AI-Enabled Mobile Food-Ordering Apps and Customer Experience: A Systematic Review and Future Research Agenda

Abstract

Artificial intelligence (AI) is reshaping mobile food-ordering apps, yet its impact on customer experience (CX) has not been fully mapped. Following systematic review guidelines (PRISMA 2020), a search of SCOPUS, Web of Science, ScienceDirect, and Google Scholar in March 2025 identified 55 studies published between 2022 and 2025. Since 2022, research has expanded from intention-based studies to include real-time app interactions and live app experiments. This shift has helped to identify five key CX dimensions: (1) instrumental usability: how quickly and smoothly users can order; (2) personalization value: AI-generated menus and meal suggestions; (3) affective engagement: emotional appeal of the app interface; (4) data trust and procedural fairness: users’ confidence in fair pricing and responsible data handling; (5) social co-experience: sharing orders and interacting through live reviews. Studies have shown that personalized recommendations and chatbots enhance relevance and enjoyment, while unclear surge pricing, repetitive menus, and algorithmic anxiety reduce trust and satisfaction. Given the limitations of this study, including its reliance on English-only sources, a cross-sectional design, and limited cultural representation, future research should investigate long-term usage patterns across diverse markets. This approach would help uncover nutritional biases, cultural variations, and sustained effects on customer experience.

1. Introduction

Mobile food delivery has evolved from an occasional convenience to a routine part of urban life. Global sales are projected to rise from USD 156 billion in 2024 to USD 173 billion in 2025 and to exceed USD 213 billion by 2030 [1]. A weekday lunch ordered on a phone in under 30 s illustrates how each additional purchase becomes a critical digital touch-point where experience quality can win or lose platform loyalty.

Whereas customer experience (CX) spans every interaction, online CX covers only digital encounters. Accordingly, this review focuses on AI-mediated online CX within mobile food-ordering apps (MFOAs). Understanding how these market forces translate into day-to-day user experiences requires examining the AI capabilities embedded in today’s apps.

Real-time personalization, dynamic pricing, and last-mile routing now shape multisensory cues (visual menus, haptic order confirmations), emotional responses (excitement, anxiety), and social behaviors (shared order-tracking chats) [2]. Classical adoption frameworks like the Technology Acceptance Model (TAM) and Unified Theory of Acceptance and Use of Technology (UTAUT) assume that all users interact with the same kind of static interface. However, modern apps work differently. They use haptic feedback to confirm orders, algorithms to instantly suggest meals, and generative-AI tools that notice delivery delays, apologize to the user, offer a free item, and update the delivery time automatically. These technologies raise new challenges such as balancing convenience with privacy, or efficiency with fairness [3]. Addressing them calls for an integrated view of the five CX factors: instrumental usability, algorithmic-personalization value, affective engagement, data trust and procedural fairness, and social co-experience.

Existing syntheses capture these dynamics only partially. Many studies still test TAM-style intentions, analyze single-platform datasets, or rely on hypothetical scenarios. The wide-ranging review by [4] mapped general online CX across retail and service contexts but did not examine AI-enabled, post-purchase dining journeys or integrate the five factors above within a unified framework. By concentrating on MFOAs and explicitly linking instrumental usability, algorithmic-personalization value, affective engagement, data trust and procedural fairness, and social co-experience, the present study addresses these unresolved questions. Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, this systematic review charts publication trends, theoretical lenses, and methods; synthesizes findings on the five factors; exposes conceptual and methodological blind spots; and proposes a research agenda for scholars, platform designers, restaurateurs, and policymakers. The resulting framework aims to advance food-delivery experiences that remain satisfying, transparent, and economically sustainable.

2. Theoretical Background

The first wave of mobile-commerce studies relied on acceptance and service-quality frameworks such as the Technology Acceptance Model (TAM), the Unified Theory of Acceptance and Use of Technology (UTAUT), Service Quality (SERVQUAL), and Mobile Service Quality (M-S-QUAL). These models suited early mobile apps, where a user simply chose an item and the app relayed the order to the restaurant. Users made every decision themselves, and the app acted as a basic courier. Ideas like usefulness, ease of use, and reliability described that experience well in the late 2000s. Today’s food ordering apps are very different. They run on ensemble recommenders—several prediction models that work together to suggest dishes—conversational agents that understand natural language, and real-time pricing engines that change delivery fees on the fly. The result is an AI-powered service, like a personal assistant who remembers you prefer ramen on rainy days, reminds you at 7 p.m., and hides items when traffic is heavy. Models built for static apps struggle because decisions are now shared between the user and the AI, and the interface changes in real time.

Researchers have tried to adapt other perspectives, but coverage remains fragmented. Studies using stimulus–organism–response logic show that push notifications and limited-time bundles boost arousal and purchase intent [5], yet they seldom track satisfaction beyond checkout. Work on the personalization–privacy paradox finds that detailed data improves recommendations but heightens worries about being watched [6,7,8], especially in location-based ordering where address, diet, and payment details are required. News reports show rising calls for algorithm transparency at firms such as Deliveroo and Uber Eats [9], while research on fairness still centers on narrow issues like equal outcomes or users’ sense of control. Other studies use cognitive–affective–conative chains—the think-feel-do sequence—to show how the perceived AI skill builds trust and then repurchase [10]. Consumption-value theory adds that the pleasure of browsing appetizing images sits alongside the practical need for quick delivery [11]. Across these streams, AI qualities such as predictive accuracy, conversational naturalness, and routing efficiency are often treated as single factors instead of evolving layers that shape every visit. Personalization can raise convenience, but too many tailored options can also exhaust users and lower satisfaction.

Drawing on the hospitality-robotics USUS model [12], this article pulls these findings into one view that treats AI both as a service employee and as part of the service setting. Five linked dimensions frame this view. First, instrumental usability covers how quickly diners learn the interface, navigate, and recover from errors, especially when screens adapt mid-session. Second, algorithmic-personalization value refers to how relevant, fresh, and surprisingly helpful the suggestions are, thanks to multiple prediction models working together. Third, affective engagement spans enjoyment and trust, but also the newer feeling of algorithmic anxiety that appears when nudges feel too strong. Fourth, data trust and procedural fairness deal with comfort in sharing data, clarity of price changes, and perceived justice in rankings and fees. Finally, social co-experience includes shared baskets, ratings-based social proof, and reputation cues that echo offline word-of-mouth. AI can either strengthen or weaken each area: a well-tailored home screen can raise relevance [13], but hidden surge pricing can erode fairness [14].

