This study used a population-based, cross-sectional design in Hong Kong with a 2-stage, stratified random sampling design. In stage 1, we conducted random sampling at the living quarters level to select residential addresses built for habitation of one or more households, stratified by the 18 districts (geographic regions) and housing type. In stage 2, we randomly selected one eligible member per sampled household at the individual level using the “last birthday” method. Inclusion criteria were aged 15 years or older, residence in Hong Kong, and the ability to speak Chinese. Exclusion criteria included inmates, those living on vessels, and foreign domestic helpers. Invitation letters, which stated the study purpose, were sent to invite the household to participate in the survey. In case of nonresponses, on-site visits were conducted to recruit eligible members for face-to-face interviews. Voluntary informed consent was obtained. Data were collected online, by phone, or face-to-face in about 10 minutes, and the online system and interviewer ensured complete responses. A pilot study was conducted among 10 community adults in January 2024 and feedback from the respondents was used to refine the questionnaire and research process. Quality control included a random check of 208 interviews, with 90.8% passing the back-check criteria (at least 3 out of 4 questions in the quality back-check questionnaire were asked during the interview).

The study was reviewed and approved by the Human Research Ethics Committee of the University of Hong Kong (reference number EA200335). All participants were informed of the study objectives and provided informed consent before completing the survey. To minimize item-level missing data, the survey was programmed such that all questions required a response for the completion of the survey. Participation was voluntary, and participants could withdraw from the survey at any time. The final analytic sample contained no missing data since no respondents withdrew from the survey because of refusal to answer a question. Data were anonymized and all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Declaration of Helsinki in 1975, as revised in 2008. The study findings were presented in accordance with the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) reporting guidelines.

Household data collection occurred between February and April 2024. We invited 3275 households, and 1800 eligible participants completed the survey. Reasons for nonresponse included noncontact (n=1399), refusal (n=63), and unoccupied households (n=13). According to the American Association for Public Opinion Research Response Rate 1 standard, the response rate was calculated by the formula: response rate = completed cases/(completed cases + partial cases + refusals + noncontacts + unknown eligibility cases) = 55.2%. The present sample was distributed across 18 districts (cluster size = 57-148) and 4 housing types: private ownership (43.9%), public rental (27.2%), subsidized ownership (19.3%), and private rental (9.6%). For the data collection mode, the number of responses collected through self-report questionnaire, phone interviews, and site visits were 553/1800 (30.7%), 765/1800 (42.5%), and 482/1800 (26.8%), respectively. The mean age was 49.3 (SD 14.9; range 17-95) years; 55.3% (995/1800) were women, 47.5% (855/1800) lived in urban districts, 68.6% (1235/1800) were married, 64.8% (1166/1800) were employed, and 85.9% (1547/1800) were born in Hong Kong. In the 2021 population census, Hong Kong residents had a median age of 46.3 (IQR 28-62) years and the labor force participation rate was 58%. The present sample showed similar sociodemographic profiles to the general Hong Kong population. Participants were categorized into 3 age groups: young adults (aged 17-39 years; n=521), middle-aged adults (aged 40-64 years; n=844), and older adults (aged 65-95 years; n=435).

GenAI use was measured by the amount of time that the respondents spent on average per day using GenAI in the past year. In this study, GenAI was defined as deep learning models such as ChatGPT, DALL-E, and Claude that can generate text, images, or other media in response to prompts and excluded traditional virtual assistants such as Siri and Alexa. The time of GenAI use was answered on a 4-point scale: 0=“not used,” 1=“<1 hour/day,” 2=“1-2 hours/day,” and 3=“>2 hours/day.” A recent meta-analysis found moderate correlations between self-reported media use time and logged data and no statistically significant differences between single-item estimates and scale-based measures, or between measures of usage duration and usage volume [29]. This provides support for using self-reported daily time use. While the measure of AI usage was specific to GenAI, the following measures refer to AI in general.

