validated across diverse cultural and market contexts. Comparative and cross-cultural studies could examine whether personalization dynamics vary across regions, service formats and guest segments, raising FRQ5: How do cultural values shape perceptions of emotional authenticity, privacy and trust in GenAI-enabled hospitality services?. Moreover, while the CLD captures dynamic processes at a conceptual level, it does not quantify threshold effects, such as the point at which personalization intensity leads to fatigue or perceived intrusion. Experimental and simulation-based approaches could help identify such critical boundaries, prompting FRQ6: At what levels does emotionally adaptive personalization shift from value-enhancing to value-eroding?. Furthermore, ethical governance and transparency mechanisms embedded within GenAI personalization warrant deeper investigation. Future research could examine how explainable AI, user-control interfaces and governance structures influence trust and sustained engagement over time, as reflected in FRQ7: How does transparency in GenAI decision-making affect guests’ willingness to share data and maintain long-term engagement?.

This research advances understanding of GenAI-driven personalization by addressing two interrelated objectives: first, to synthesize and assess existing research, with attention to emotional, cultural and relational mechanisms in human–AI interactions; and second, to integrate these insights into a coherent conceptual framework that explains the dynamic pathways shaping experiential and relational outcomes. In relation to RO1, the critical synthesis reveals that the literature on GenAI-enabled personalization has expanded rapidly but unevenly. While foundational models from technology adoption, service quality and consumer behavior continue to dominate, they only partially capture the complexities introduced by emotionally adaptive and conversational AI systems. Recent studies increasingly acknowledge affective responses, trust formation and cultural sensitivity, yet these elements remain inconsistently theorized and operationalized. Methodologically, the field demonstrates growing sophistication through the use of computational analytics, machine learning and large-scale user-generated data; however, this progress is often accompanied by limited interpretive depth and insufficient linkage to human-centric theories. These patterns indicate that current research reasonably well explains what personalization does, but is less equipped to explain how and why GenAI-driven personalization shapes long-term relational outcomes. Addressing RO2, this research integrates dispersed insights into a unified conceptual framework that explicates how GenAI capability, emotional adaptability, interaction quality and transparency influence personalization outcomes through mediating and moderating mechanisms. By adopting a systems perspective, the framework highlights personalization as a dynamic process governed by reinforcing and balancing feedback loops, in which emotional engagement and satisfaction coexist alongside risks such as cognitive overload, fatigue and perceived bias. This integration demonstrates that personalization effectiveness is not linear but contingent on evolving contextual, cultural and psychological conditions.

Dr Rachana Jaiswal is an Associate Professor in the Department of Business Administration, School of Management Sciences, Central University of Jharkhand, Ranchi, Jharkhand, India.

This research received no external funding.

Rachana Jaiswal conceptualization, supervision, methodology, software, writing - reviewing and editing, formal analysis, revision, review, project administration; Shashank Gupta conceptualization, methodology, data curation, writing- original draft preparation, revisions, review, visualization, investigation; Po-Ju Chen conceptualization, supervision, validation, reviewing and editing, revision, review, project administration.

The supplementary material for this article can be found online.