Three gaps need attention: the lack of longitudinal studies that follow satisfaction beyond the first purchase, the under-representation of users with low digital literacy or high privacy concerns, and the limited measurement of algorithmic fairness in consumer metrics. By framing AI at once as both employee and environment, the proposed framework offers a scaffold for future theory building and a common language for practitioners who must stage responsible, value-creating AI encounters.

3. Review Methodology

This review follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework. Each stage is systematically organized under clear sub-headings to enhance the transparency and traceability of the decision-making process (see Figure 1 for the flow diagram). Although a hybrid method that combines quantitative bibliometric analysis with qualitative systematic review techniques [15] was initially considered, the final collection of studies obtained through screening is recent and lacks sufficient citations. This limitation makes quantitative network analysis unreliable at present. Thus, a purely qualitative PRISMA-based systematic review provides deeper and more accurate thematic insights suitable for the current developmental stage of research in this field.

3.1. Identification

The search was executed on 31 March 2025 in SCOPUS, Web of Science, and ScienceDirect, with a complementary hand sweep of Google Scholar. A Boolean string (a keyword line that joins terms with AND/OR logic) that combined terms for customer experience, artificial intelligence, and mobile food-ordering apps (reported verbatim in Supplementary File S1, Table S1) returned 278 records. Searches were limited to English-language publications. Reporting the full string and database mix ensures that future researchers can replicate or extend the search with minimal effort, meeting PRISMA’s transparency ideal.

3.2. Duplicate Removal Procedure

All references were exported into EndNote X20. The software’s ‘Find Duplicates’ routine compared the DOI, title, and author fields. Flagged items were deleted, and a manual, record-by-record inspection confirmed the result. The second pass in Covidence (an online screening tool that flags near-duplicate titles) captured partial-title mismatches that EndNote can overlook. Using two tools added an extra layer of rigor. These combined steps eliminated 53 duplicates and reduced the pool to 225 unique records.

3.3. Screening I—Accessibility Check

Before relevance screening, each of the 225 records was checked for full-text availability. Nineteen articles could not be retrieved despite institutional subscriptions and repeated author contact. Because inaccessible papers cannot be appraised for quality, they were removed, leaving 206 accessible articles. All decisions were made by the first author and resolved with the co-authors whenever any uncertainty arose.

3.4. Screening II—Title and Abstract Check

The remaining 206 articles were assessed on four inclusion criteria: the paper had to be written in English; be scholarly in nature (empirical, conceptual, methodological, or review); examine customer experience; and contain an AI-enabled feature such as chatbots, personalized recommendations, or predictive algorithms. Applying these rules excluded 81 papers and retained 125 for full-text eligibility review. This step prevented obviously irrelevant studies from consuming resources in later stages while limiting subjective bias because the criteria were explicit and repeatable.

3.5. Eligibility Assessment

Full texts of the 125 retained studies were read in detail. Articles were excluded if they strayed from a customer experience, if they lacked peer-review status, if their methodological reporting was insufficiently transparent to permit appraisal, or if they examined only back-end logistics without a user-facing interface. These exclusions removed 67 further articles and yielded a final corpus of 55 studies. Stating these rules and their rationale—topic relevance, scholarly quality, methodological clarity, and user focus—helps readers understand how the review balances inclusivity with rigor.

3.6. Data Extraction and Thematic Synthesis

Each eligible study was entered into an Excel coding matrix, chosen for its ease of sharing and auditability; formula cells were locked to prevent accidental edits. Data were classified under five domains adapted from prior customer experience research: instrumental usability, algorithmic-personalization value, affective engagement, data trust and procedural fairness, and social co-experience. For every study, the matrix also captured contextual descriptors such as publication year, geographical setting, sample size, design, platform examined, theoretical lens, and any declared funding or conflicts of interest. Missing information was left blank rather than imputed. Two reviewers independently checked coding decisions and resolved discrepancies by consensus, ensuring reliability.

Given the early stage of research in this area, the review synthesized findings using the five-factor CX framework rather than the broader TCCM grid, where TCCM refers to mapping studies by their theories, characteristics, contexts, and methods, in order to preserve thematic depth and conceptual clarity.

3.7. Citation Snowballing

To minimize the risk of missing relevant work, forward- and backward-snowballing was carried out in Google Scholar and Scopus. For each of the 55 eligible articles, reference lists were screened (backward step) and “cited-by” records up to April 2025 were inspected (forward step), using the same inclusion criteria applied in Section 3.4 and Section 3.5. No new eligible records were found, confirming that the database search had captured the full corpus for this review.

3.8. Risk-of-Bias Assessment

The reviewers independently assessed each study across five domains: sampling, measurement, reporting, confounding, and ethical procedural transparency, using tailored questions described in Supplementary File S2. Ratings of low, moderate, or high risks were agreed upon through discussion. The inter-rater agreement was substantial (Cohen’s κ = 0.74) [16]. Following standard practice [17], the worst rating across the domains determined each study’s overall quality. Due to diverse study designs and outcomes, no meta-analysis or heterogeneity testing was conducted. Publication bias was qualitatively assessed via multi-database searches and reference chaining; results appear as a color-coded summary in Supplementary Figure S2.

4. Literature Review

This section first traces publication trends and dominant methodological choices, then synthesizes key theories and empirical findings through a five-factor customer experience (CX) lens.

4.1. Publication Trends

Academic work on AI-enabled mobile food ordering apps has accelerated markedly over the last four years. Ten relevant studies were published in 2022, rising slightly to eleven in 2023. Output then more than doubled to twenty-two publications in 2024, and the first four months of 2025 have already produced twelve additional papers. This post-2022 surge shows that scholarly interest is now catching up with the rapid commercial expansion of the sector, opening a timely window for deeper theory building and stronger collaboration between researchers and industry practitioners (Figure 2).

4.2. Methodological Choices and Data Sources

Research on AI-enabled mobile food-ordering apps still leans most heavily on attitudinal techniques. Almost half of the papers (49%) use online or paper questionnaires analyzed with structural-equation models—statistical tools that map links between variables—often in the PLS-SEM variant (partial least-squares SEM) seen in [18,19]. Behavior-focused work forms the next largest share: 22% mine user-generated or system logs, such as App Store reviews and clickstreams (sequences of user clicks), to reveal real-time pain points and delights [20,21,22]. Another 11% adopt mixed designs that pair those behavioral traces with interviews or follow-up surveys, linking what users do with why they do it [23,24]. Controlled experiments and usability tests—eye-tracking in the lab or A/B trials inside live apps—also account for 11% but remain rare despite their strong causal power [25,26]. The small remainder consists of qualitative interviews and engineering case studies, each below five per cent (e.g., [27,28]). A summary of these study designs is presented in

Table 1. Study Design Summary (n = 55).