AI literacy was evaluated using a 3-item short version of the 12-item Artificial Intelligence Literacy Scale (AILS) [30]. The AILS was developed in the Chinese context to measure users’ competence to understand, apply, and evaluate AI technology. This study selected 3 items from the AILS across 3 subscales: “I can distinguish between smart devices and non-smart devices” (Awareness), “I can skillfully use AI applications or products to help me with my daily work” (Usage), and “I can choose a proper solution from various solutions provided by a smart agent” (Evaluation). The use of a short version was to reduce participant burden and ensure brevity of the phone survey. The 3 items were answered on a 7-point Likert scale (1=“strongly disagree,” 2=“disagree,” 3=“slightly disagree,” 4=“neither agree nor disagree,” 5=“slightly agree,” 6=“agree,” and 7=“strongly agree”), with higher scores denoting greater AI literacy. Excellent reliability was found for the 3-item brief version of AILS (Cronbach α=0.93) in the present sample.

Perceived AI usefulness was evaluated by the following 2 items: “I believe that AI will improve my life” and “I believe that AI will improve my work”. Behavioral intention to use AI was assessed by the item: “I think I will use AI technology in the future.” These 3 items are selected from the validated 4-item Artificial Intelligence Attitude Scale (AIAS-4) [31] according to their theoretical alignment with the TAM. The AIAS-4 is a brief common measure of AI attitudes and has recently been validated in the Chinese context [32]. The items were rated on a 10-point Likert scale (1=“strongly disagree,” 2-3=“disagree,” 4=“slightly disagree,” 5-6=“neither agree nor disagree,” 7=“slightly agree,” 8-9=“agree,” to 10=“strongly agree”). Excellent composite reliability was found for perceived AI usefulness (Cronbach α=0.91) in the present sample.

Life satisfaction was measured by the 5-item Chinese version of the Satisfaction with Life Scale (SWLS) [33]. Example items include “I am satisfied with my life” and “So far I have gotten the important things I want in life.” The 5 items are rated on a 7-point Likert scale (1=“strongly disagree,” 2=“disagree,” 3=“slightly disagree,” 4=“neither agree nor disagree,” 5=“slightly agree,” 6=“agree,” and 7=“strongly agree”). The total life satisfaction score had a theoretical range from 5 to 35, with higher scores denoting greater life satisfaction. The SWLS showed excellent reliability (ω=0.93) in the present sample.

Sociodemographic characteristics included age, gender (men and women), education level (9-point scale; 1=“no formal education,” 2=“primary school,” 3=“lower secondary school,” 4=“upper secondary school,” 5=“diploma,” 6=“associate degree,” 7=“Bachelor degree,” 8=“Master’s degree,” and 9=“Doctoral degree”), place of birth (Hong Kong or other place), presence of chronic health conditions (such as restriction in body movement, mental disorder, hypertension, and diabetes mellitus), household size, work status (have full-time or part-time jobs), and monthly household income (12-point scale; 1=“less than HKD10,000,” 2=“HKD10,000-19,999,” 3=“HKD20,000-29,999,” 4=“HKD30,000-39,999,” 5=“HKD40,000-49,999,” 6=“HKD50,000-59,999,” 7=“HKD60,000-69,999,” 8=“HKD70,000-79,999,” 9=“HKD80,000-89,999,” 10=“HKD90,000-99,999,” 11=“HKD100,000-119,999,” and 12=“HKD120,000 or more”).

First, we examined the factorial validity of the 3 latent factors (perceived AI usefulness, AI literacy, and SWLS) using confirmatory factor analysis (CFA). Model fit was evaluated following standard criteria on fit indices [34]: comparative fit index (CFI) and Tucker-Lewis index (TLI) ≥ 0.95, and root-mean-square error of approximation (RMSEA) and standardized root-mean-square residuals (SRMRs) ≤ 0.06. Scalar measurement invariance of the factors was compared across gender and age groups between the configural invariance model (with free factor loadings and item intercepts) and the scalar invariance model (with equal factor loadings and item intercepts). Model comparison was performed using chi-square difference tests, Bayesian information criterion (BIC), and changes in fit indices with ΔCFI ≥ –0.01, ΔRMSEA ≤ 0.01, and ΔBIC ≥ 10, favoring the more parsimonious model [35].

Second, we conducted chi-square tests and multivariate analysis of variance (MANOVA) to compare sociodemographic profiles and GenAI use across gender and age groups. We performed MANCOVA to compare perceived AI usefulness, behavioral intention, AI literacy, and life satisfaction across GenAI use. MANCOVA controlled for the following sociodemographic covariates: gender, age, education level, place of birth, chronic condition, household size, work status, and household income. Effect sizes were denoted by partial eta-squares (η²) with 0.01, 0.06, and 0.14 denoting small, moderate, and large sizes, respectively [36]. Post hoc tests were conducted with Sidak adjustment. Statistical significance was taken at the .05 level.

