According to the World Health Organization (1), approximately 39 and 13% of the adult population worldwide were living with overweight and obesity, respectively. Among the four million deaths related to having a high body mass index (BMI), more than 66% of them were attributed to cardiovascular diseases among other common chronic diseases, including diabetes mellitus, chronic renal diseases, cancers, and musculoskeletal disorders (2). While pharmacological and surgical interventions have been demonstrated to lead to quick and successful weight reduction, behavior change remains the safest first-line weight management option (3). Several factors have been identified to improve sustained weight loss including adhering to high levels of physical activity, calorie restriction, weight and dietary self-monitoring, self-restraint, confidence, and the low personality trait of novelty seeking (4). However, weight management programs focused on calorie restriction and regular physical exercise are mentally and physically demanding, especially for those with a higher BMI, leading to high non-adherence and attrition rates (5). A 10% increase in adherence would reduce one’s BMI by 2.59 points, suggesting the importance of increasing the acceptability, uptake, and engagement in such programs (6). Various behavior change strategies such as counseling (e.g., motivational interviewing) (7), psychotherapy (e.g., cognitive behavioral therapy) (8), special diet patterns (e.g., time-restricted eating) (9), and exercise regimes (e.g., high-intensity training) (10) have been used to increase one’s motivation and commitment to behavioral change. Various modes of delivery such as face-to-face, web-based, mobile phone apps, and group sessions have also been trialed to improve the effectiveness of various weight loss programs (11, 12). However, the average adherence rate of such programs was reportedly only 60.5%, including those who may have attended every program session but have not necessarily performed the lifestyle change recommended (13).

Alongside the increase in smartphone penetration worldwide, weight loss apps provide users with the convenience of accessing weight loss programs anywhere, at any time. Examples include MyFitnessPal (14), which is essentially a calorie counter to monitor energy intake and expenditure; WW (Weight Watchers) (15), which is an incentive-based points system that encourages users to adhere to their daily meal plans; and Noom (16), which provides health coaching on top of calorie counting. While users of such apps report significant weight loss (17), the first hurdle toward success is the adoption of such programs by the mass public, especially those with overweight and obesity. Common reasons for the non-adoption of such programs include depleting motivation, the lack of satisfactory results, dietary constraints, logistical constraints, low level of supervision, and low provision of social support (13, 18). However, these functions are resource-intensive (e.g., manpower, technical support, and infrastructure) to deploy and maintain.

Artificial intelligence (AI) has been increasingly popular in enhancing the resource efficiency of various activities such as entertainment (e.g., personalized video recommendations on Netflix) and contexts such as smart cities (e.g., energy consumption prediction). AI is capable of offloading the demand for manpower through human–AI interventions, which imitates human intelligence and communications to perform human tasks faster, more accurately, and more efficiently (19, 20). However, the effectiveness of AI-assisted weight loss apps remains unclear and is largely contingent upon the user’s engagement with the app’s contents. Several studies have examined the acceptability of weight loss apps, but few studies have considered those with AI enhancements, the influencing factors of acceptability in a Southeast Asian multi-ethnic population (21–23). For example, one study on college students reported that out of the seven influencing factors of usage intention included in Unified Theory of Acceptance and Use of Technology 2 (UTAUT2), only performance expectancy, hedonic motivations, price value, and habit significantly predicted app acceptability (24) On the contrary, a study on people with overweight and obesity showed that only performance expectancy, effort expectancy, and social influence influenced app acceptability (25).

While studies have shown positive perceptions of the use of AI in healthcare such as increased availability, user-friendliness, and cost-efficiency, common barriers include concerns about data privacy, credibility, patient safety, and technological maturity (26). With the increasing use of AI in weight loss apps, it is timely to assess the needs, preferences, and influencing factors of AI and non-AI smartphone-based weight management apps in a multi-ethnic context like Singapore. Therefore, we aimed to examine the user perceptions of artificial intelligence (AI)-based mobile apps for weight management in people with overweight and obesity. Broadly, this study would also indicate the technological readiness of Singaporeans who are overweight and obese to adopt AI-based technology.

A cross-sectional study is reported as part of a sequential explanatory study. This study was approved by the National Healthcare Group (NHG) Domain Specific Review Board (DSRB) Ethics Review Board (ref: 2021/00834) and registered with ClinicalTrials.gov (ref. NCT05257239). The results are reported according to the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) checklist for cross-sectional studies (Supplementary Table S1).

To the best of our knowledge, this is the first study to examine the public acceptance and intention to use AI-assisted weight management apps in Singapore, a high-income multi-ethnic country that was ranked fourth in the International Institute for Management Development (IMD) World Digital Competitiveness Ranking 2022, following Denmark, the United States, and Sweden (first) in increasing rank order (39). Recognizing the important role of AI in the future, Singapore also prioritizes AI research and launched a national AI strategy in 2019 (40).