Design Type (Operational Definition)	Studies (n)	Share (%)
Surveys + SEM/PLS-SEM (attitudinal questionnaires analyzed with structural-equation or regression models)	27	49%
Log-mining and secondary behavioral data (mining App Store reviews, clickstreams, transactions, sensor logs)	12	22%
Mixed methods (behavioral traces paired with interviews, or follow-up surveys)	6	11%
Controlled experiments/usability tests (lab eye-tracking, online A/B, task-based tests)	6	11%
Qualitative interviews/grounded theory	3	5%
Engineering case studies/prototypes	1	2%
Total	55	100%

.

Together, log-data studies expose operational snags, surveys capture user feelings, and mixed methods connect the two; the scarcity of field experiments signals room for closer work between scholars and platform operators. Publication venues reflect this spread. As shown in

Table 2. Publication Outlets.

Top Publication Outlets	Number of Studies	ABDC (2019)
International Journal of Hospitality Management	2	A*
International Journal of Human–Computer Interaction	2	A
Journal of Retailing & Consumer Services	2	A
British Food Journal	3	B
Journal of Foodservice Business Research	3	C
Journal of Association of Arab Universities for Tourism & Hospitality	2	–
Conference Proceedings & Workshops †	6	–

Note. * “A-star” (A*) is the highest tier in the 2019 ABDC journal-quality list. “†” indicates the outlet is not listed in the 2019 ABDC journal-quality list.

, Top Publication Outlets (n = 55), articles appear in outlets from the ABDC A*–rated International Journal of Hospitality Management to the practice-oriented Journal of Foodservice Business Research (ABDC C), underlining both academic rigor and industry relevance in this fast-growing line of research.

4.3. Theoretical Anchors and Key Findings

Theories rooted in technology adoption still shape much of the research on AI-enabled food ordering; however, experience-centered lenses are increasingly visible. As summarized in

Table 3. Theoretical Frameworks and Models.

Theoretical Frameworks/Models	Frequency
Technology Acceptance Model (TAM)	9
Unified Theory of Acceptance and Use of Technology (UTAUT)	6
Expectation–Confirmation Theory (ECT)	4
Stimulus–Organism–Response (SOR) Framework	4
Trust–Risk Calculus Models	3
Privacy Calculus Theory	2
SERVQUAL/E-Service Quality Models	2
Grounded Theory (inductive)	2
Affective–Cognitive Dual-Process Models	1
Atheoretical/Data-Driven	12

, Most-Used Theoretical Frameworks, about half of the reviewed studies adopt either the Technology Acceptance Model (TAM), or its extension, the Unified Theory of Acceptance and Use of Technology (UTAUT). Examples include work on mobile usability [21], trust formation in Vietnamese delivery apps [29], and first-time loyalty intentions in India [18]. These models link usefulness and ease-of -use to adoption, confirming the field’s origins in classic acceptance logic.

Approximately one-third of the corpus turns to richer, socio-emotional perspectives. The Stimulus–Organism–Response (SOR) traces how push-notification cues trigger arousal and purchase intent [30]. Expectation Confirmation Theory (ECT) explains satisfaction shifts when service promises are met or missed in cross-border samples [31], while Trust–Risk Calculus models weigh privacy costs against convenience gains in location-based ordering [32]. These frameworks move beyond the “will they use it?” question to ask how feelings, expectations, and perceived risks shape customer experience.

A smaller segment, around one-fifth of the studies, reports data-driven results without an explicit guiding theory, such as panel analyses linking app-order shares to restaurant sales [22] or deep-learning evaluations of novel recommender algorithms [33]. This atheoretical stance highlights the pull of big data and machine learning but also underscores the need for stronger conceptual integration.

Overall, the frequent use of TAM and UTAUT shows that much of this research still focuses on understanding why people start using food-ordering apps. At the same time, the growing use of theories like SOR, ECT, and Trust–Risk models shows a shift toward exploring users’ feelings, expectations, and social interactions—key parts of the customer experience. The mix of theory-based and data-driven studies suggests that research is moving beyond just predicting app use. It is starting to explore how AI actually shapes what people think, feel, and do while using these platforms.

4.4. Thematic Review of Customer Experience with AI-Enabled Mobile Food-Ordering Apps

Building on the descriptive profile of the 55 extant studies, this section integrates their findings through the five-dimensional lens introduced earlier—instrumental usability, algorithmic-personalization value, affective engagement, data-trust and procedural fairness, and social co-experience. Although inspired by the USUS framework’s concern with embodiment, emotion, and co-experience, the present synthesis tailors each dimension to the unique affordances and tensions of AI-mediated meal ordering.

Table 4. Thematic Summary of Findings from the 55 Reviewed Studies.

Theme	Key Points Drawn from the 55 Studies
Instrumental Usability	Easy ordering and stable layouts remain essential: features like one-tap reordering and biometric payment continue to satisfy users. AI-based menu reordering improves relevance but can confuse users when dish positions change between visits (“orientation loss”). Less is more: showing 5–7 well-chosen dishes works better for clicks and satisfaction than longer, fully adaptive lists. Long-term effects are under-studied: only three studies follow how user perceptions of usability change as algorithms learn over time.
Algorithmic-Personalization Value	Relevance, novelty, and surprise influence five-star ratings more than price or delivery speed, according to review-mining studies. Heavy users see less variety over time known as “recommender flattening” which leads to browsing fatigue and smaller, less diverse orders. Counter-personalization tactics (random injection, diversity sliders) remain under-tested and yield mixed results. Long-term goals are overlooked: few studies connect recommender systems to outcomes like healthy eating or sustainability.
Affective Engagement	Conversational chatbots enhance perceived competence and user enjoyment during routine interactions but tend to underperform when handling complex or emotionally sensitive complaints that require empathy. Push-notification experiments show an arousal-to-annoyance shift after ~3 exposures in a single week. Anthropomorphic bot avatars can strengthen trust, yet eye-tracking evidence suggests they lengthen checkout times by diverting attention. “Algorithmic anxiety”, a discomfort with opaque recommendation logic, has been identified in only four studies, and its effect on user churn remains largely unexamined.
Data-Trust and Procedural Fairness	Users trade data for convenience until a privacy-overexposure threshold is crossed; afterwards, trust erodes sharply. Lack of transparency in surge pricing consistently triggers high levels of user backlash on social media, particularly during holidays or extreme weather conditions. Simulation studies detect nutritional bias: profit-maximizing recommenders steer price-sensitive segments toward calorie-dense items. Few papers test transparency interventions (e.g., cost-breakdown screens, explainable-AI pop-ups), despite their policy relevance.
Social Co-Experience	Cart-sharing and split-payment affordances raise enjoyment through shared agency yet appear in <20% of leading apps. Algorithmic social proof (“popular near you”) lifts conversion, but credibility drops when cues feel artificially inflated. Cross-platform influences (e.g., TikTok food trends, live unboxing videos) are increasingly shaping user expectations, yet these dynamics are rarely incorporated into customer experience models. Cultural differences moderate social influence effects: collectivist users respond more to family endorsements, while individualist users are more influenced by digital creators. However, large-scale cross-cultural studies remain scarce.