Third, we investigated the pathways specified in the conceptual model via structural equation modeling (SEM). Figure 1 illustrates the conceptual model of this study. Black arrows denote the paths tested in the main model. Dashed blue arrows refer to the moderating effects in multigroup analysis. Age, education level, place of birth, household size, and household income are the control variables in the SEM. Gender, chronic conditions, and work status act as the control variables in the SEM or the multigroup variables in the multigroup SEM.

In the main SEM, perceived AI usefulness, AI literacy, and life satisfaction were modeled as latent factors measured by the respective indicators. Behavioral intention to use AI and GenAI use were treated as an observed continuous and an observed ordinal variable, respectively. The model included the sociodemographic variables as covariates. We evaluated the indirect effect of perceived AI usefulness on life satisfaction through sequential mediators (behavioral intention, GenAI use, and AI literacy) using 10,000 bootstrap samples to account for the potential skewed distribution of the indirect effects. Standardized path coefficients and R² denoted the effect sizes of the associations. Fourth, we conducted multigroup SEM across sociodemographic subgroups, namely, gender, age groups, chronic condition, and work status. In the multigroup SEM, we empirically tested the differences in the path coefficients using Model Constraint and compared the 95% CI of the indirect effects across subgroups.

Sampling weights were applied in the main analyses, namely, CFA, MANCOVA, and SEM, to align the sample distribution with the 2021 Hong Kong census distribution across 16 age × gender strata. These weights were not design-based sampling weights based on individual selection probabilities. CFA was conducted in Mplus 8.6 under Type = Complex using the robust maximum likelihood estimator [37]; MANCOVA was performed using the Complex Samples module in SPSS 28.0 (IBM Corp); and SEM was estimated under Type = Complex using the robust weighted least squares estimator to account for GenAI use as a categorical endogenous variable. To account for the complex survey design and nonindependence of observations, the residential district was entered as the cluster variable. Robust SEs were computed in CFA and SEM via the sandwich estimator, which was equivalent to Taylor series linearization.

The 3-factor CFA model showed a good fit to the data (CFI=0.985, TLI=0.978, RMSEA=0.044, and SRMR=0.027), with strong item loadings on perceived AI usefulness, AI literacy, and life satisfaction (λ=0.77-0.95, P<.001). Across gender, the scalar invariance model showed a comparable fit (CFI=0.985, TLI=0.982, RMSEA=0.041, SRMR=0.033, and BIC=51926) as the configural invariance model (CFI=0.986, TLI=0.980, RMSEA=0.044, SRMR=0.029, and BIC=52012) and the chi-square difference test was not significant (Δχ²₁₄=14.5, P=.41). Across age groups, the scalar invariance model showed a comparable fit (CFI=0.977, TLI=0.975, RMSEA=0.047, SRMR=0.043, and BIC=51364) as the configural invariance model (CFI=0.981, TLI=0.974, RMSEA=0.049, SRMR=0.030, and BIC=51501). Despite a significant chi-square difference test (Δχ²₂₈=54.4, P=.002), the substantially lower BIC for scalar invariance models supported equal loadings and intercepts across groups.

Tables 1 and 2 show the sample profile and comparison across gender. Overall, 61.5% (1107/1800) of the respondents did not use GenAI in the past year and 7.2% (130/1800) of them used GenAI for more than 2 hours per day. Men were significantly more likely to be working (P<.001) and had higher education level than women (η²=0.007, P<.001), and data collection was significantly more likely to be conducted via phone interviews (P=.002) in women than in men. No significant gender differences were found in place of birth (P=.91), chronic condition (P=.07), GenAI use (P=.11), and age (P=.56). Men showed significantly higher AI literacy (η²=0.008, P<.001) than women.

Tables 3 and 4 show the comparison among the 3 age groups. Older adults were significantly less likely to be working and to use GenAI in the past year (P<.001). There were moderate to strong age differences in education level, household income, perceived AI usefulness, behavioral intention to use AI, and AI literacy (η²=0.11-0.35, P<.001), with all variables declining with age. Life satisfaction did not differ across age groups. Supplementary analysis found no significant differences among the 3 data collection modes in other sociodemographic characteristics (η²=0.000-0.002, P=.17-.76), perceived AI usefulness (η²=0.001, P=.55), behavioral intention to use AI (η²=0.001, P=.62), GenAI use (χ²₆=3.37; P=.76), AI literacy (η²=0.000, P=.79), and life satisfaction (η²=0.001, P=.46).