Generally, the seven UTAUT2 constructs explained 63.3% of the variance in our sample’s intention to use AI-assisted weight management apps, similar to another study on 439 German users with overweight and obesity where the UTAUT model explained 60.0% of the variance (25). However, this study was not specific to AI-assisted apps, where the incorporation of AI into healthcare devices and services has always been contended for its potential to breach data privacy and trust issues (26). However, replacing the price value construct with trust and privacy concerns only explained 20% of the variance in the intention to use non-AI-specific lifestyle and therapy apps (41). Although this could be due to a larger predictive strength of price value than trust and privacy concerns on intention, this is more likely due to testing the UTAUT model on a general population as compared to a targeted population in the former two studies on adults with overweight and obesity. In terms of adapted UTAUT models, one study on a Taiwanese sample reported that the UTAUT2 constructs alongside personal innovativeness and network externality explained 75.5% of the AI-assisted weight loss app acceptability among the general public (42). These findings suggest that the application of UTAUT2 in developing and evaluating technology-related healthcare services may not be generalizable across populations and the effectiveness of such interventions should be customized to the needs of certain populations. The integration of the qualitative findings from the larger study is reported elsewhere.

Interestingly, only performance expectancy, hedonic motivation, and the habit of using AI-assisted apps were significant predictors of intention. Our literature search revealed vast variabilities in the determinants of weight loss app acceptability. Concurring with our findings, the most common determinants were habit and performance expectancy, which were identified by six studies in both the general and overweight/obese populations (17, 24, 42–45). However, three studies (17, 43, 45) also found effort expectancy and social influence to be significant predictors of behavioral intention while another study on the general population identified performance expectancy, effort expectancy, and social influence as significant determinants (25). Two studies (17, 24) found hedonic motivation and price value to be significant predictors of behavioral intention in the general population, whereas facilitating condition was identified as one of the key predictors of behavioral intention in a sample attending cardiac rehabilitation (43). Such variability could be due to the different determinants that influence AI-assisted and non-AI-assisted weight loss apps differently. Our findings suggest that users were willing to use AI-assisted apps if they perceived a high chance of weight loss success and used them repeatedly, regardless of the effort, price, resources, and social influence present. As both Singapore and Taiwan are developed countries with relative financial stability (42), users may be less sensitive to the price, effort, and resources available. Due to the relative novelty of AI-assisted apps, the absence of social influence as a significant predictor could also be due to the lack of experience and knowledge about it as compared to general weight loss apps as indicated by the low scores on the habit of use (3.43 ± 1.56).

Interestingly, while 86.3% of the participants expressed the intention to cut down on tempting food consumption, only 13.7% responded to eating healthily. This presents an irony where users may intend to reduce the intake of only certain junk foods and not compensate for this with healthy food (e.g., fruits and vegetables) intake (46). The top three most tempting foods were sweets (56.7%), chocolate (55.2%), and ice cream (50.7%), which are calorie-dense obesogenic foods that highlight the need for interventions to reduce the consumption of such foods. Interestingly, self-regulation, depression risk, BMI, and waist circumference were not associated with intention to use AI-assisted weight management apps, potentially because these constructs are more distal than the UTAUT2 constructs in terms of their influence over the intention. Moreover, although 24.4% and 19.3% of the participants were at risk of having anxiety and depression symptoms, these risks were not significantly associated with BMI or waist circumference, which contradicts existing studies (47). This could be because the association only becomes significant when the anxiety and depression risk is confirmed with a diagnosis or because these psychiatric disorders are such a taboo in an Asian country that the prevalence was underreported in this study (48).

There were several limitations to this study. First, although the ethnic sample composition is representative of the local population, the sample is still relatively small as compared to census data. This alongside the use of self-reports would limit the generalizability of our findings. Second, although we explained the differences between AI and non-AI weight loss apps, participants may not fully appreciate the differences and hence lower the accuracy of our findings. Future studies could consider conducting a short test to ensure that participants fully understand the differences between AI and non-AI applications. Longitudinal studies could also be conducted to observe changes in user perceptions over time, and experimental studies could be conducted to evaluate the effectiveness of specific app features. This would inform the development of weight loss apps to be comprehensive yet limited redundancy. As only performance expectancy, hedonic motivation, and the habit of using AI-assisted apps were significant predictors of intention, weight management apps could be designed to track and visualize one’s weight loss progress, and intrinsic motivation, and promote repetition of healthy eating habits. Large language models (LLMs) could also be used to explore app feedback and social media comments to further understand the various predictors of user adoption. LLMs could also be used to provide personalized plans and feedback to enhance the effectiveness and user experience of AI-assisted weight management apps (19).

The UTAUT2 was found to produce a good fit in explaining the acceptability of AI-assisted apps among a multi-ethnic, developed, southeast Asian sample with overweight and obesity. Therefore, the UTAUT2 model is recommended to guide the development of AI-assisted weight management apps among people with overweight and obesity.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving humans were approved by National Healthcare Group Domain Specific Review Board. The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

HC: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing. PA: Conceptualization, Writing – original draft, Writing – review & editing. MD: Writing – review & editing, Validation. NC: Validation, Writing – review & editing. YC: Validation, Writing – review & editing. YG: Validation, Writing – review & editing. BJ: Validation, Writing – review & editing. AS: Validation, Writing – review & editing. LP: Validation, Writing – review & editing. KN: Validation, Writing – review & editing.