consolidates the empirical signals discussed below and indicates the weight of evidence supporting—or challenging—each proposition.

4.4.1. Instrumental Usability

Instrumental usability refers to how easily users can find, customize, and pay for meals through mobile food-ordering apps (MFOAs), with minimal mental effort. Across the reviewed studies, this factor consistently stands out as a strong predictor of post-purchase satisfaction [5,32,34]. Early work on digital convenience focused on static features such as clear menus, fast-loading pages, and secure checkout [34]. However, as adaptive AI systems now personalize app interfaces in real time, usability has become more dynamic and less predictable.

Recent large-scale analyses of user data confirm the ongoing importance of usability but also reveal new challenges. For example, ref. [18] found that for first-time users, ease-of-use influences loyalty more strongly than enjoyment or price. Similarly, studies of user reviews show that “easy navigation” remains the most common reason for high ratings [21,34]. However, some users report feeling disoriented when app layouts change between visits, especially when items are re-ranked or new promotions are added. Experimental work suggests that while algorithmic reordering can improve relevance, it may also increase mental effort and decision time [25]. Ref. [35] found that while clear navigation improves purchase likelihood, showing too many options at once can lead to confusion and longer decision-making.

Payment processes are also central to usability. Technical improvements like backend refactoring have been shown to reduce page loading times significantly [28], which directly improves user experience. Features like biometric log-in and stored payment details can make checkout faster and smoother [36,37]. Yet, this convenience does not benefit all users equally. For those concerned about data privacy, these same features can cause discomfort, reducing overall satisfaction even when the interface performs well [38]. One study found that when users believe their data could not be changed or tampered with, the link between usability and future purchases grows stronger [39], suggesting that trust in data security enhances the effect of usability.

The number of choices presented is another key issue. While recommender systems were once praised for offering a broad selection, recent evidence suggests that more is not always better. Shorter, well-curated lists lead to higher purchase rates than longer ones [40,41]. Studies using eye-tracking technology show that long lists increase mental effort and lead to decision fatigue [24,26]. Only a few studies have looked at how users adapt over time. Research by [42,43] shows that users tend to experience the most confusion early on, but this effect decreases after repeated use. These findings suggest that apps could benefit from simple onboarding instructions that explain why the menu changes.

Despite growing insights, experimental evidence remains limited. Only a small portion of the reviewed studies use real-world experiments to test usability, and very few connect it to other outcomes like delivery accuracy or customer complaints. Another gap is cross-device behavior. Many users start an order on one device, like a smartwatch, and complete it on another, like a phone or smart speaker. Yet, little research explores how switching between devices affects the user experience [44].

In summary, the literature suggests three key points. First, while AI brings new capabilities, basic usability principles—clarity, speed, and easy payment—still matter. Second, helpful recommendations can backfire if they create confusion or increase effort. Third, user trust in data handling affects how much value they get from usability improvements. As AI continues to evolve, future research should explore how different users experience adaptive interfaces over time and across devices.

4.4.2. Algorithmic-Personalization Value

Algorithmic personalization is a core innovation in mobile food-ordering apps. By collecting real-time signals—location, time of day, previous purchases, and even scrolling patterns—AI systems create menus, bundles, and price nudges that feel tailor-made for each diner. Deep-learning and combined (ensemble) recommender models now drive much of this work, learning individual “taste signatures” and ranking dishes accordingly [33,44]. Restaurant operators see clear financial gains: panel data from a multi-city chain show that items pushed to the top of the menu generate higher same-store sales [22].

Findings from the user perspective support the idea that personalization adds real value for both customers and businesses. Survey studies link personalized offers with higher satisfaction, stronger engagement, and a greater intention to reorder [19]. A large-scale analysis of millions of app reviews shows that relevance, novelty, and pleasant surprise influence five-star ratings more than discounts or delivery speed [20,45]. When suggested dishes match users’ dietary needs or moods, the system is seen as both capable and thoughtful, which increases affective trust [21].

However, these benefits may decrease over time. Many users experience what is called “recommender flattening”, where the app keeps showing the same familiar dishes. This reduces variety and makes the menu feel repetitive [29]. As a result, users can get bored and lose trust in the app’s suggestions. Some lab studies suggest that adding sliding discovery features or short-term new item highlights can help users explore more options and reduce this problem [23].

Short-term engagement improvements can hide longer-term concerns about health and ethics. Studies using mobile tracking show that late-night personalized messages often lead young adults to order high-calorie snacks [46]. Simulations also suggest that apps focused mainly on engagement may push cheaper, calorie-heavy meals to budget-conscious users, which can encourage unhealthy eating habits [47,48]. Some possible solutions—such as randomly adding different items, letting users adjust the variety settings, or showing rotating new options—try to bring back more choice and control. However, research results are mixed, and these tools are not widely used in practice [49,50].

Taken together, the evidence underscores both promise and peril. Personalized lists boost satisfaction when they deliver relevant surprises, but they risk narrowing choices, promoting unhealthy food, and eroding trust when novelty wanes. Sustaining value will require balancing short-run convenience with broader well-being. Future studies should follow users over time, test recommender goals beyond clicks—such as nutrition or sustainability—and adopt explainable AI tools that show how suggestions are made. Transparent designs and ethically aligned algorithms are likely to prove essential for keeping personalization valuable to diners and profitable for providers.