MANCOVA across GenAI use revealed moderate between-subject difference for perceived AI usefulness and behavioral intention to use AI (η²=0.084-0.105, P<.001), a large difference for AI literacy (η²=0.209, P<.001), and a small difference for life satisfaction (η²=0.007, P=.005). Figure 2 displays the estimated marginal means of AI-related constructs and life satisfaction across the 4 levels of GenAI use. In the post hoc tests, GenAI nonusers showed significantly lower levels of perceived AI usefulness (mean difference –1.11 to –1.40, P<.001), behavioral intention to use AI (mean difference –1.45 to –1.50, P<.001), and AI literacy (mean difference –1.09 to –1.68, P<.001) than GenAI users, regardless of duration of use. Users who engaged with GenAI for more than 2 hours per day reported higher AI literacy (mean difference 0.45-0.59, P=.001-.003) than other GenAI users, and they also showed higher life satisfaction (mean difference 1.60, P=.02) than nonusers.

The original cascading model showed a mediocre fit (CFI=0.855, TLI=0.782, RMSEA=0.060, SRMR=0.048). There was a large modification index for a direct effect from behavioral intention to use AI to AI literacy. The modified SEM with the added effect showed a good fit (CFI=0.974, TLI=0.960, RMSEA=0.026, SRMR=0.036) to the data. As Table 5 shows, women showed lower AI literacy (β =–0.06, P=.005) than men. AI-related constructs showed negative associations with age (β=–0.22 to –0.40, P<.001) and chronic condition (β=–0.12 to –0.13, P<.001) and positive associations with education level (β=0.14-0.24, P<.001). Household income was positively associated with AI-related constructs and life satisfaction (β=0.07-0.23, P<.001). The eight covariates explained 5.4% to 44.2% of the total variance of the study variables.

Figure 3 displays the SEM results for the main model, where the 3 latent factors showed substantial factor loadings (λ=0.74-0.93, P<.001). Controlling for the covariates, perceived AI usefulness had a very strong association (β=1.00, P<.001) with behavioral intention to use AI; behavioral intention to use AI was moderately associated (β=0.41, P<.001) with GenAI use; behavioral intention to use AI and GenAI use were moderately associated (β=0.32-0.47, P<.001) with AI literacy; and AI literacy was moderately associated (β=0.31, P<.001) with life satisfaction. The model explained an additional 64.8%, 10.5%, 29.5%, and 5.3% of the variance in behavioral intention to use AI, GenAI use, AI literacy, and life satisfaction, respectively.

Multigroup SEM across gender, age groups, work status, and chronic condition provided adequate fit to the data (CFI=0.963-0.974, TLI=0.949-0.964, and RMSEA=0.025-0.030). All path coefficients showed no significant differences across work status and chronic condition. Across gender, the path of behavioral intention to use AI → AI literacy was significantly stronger (P=.02) in men (β=0.51) than in women (β=0.40); and the path of AI literacy → life satisfaction was significantly stronger (P=.02) in men (β=0.40) than in women (β=0.21). Across age groups, the path of GenAI use → AI literacy in middle-aged adults (β=0.47) was significantly stronger (P<.001) than in young adults (β=0.24) and significantly stronger (P=.01) than in older adults (β=0.33). The behavioral intention to use AI → AI literacy path was significantly stronger (P=.001) in young adults (β=0.60) than in middle-aged adults (β=0.41).

Table 6 lists the indirect effects of perceived AI usefulness on life satisfaction via behavioral intention to use AI, GenAI use, and AI literacy across sociodemographic subgroups. In the overall sample, perceived AI usefulness showed a significant positive total indirect effect (αβ=0.186, 95% CI 0.134-0.242) on life satisfaction. The indirect effects were consistent across work status and chronic conditions, with highly overlapping 95% CIs. Across gender, the indirect effects were substantially stronger in men (αβ=0.260, 95% CI 0.177-0.348) than in women (αβ=0.112, 95% CI 0.055-0.180). Across age groups, the indirect effects were substantially stronger in older adults (αβ=0.227, 95% CI 0.113-0.367) than in young adults (αβ=0.124, 95% CI 0.038-0.214).