4.4.3. Affective Engagement

Affective engagement describes the mix of feelings, body reactions, and behaviors that arise while diners use AI-driven food-ordering apps. Under appraisal theory, these emotions appear when an on-screen event such as a personalized coupon, a changing delivery countdown, or a chatbot apology, seems to help or hinder a personal goal, pushing users either to continue or to pull back. A text-mining study of 2.1 million Google Play reviews shows how central emotion is: joyful phrases like “saved my lunch break” predict five-star ratings more strongly than price or speed [51]. Survey work based on the Pleasure–Arousal–Dominance model finds that pleasure and a sense of control, rather than sheer excitement, best explain the wish to reuse an app [52].

Positive feelings dominate most interactions. Chatbots that answer in warm, first-person language raise trust and competence scores, which in turn lift satisfaction and the intention to reorder; an experiment with 525 users found a path coefficient of β = 0.41 from empathetic wording to overall satisfaction, even after accounting for discount levels [53]. A survey of 632 shoppers in Saudi Arabia confirms that a pleasant emotional experience has a direct effect on repeat purchase, especially for expert users [54]. Small visual touches matter as well: animated confetti when an order ships or a speeding rider icon during preparation can trigger short spikes in physiological arousal that translate into larger tips and more social sharing. Value-matched price promotions show a similar pattern, raising cognitive involvement (β = 0.37) and advocacy intentions (β = 0.29) [55]. These findings point to clear economic gains from a well-timed, emotion-aware design.

Although less frequent, negative emotions can quickly and strongly affect user behavior. On an Indian platform, two promotional push alerts within 48 h increased orders by 14 per cent, but a third alert reversed the effect and shifted comments from “tempting” to “intrusive” [26]. Even small errors in delivery-time estimates can cause “algorithmic helplessness,” a stress reaction linked to app abandonment when delays repeat [56]. Survey data grounded in stress theory confirm this: a perceived threat from algorithms increases stress levels, which fully explains the drop in adoption under risk [57]. Social media analysis during price surges shows up to a tenfold spike in anger-coded emoji, highlighting how fairness concerns drive emotional reactions online [58]. During the COVID-19 pandemic, anxiety about dining risk reduced satisfaction and reuse intentions, even when other app features remained positive [59]. Emotions also shift rapidly. Excitement over a discounted bundle may turn to frustration if a courier is late, a swing described as “emotional whiplash”. Arousal leads to positive experiences only when paired with clear information or quick error correction, suggesting that attention-grabbing features need to be balanced with tools like live rider chat or clear explanations for price changes [26].

Important gaps remain. Few studies map emotions over long periods: passive-sensing research tracks phone moods for a year [60], yet no mobile-food study follows users beyond ninety days, leaving fatigue effects unclear. Most work still uses surveys or text sentiments; richer sensing methods—facial micro-expressions or heart-rate data—common in mental-health apps [61] are almost absent here. Cross-cultural insight is also thin. Early studies suggest collectivist diners value shared enjoyment, while individualists prize efficiency [62], and Indonesian coffee shop data show stronger links between emotion and loyalty under collectivist norms [63], but broad tests across markets are lacking.

Overall, the current evidence agrees that positive emotions boost satisfaction and spending, whereas poorly managed negative emotions can erase those gains. Designers should pair engaging visuals and promotions with clear information and quick recovery options to keep excitement from drifting into anxiety. Future research needs longer time frames, multi-modal emotion tracking, and deeper cross-cultural work to show how AI design choices shape diners’ feelings and loyalty over time.

4.4.4. Data Trust and Procedural Fairness

Data trust and procedural fairness form the ethical backbone of AI-driven food-ordering apps. Features like personalized menus, surge pricing, and one-tap biometric payments rely on the constant collection and use of detailed personal data—such as a user’s location, dietary preferences, past orders, and even how they scroll through the app. As this data use increases, so do the risks of privacy breaches, misuse, and unfair treatment. Regulators have started to respond. In Hong Kong, the Privacy Commissioner reported a 50 per cent increase in data breach reports from food delivery companies between 2022 and 2023 and urged users to be careful when sharing their address or payment information [64]. Survey research from five MENA countries also shows that trust in how companies manage data explains nearly a third of the connection between personalization and user satisfaction, showing that convenience depends heavily on strong privacy protections.

Price transparency forms the second fairness pillar. Fee multipliers that respond to demand, bad weather, or holiday peaks can triple costs within minutes. Companies defend these surges as efficient capacity tools, but many users label them “price gouging” and question the fairness of the algorithms [65]. A U.S. survey found that a perceived unfair price cut reorder plans by 18 per cent in the following week [66]. Simulations based on two million past orders raise another concern: profit-focused recommendation systems tend to steer budget-conscious users toward cheaper and calorie-dense meals, a pattern that could worsen health inequalities [67].

Early tests suggest transparency can help, yet it is still rare. On one U.S. platform, a cost-breakdown pop-up that named distance, demand, and weather reduced fairness complaints by 27 per cent without lowering sales. In Indonesia’s Railfood app, showing the queue length and driver-allocation logic raised perceived procedural justice by nearly half a standard deviation and lifted satisfaction by 11 per cent [14]. Nutrient-based explanations such as “recommended because it meets your protein goal” also improve fairness ratings in healthy-eating contexts [67]. Even so, fewer than 15 per cent of the leading apps offer any explainable-AI overlay. Technical robustness matters as well: when a micro-service architecture survives peak traffic with almost no errors, users interpret the stability as proof of a fairer system [28].

Fairness debates extend beyond diners to restaurants and couriers. Delivery-fee splits that once hid extra margins now face scrutiny, and research at Duke University shows that fixed-fee revenue sharing could raise overall welfare [68]. Because fairness spans privacy, nutritional equity, and partner income, future studies should measure these dimensions together instead of tackling them one by one. An integrated framework that joins data governance, economic justice, and well-being would give designers and regulators clearer guidance on how to keep AI-powered food delivery both profitable and fair.

4.4.5. Social Co-Experience

Mobile food ordering is often treated as a solo convenience yet eating and choosing food remain social acts. Modern apps now weave that sociability into the interface. Swiggy’s Group Cart, for example, lets several friends add dishes from different restaurants to one order; ethnographic work shows the feature eases “decision bottlenecks” and raises enjoyment when tastes diverge, while platform data record baskets about twenty per cent larger than solo equivalents [27]. A survey of Indonesian coffee shop users finds that the perceived quality of in-app group tools—shared-cart editing and chat—directly predicts continued use, with a path strength of β = 0.43 [69].

Social-proof cues add another layer to the experience. Labels like “Popular near you”, live order updates, and streak counters can speed up decisions by appealing to social influence. However, trust drops quickly if users doubt these features are real. Eye-tracking research shows people spend only about three seconds looking at source labels before deciding if they trust them, and badges that seem fake increase the chance they will leave the app [70]. A mixed-methods study in Portugal found that real-time endorsements from friends can restore trust and reduce the risk of users dropping out [71].

Short-form video is an emerging social channel. Uber Eats is testing a TikTok-style feed where diners swipe vertical clips and order directly; early data suggest that seeing dish texture and portion size boosts confidence in trying new cuisine [72]. Gopuff market reports confirm that viral TikTok dishes lift sales, indicating feedback loops between off-platform trends and in-app purchases [73]. Diary studies with young adults echo this pattern: forty-two per cent of late-night orders follow peer-shared food videos within a day of viewing [74].

Social benefits are not uniform across cultures or age groups. Comparative studies show collectivist consumers gain more satisfaction from group ordering and split payments, whereas individualists prefer self-customization tools such as nutrition filters [62,75]. Usability tests with older adults reveal discomfort when several people edit one cart at once, pointing to the need for gentler interfaces for less tech-savvy users [76].

Co-experience also extends beyond checkout. Posting unboxing videos, joint rating sessions, and tagging friends generate user content that recommender systems reuse to refine future suggestions [11,47]. Household-level features such as “Dining Circle” offer richer relational data but raise privacy questions. Opt-out controls are vital, because shared accounts can blur consent boundaries, a risk highlighted in conceptual analyses of food-sharing apps [6,77].

Important gaps remain. Social-proof features might reinforce popular menu items through algorithmic bias, or, in contrast, encourage users to try new dishes through peer influence. Although both effects are noted in recommender system studies [20,48], the ways that diners, couriers, and algorithms interact remain poorly understood, even though communication across these actors can shape user satisfaction [44]. Exploring these areas will help move the field beyond simple efficiency toward a deeper understanding of how food apps support shared and meaningful dining experiences.

5. Discussion

Customer experience (CX) has long driven loyalty in mobile commerce, but in AI-enabled mobile food-ordering apps (MFOAs), the decisive moments now occur inside algorithmic black boxes. Service quality and usability still matter [78,79], yet today every tailored menu, price nudge, and chatbot reply are generated in real time by context-aware recommender systems and demand-forecast engines [13,80]. These “always-on” interactions shape perceptions and behavior more than any static interface can [81,82].

Scholarly work, however, remains rooted in Technology-Acceptance Thinking (TAM, UTAUT), which asks only whether a tool is useful and easy [83,84]. To push beyond adoption logic, we re-read 55 empirical studies through four experiential themes—algorithmic-personalization value, affective engagement, data trust and procedural fairness, and social co-experience—revealing how AI can heighten value and delight while just as quickly sparking fatigue, privacy anxiety, and fairness outrage [85].

5.1. Key Findings

5.1.1. KF-1 Algorithmic-Personalization Value

Personalized menus are the main battleground for MFOA differentiation. By analyzing location, order history, and even scrolling pauses, recommender engines surface dishes that cut search effort and lift conversion [13]. Yet, longitudinal analyses uncover a paradox: frequent diners face “recommender flattening,” where novelty drops, basket diversity shrinks, and browsing fatigue rises [86,87].

The current evidence celebrates short-run gains—higher relevance, faster checkout—but rarely tests long-term outcomes such as nutritional balance or sustainability. Only isolated studies flag calorie drift and menu monotony. Future work should therefore (1) build diversity controls (e.g., variety sliders, random-injection protocols) that keep the surprise alive, and (2) expand success metrics beyond clicks to include wellbeing and societal impact. Without these safeguards, MFOAs risk trading immediate convenience for long-term disengagement and public-health costs.

5.1.2. KF-2 Affective Engagement

Warm first-person chatbots, playful “order-shipped” confetti, and value-matched coupons reliably spark joy, trust, and higher tips [55,88,89]. Surveys and field-experiment work show these micro-moments matter more to retention than page-load speed or even price discounts [34,90]. Yet, the emotional upside is fragile: a third promotional push-alert within 48 h reverses the effect, shifting sentiment from “tempting” to “intrusive” and cutting orders by 7–14% [26,30]. Similar volatility appears when surge pricing strikes: frustration spikes within minutes and drives “rage-quit” churn [91]. The evidence therefore positions emotional swing management—not technical performance—as the key predictor of long-term engagement; platforms need sentiment analytics that throttle promotions and apologize fast when algorithms misfire.

5.1.3. KF-3 Data Trust Procedural Fairness

Consumers gladly trade location and taste data for one-tap re-ordering, until they meet an unexplained fee or surge price. At that moment, trust and loyalty plunge [92,93]. Social-media analyses record 10× spikes in anger emojis within hours of opaque pricing [94]. Early mitigation studies are promising: a single cost-breakdown pop-up (distance + demand + weather) or a brief “why-this-dish” explainer reduces fairness complaints by roughly half without hurting sales [14]. Conversely, continued opacity fuels perceptions of exploitation and of calorie-dense “pushes” aimed at profit, not health [31]. Participatory governance ideas such as Rahwan’s society-in-the-loop approach—giving diners a voice in the algorithm design—are still conceptual but emerge as the logical next step toward durable data trust [95,96].

5.1.4. KF-4 Social Co-Experience

Group-ordering turns a solo convenience into a shared ritual: Swiggy-style carts and split-payment mechanics lift basket values by about 20% and boost enjoyment [27,69]. Yet, social influence cuts both ways. “Popular near you” badges accelerate choice, but eye-tracking shows credibility collapses in under three seconds if cues feel inflated [70,97]. Peer pressure can also override personal dietary goals, as joint decisions tilt toward indulgent items [98]. Cross-platform contagion amplifies the effect: TikTok food trends redirect late-night orders within 24 h [99]. The upshot is a design dilemma: celebrate the fun of communal ordering while building safeguards—visible source labels, equitable fee-splits, opt-outs—to protect individual preference and trust.