This study examined the age differences in perceived AI usefulness, behavioral intention to use AI, GenAI use, and AI literacy and their relations to life satisfaction in a population-based sample of Hong Kong adults in 2024. This study synthesized the simplified TAM and self-determination theory into the GenAI context and elucidated the interrelationships between the study variables. About Hypothesis 1, although women reported lower levels across all AI-related constructs, the difference was significant only in AI literacy but not in perceived AI usefulness, behavioral intention to use AI, and GenAI use. Consistent with Hypothesis 2, there was a substantial decline in GenAI use with age, which showed negative associations with perceived AI usefulness, behavioral intention to use AI, and AI literacy. The MANCOVA results found significantly higher levels of perceived AI usefulness and behavioral intention to use AI for the GenAI users over the nonusers, supporting Hypothesis 3. AI literacy and life satisfaction showed linear increasing trends across GenAI use levels in MANCOVA. These results provide empirical evidence for Hypothesis 4. Finally, an indirect association was found between GenAI use and life satisfaction via AI literacy, lending support to Hypothesis 5.

In the present sample, 38.5% (693/1800) of the respondents used GenAI in 2024, which was comparable to a 39% usage rate among US respondents aged 18-64 years at the same time [38]. In contrast to the finding of a meta-analytic review [39], men in the present sample reported higher AI literacy than women. The gender discrepancy could be attributed to reasons such as higher education level and working proportion. Besides, men tend to overestimate their AI knowledge and skills while women tend to underestimate themselves [40]. This study found substantial age differences in GenAI use, with the usage rate dropping from more than half in young adults to less than one-sixth in older adults. The negative association of age with GenAI use and AI-related constructs remained significant after controlling for known determinants (education level and socioeconomic status) [41]. The robust associations likely reflect generational differences in technology receptiveness and familiarity and could be attributed to higher openness personality trait in younger generations [42] and greater technology anxiety in older adults [17].

In line with similar studies [5,6,43], the SEM found a strong association between subjective perceptions of AI and behavioral intention to use AI, and the latter served as a pathway through which perceived AI usefulness was indirectly associated with GenAI use. Through GenAI use, users could gradually familiarize themselves with GenAI tools, acquire skills in human-AI communication and prompting, and develop critical thinking abilities to recognize stereotypes potentially induced by GenAI [44]. Self-determination theory suggests that learning is more effective when driven by autonomous motivation rather than external pressure, leading to greater engagement and better outcomes in GenAI use [10]. People with higher AI literacy would compare outputs across multiple GenAI platforms, adopt a critical disposition towards the outputs’ reliability, and verify information through lateral reading [45].

This study provides a nuanced understanding of GenAI use and life satisfaction. Our study found a positive but small relationship between GenAI use and life satisfaction, which is consistent with previous findings [15,16]. The direct effect between GenAI use and life satisfaction became nonsignificant after accounting for AI literacy in the SEM. In the SEM, a path was added from behavioral intention to use AI to AI literacy because of high modification index. This added path aligns with self-determination theory in which autonomous and self-directed learning results in higher AI competence and better digital well-being [46]. Besides, the bridging role of AI literacy as an indirect statistical pathway was robust across sociodemographic subgroups. These results underscore that GenAI use is associated with life satisfaction predominantly via its relationship with AI literacy. Existing studies have linked general self-efficacy with higher life satisfaction [47,48]. AI literacy has shown a mediating role in critical thinking and future work readiness in education settings [49,50] via factors such as greater adaptability, problem-solving skills, and digital confidence [51,52].

Subgroup analysis revealed preliminary findings on potential heterogeneity in the indirect effects across gender and age groups. The gender discrepancy in the indirect effects of AI literacy is in line with the results of a meta-analysis [26]. Men are typically overrepresented in Science, Technology, Engineering, and Mathematics–related fields and have greater exposure to on-the-job training in AI literacy as reflected in their higher levels of AI literacy than women. Women are more likely to face systemic barriers, such as algorithmic bias that tends to marginalize them, resulting in a digital divide and a smaller mediating role of AI literacy for them [25]. Furthermore, AI literacy mainly taps into the psychological needs of autonomy and competence while women value relatedness more in the evaluation of life satisfaction [53].