5.2. Theoretical Implications and Conceptual Mapping

5.2.1. Integration with the USUS Framework

Building on Scholtz’s (2006) USUS model, our conceptual map (Figure 3) specifies how five experiential pillars—instrumental usability, algorithmic-personalization value, affective engagement, data trust and procedural fairness, and social co-experience—converge to shape customer experience in AI-mediated food ordering. By attaching concrete sub-attributes to each pillar (e.g., “navigation clarity,” “surge-pricing perception”), the map gathers otherwise scattered findings on algorithmic CX [100] and highlights boundary conditions—such as usage frequency, digital literacy, or culture—those earlier applications left implicit [101].

5.2.2. Extending Theory Beyond Adoption Logic

Where most studies still rely on TAM or UTAUT and treat usefulness and ease as primary triggers [102,103], the present synthesis draws on logs, field experiments, and unsolicited reviews [44] to show that post-adoption realities demand additional constructs. Affect, trust, and social influence rival instrumental efficiency in driving loyalty, corroborating [81] call to view food apps as “always-on, algorithmic relationships” rather than one-off acceptance events.

5.2.3. Mechanisms Illuminated by the Map

Across the five pillars, long-term satisfaction hinges less on the novelty of recommendations than on their reliability and ethical alignment [102]. Early bursts of relevance and serendipity can devolve into “recommender flattening,” provoking browsing fatigue and caloric drift when engines over-optimize for familiarity [89]. Real-time menu reordering and price recalculations compress appraisal cycles to minutes, turning customer emotion into a chain of micro-transitions [44].

5.2.4. Coupling Privacy and Procedural Justice

The synthesis refines privacy-calculus and fairness-heuristic perspectives by demonstrating that data disclosure, price perceptions, and surge-pricing reactions form mutually reinforcing judgements [89,102]. Customers tolerate granular data capture until an opaque fee violates distributive justice; thereafter, trust collapses faster than convenience benefits can recover it. This tight coupling argues for integrating informational and procedural justice in future theorizing.

5.2.5. Social Co-Experience as a Moderating Force

Group carts and algorithmic social-proof cues magnify or buffer negative affect depending on group composition, cultural norms, and cue transparency [99]. By positioning social co-experience as a moderator, the map invites multi-actor decision models that move beyond the individualist assumptions of classic consumer-behavior frameworks.

5.2.6. Methodological Agenda

The current dependence on user-generated text overlooks silent churners and marginalized segments [101]. The map therefore signals the need for longitudinal panels, multi-modal emotion sensing, and mixed-method designs that combine behavioral telemetry with qualitative depth. Cataloguing constructs such as recommender flattening, algorithmic anxiety [100], nutritional bias, and cross-platform contagion [99] supply a vocabulary for scale development akin to the Service Robot Acceptance Model in hospitality [104]. Mediator and moderator pathways—digital literacy, cultural tightness–looseness [105], household ordering norms—emerge as prime targets for empirical testing.

5.3. Practical Implications

The speed at which artificial intelligence has been embedded in mobile food-ordering apps now exceeds the pace of established managerial guidance, forcing platforms and restaurant partners into costly trial-and-error cycles [106,107]. By distilling empirical findings, this review offers five actionable priorities.

5.3.1. Prioritize Reliability over Novelty in Personalization

Recommender engines should aim first for consistent accuracy. Diners welcome relevant suggestions, yet enthusiasm fades once menus become repetitive, nutritionally skewed, or overtly manipulative [108]. Calibrating algorithms to respect diversity constraints, incorporate health metrics, and provide concise “why-this-dish” explanations extends engagement and protects trust.

5.3.2. Calibrate Emotion with Real-Time Feedback

User journeys are emotionally volatile: a burst of confetti or a witty chatbot can lift the user’s mood, but overexposure quickly breeds fatigue. Sentiment analytics should therefore throttle joy triggers—confetti bursts, playful greetings, surprise coupons—before irritation sets in [89].

5.3.3. Make Data Trust and Fairness Visible

Opaque fees and surge prices remain the chief trust breakers. Granular cost breakdowns, clear surge rationales, and privacy dashboards convert latent anxiety into informed choice, while meaningful consent toggles reinforce users’ sense of control [106,107]. The result is a measurable “trust dividend” that outlasts short-term promotions.

5.3.4. Design Collaborative Tools for Equity, Not Just Convenience

Shared carts, split payments, and vote-based dish selection promote social enjoyment but can also provoke conflict if costs feel uneven. Interfaces should display role permissions and fee allocations transparently, especially in heterogeneous groups [108], to avoid post-purchase friction.

5.3.5. Harness, Yet Temper, Cross-Platform Influence

TikTok, Instagram, and Snapchat serve as powerful discovery funnels but magnify peer pressure and indulgent choices. Responsible link-outs that preview portion size or nutritional content can moderate impulse orders without undermining virality [108].

In sum, the evidence shows that successful mobile food-ordering apps now hinge on more than technical efficiency: long-term value arises when algorithmic personalization remains diverse and reliable, emotional cues are calibrated to avoid fatigue, pricing and data practices are transparent and fair, and social features are designed to empower—not pressure—users. Framing these insights within the USUS-based map clarifies how instrumental usability, personalization value, affective engagement, data trust, and social co-experience interact, while the five practical priorities translate theory into concrete design guidance. Together, these points underscore a central message: a sustained competitive advantage in AI-mediated food delivery will come from algorithms that are not only smart, but also ethical, empathetic, and publicly accountable.

6. Limitations and Directions for Future Research

Although this review integrates the most recent evidence on AI-enabled mobile food-ordering apps, its conclusions are inevitably constrained by gaps in the underlying literature. In what follows, we outline six principal limitations and propose concrete research strategies—anchored to newly available instruments such as the Mobile-App-Convenience Scale—to advance the field.

6.1. Narrow Data Sources and the “Silent-Churner” Problem

Most published studies rely on user-generated text such as App Store reviews, social-media posts, or brief surveys. This lens excludes silent churners and under-represented groups who do not voice their concerns online, thereby biasing our understanding of customer experience. Future work should triangulate textual data with (i) transactional logs that reveal actual spend and repeat behavior, (ii) longitudinal survey panels, and (iii) in-depth interviews targeting low-voice segments. Embedding the cross-culturally validated Mobile-App-Convenience (M-App-Conv) Scale [109] in such mixed designs would enable researchers to test whether perceived convenience mediates the path from algorithmic personalization to continued use within a Stimulus–Organism–Response framework.

6.2. Geographical Imbalance

Empirical evidence is heavily skewed toward North American and East Asian markets [110], limiting external validity. Intentional sampling in Latin America, Africa, South Asia, and the Middle East is therefore essential. Because M-App-Conv has already demonstrated cross-country reliability, its deployment in emerging economies can illuminate how local payment infrastructures and cultural norms modulate AI-driven notions of convenience.

6.3. Short Observational Horizons

Fewer than ten papers trace user behavior beyond three months, leaving the field blind to long-run effects such as recommender “honeymoon” decay, cumulative nutritional drift, and trust recovery. Longitudinal diaries, multi-year panel data, or retailer partnerships that grant access to historical order logs (e.g., [22]) can map these trajectories. Repeated M-App-Conv administrations will clarify whether perceived convenience decays (adaptation) or compounds (reinforcement) with continued exposure.

6.4. Scarcity of Field Experiments

Randomized controlled trials remain rare despite their capacity to establish causality [111]. Future agendas should prioritize platform-embedded experiments that manipulate transparency layers, surge-pricing rationales, or social-proof cues. In such trials, M-App-Conv can serve as an outcome variable, while fairness or trust items act as mediators—permitting rigorous tests of the hypothesized “convenience-versus-fairness” trade-off.

6.5. Measurement Voids for Emerging Constructs

Concepts such as algorithmic anxiety, menu-flattening fatigue, and perceived nutritional bias lack validated scales. Researchers should invest in systematic instrument development: adapt relevant M-App-Conv items (e.g., “time saved,” “effort saved”), establish clear construct definitions, and pursue cross-study replications to secure reliability and cultural invariance [112].

6.6. Under-Developed Interdisciplinary Links

The ethical and health ramifications of AI-curated menus remain largely unexplored. Collaborative research with nutrition scientists, public health scholars, and behavioral economists could reveal whether personalization nudges diners toward or away from balanced diets, and under which conditions. Comparative trials of human-curated versus AI-curated menus—or hybrid service flows [113]—executed in such interdisciplinary settings would be particularly valuable.

6.7. Roadmap

Table 5. Roadmap for Future Research on AI-Enabled Mobile Food-Ordering Apps.

Theme	Research Questions
Instrumental Usability	How can menus stay relevant and spatially stable? Do biometric/tokenized payments change the convenience–privacy trade-off? Which micro-cues (progress bars, ETAs, push pacing) best cut fulfilment anxiety? How do phone-to-watch or car-dashboard shifts reshape usability?
Algorithmic Personalization	Can health and sustainability goals be built into recommenders without killing novelty? What safeguards stop long-term menu homogenization for heavy users? Which “why-this-dish” formats deliver clarity without overload? How do local food norms and price sensitivities moderate personalization effects?
Affective Engagement	How do emotions evolve from pre-order to post-meal under adaptive algorithms? When do delight cues (confetti, chatbots) tip into annoyance? Can explainable AI ease algorithmic anxiety and reduce churn? How do cross-cultural norms shift acceptable humor or apology tone?
Data Trust Fairness	Do fine-grained cost or surge explanations lift perceived fairness? Can participatory data trusts build durable platform trust? How do hidden pricing biases widen dietary inequities, and what fixes work? How do national privacy norms affect tolerance for data capture?
Social Co-Experience	How do shared carts and split payments alter group satisfaction and conformity pressure? What design rules ensure fair fee allocation in joint orders? How does social proof feel authentic in collectivist vs. individualist cultures? What safeguards include older, young, or low-literacy diners in group ordering?

aligns each experiential pillar with specific unanswered questions and recommended methods. By pairing rigorous multi-method designs with standardized instruments, the next wave of research can move from descriptive insight to causally robust, generalizable knowledge on AI-mediated food delivery.

7. Conclusions

A review of fifty-five peer-reviewed studies shows that scholarship on AI-enabled mobile food-ordering apps has moved from technical proofs of concept to a customer experience agenda. Annual output rose from ten papers in 2022 to twenty-two in 2024, with twelve more in the first quarter of 2025 [102]. Yet, methods lag: thirty-eight per cent of studies analyze user-generated text, thirty-three per cent rely on one-shot surveys, and fewer than five per cent run live field experiments, leaving long-term, real-world dynamics largely unexplored.

Five intertwined levels now explain competitive advantage. First, instrumental usability: menus trimmed to five–seven dishes cut cognitive load and raise satisfaction [21,114]. Second, algorithmic-personalization value: early relevance fades when recommender novelty drops, signaling the need for diversity safeguards. Third, affective engagement: a third push-alert within a week flips delight to annoyance and decreases orders [52]. Fourth, data trust and procedural fairness: full cost breakdowns boost justice perceptions and cut complaints by more than a quarter without harming conversion [91]. Fifth, social co-experience: group-order and split-payment tools lift basket values and engagement in case studies [27], yet fewer than one-fifth of leading apps provide them.

These findings yield clear actions. Developers should anchor menus with stable tiles, keep adaptive lists within the five-to-seven range, cap promotional alerts below irritation thresholds, and add concise “why this price?” explainers. Policymakers can require algorithm-impact audits and transparency screens to curb nutritional bias and protect distributive justice, echoing calls for ethical data governance [6]. Restaurant partners should bargain for visibility that rewards nutritionally diverse menus and experiment with group bundles, while privacy authorities warn that opaque data flows continue to erode trust [65].

Research gaps remain. Only five per cent of studies use field experiments, so scholars should follow the same users for a year or more to test whether early convenience gains outweigh nutritional drift and trust erosion. Fairness work needs validated metrics for popularity and exposure bias [48]. Comparative trials of privacy-preserving techniques—differential privacy versus federated learning—are overdue, and cross-cultural emotion mapping should probe how collectivist and individualist audiences interpret social-proof cues differently [62]. Throughout, technical terms must stay clear: cognitive simplicity means minimizing clicks and mental effort, while ethically responsible data use means collecting only what is necessary, guarding it rigorously, and applying it for disclosed, user-benefitting ends.

Limitations temper these conclusions. All fifty-five studies focus on North American or East/Southeast Asian samples, and reliance on self-reported satisfaction leaves silent churners invisible. Broader sampling and passive telemetry will sharpen both theory and practice. Ultimately, the sustainable advantage in algorithmic food delivery rests on blending technology, emotion, fairness, and social connection into a transparent, human-centered experience. Platforms that adopt these principles—transparent pricing, culturally adaptive interfaces, and fairness-aware recommender engines—will secure today’s orders and earn the trust that the next decade of AI-mediated dining demands.