Despite lower levels of GenAI use and AI literacy among older adults, our study found a stronger bridging role for AI literacy among this age subgroup than among young adults. This discrepancy may appear counterintuitive since young adults are considered digital natives. The higher baseline digital literacy in young adults could denote a ceiling effect where AI literacy plays a lesser role, and AI literacy could provide greater marginal benefits to older adults given their lower digital fluency at baseline. Older adults are more risk-averse about data privacy and security issues than young adults, and AI literacy could play a greater role in helping them to trust and understand the rules on GenAI use. However, the very small number of older adults who used GenAI for more than 2 hours per day could make the subgroup results less stable and more vulnerable to sampling error. Given the exploratory nature of the subgroup analyses, future research should clarify the age differences in the bridging role of AI literacy in the relationship between GenAI use and life satisfaction in other samples.

Findings have some implications for digital-literacy policy and support targeted programs to mitigate digital divides among sociodemographic subgroups. Policymakers should strive for digital inclusion and active literacy cultivation in the digital age and ensure equitable access to GenAI among those with lower socioeconomic status [54]. Hands-on assistance from family and peers should be encouraged to build AI literacy, which was found to mitigate the relationship between perceived anthropomorphism and fear of AI [55]. Further enhancements of the usability of GenAI, in terms of accuracy, transparency, and reliability, could build up related literacy and promote its implementation into health care information seeking and self-care management among the general public [56]. As women tend to underestimate their digital literacy while their objective digital capabilities could be as good as men’s [24,25], their self-efficacy and sense of control in using AI could be promoted. Education on using GenAI for older adults could focus more on digital content creation, safety, and problem-solving ability, and GenAI tools can be developed with less emotional distance and age-friendliness to build their trust [57].

This study had a few limitations. First, the current cross-sectional design did not allow causal inference about the effects of GenAI use on life satisfaction. Respondents who felt more satisfied with life could be more likely to use GenAI, and higher AI literacy could predict more frequent GenAI use. Longitudinal studies are needed to confirm the mediating role of AI literacy and clarify the reciprocal effects of AI literacy and life satisfaction on GenAI use. Second, this study used short forms of validated scales (AIAS-4 and AILS) to measure perceived AI usefulness, behavioral intention to use AI, and AI literacy. These adapted scales showed strong factor loadings and measurement invariance across demographic subgroups in the present sample, which lent support to their construct validity. However, the current 3-item brief version did not include any items from the Ethics subscale of AILS. This raises potential concerns over the content validity of the adapted measure. Further research is needed to elucidate the ethical domain of AI literacy in the relationship with GenAI use and life satisfaction. Third, this study only measured the respondents’ GenAI use using a single self-reported item on their daily time. The single-item measure for GenAI use could not account for potential measurement errors and may introduce biases in the results. Besides, we did not assess the type of GenAI tool, purpose, and accuracy of use [11]. Further research is required to evaluate the antecedents and contextual determinants of ethical use of GenAI.

Fourth, the nonresponses in this study could be due to lower technological engagement and enthusiasm toward GenAI. This would induce a positive self-selection bias that would overestimate GenAI use and AI literacy in the sample. Unequal selection probabilities at geographic or household levels were not accounted for, leading to overrepresentation of the individuals living alone or in smaller districts. The use of self-report measures denotes common method bias. Fifth, this study focused on positive attributes of AI (usefulness and literacy) and did not assess fear of AI as a barrier to GenAI adoption. Technology anxiety has shown significant relationships with acceptance, literacy, and attitudes toward AI [58]. Compulsive use of ChatGPT has shown a negative indirect effect on life satisfaction via technostress [59]. Further studies should evaluate the role of technology anxiety and technostress in GenAI adoption for health-enhancing purposes. Six, this study did not include other psychological factors of life satisfaction, such as locus of control, intolerance of uncertainty, hope, and resilience. The omitted variables could lead to confounding bias in the present results. Future studies should consider these factors as alternative pathways that help explain the relationship between GenAI use and life satisfaction.

This study found relatively low GenAI use among Hong Kong adults in 2024, with a notable decreasing trend in GenAI use and AI literacy across age groups, especially among older adults. The findings showed that AI literacy was statistically positioned in the positive indirect pathway between GenAI use and life satisfaction and longitudinal studies are needed to test the causal direction of the relationships. These findings have potential implications for informing digital health initiatives and policies to leverage GenAI tools.