Family caregivers’ acceptance of Artificial Intelligence-enabled technologies for providing care to older adults Authors: Amanda Yee, Mark J Yaffe, Tibor Schuster, Sylvie Lambert, Samira Abbasgholizadeh-Rahimi Published: 2025 ABSTRACT AbstractBackgroundArtificial intelligence (AI)-enabled technologies hold promise for assisting in the care of an aging population. Few studies have focused on exploring family caregivers’ (FCGs) behavioural intention of using such innovation, and even fewer have employed a technology acceptance framework.ObjectiveThis study examined FCGs of older adults’ behavioural intention of using AI-enabled technologies for caregiving.Research design and methodsWe conducted a theory-based cross-sectional quantitative survey. Eligible FCGs for this study were: (1) aged 45–64; (2) residing in Quebec, Canada; (3) providing care for at least one older adult (65+); (4) having access to a computer or smartphone with internet connectivity; and, (5) having proficiency in reading and comprehending English or French.We adapted and expanded the Unified Theory of Acceptance and Use of Technology (UTAUT) framework to measure their behavioural intention of using AI-enabled technologies for caregiving. We used descriptive statistics and a random forest model to assess the most important predictive factors across nine variables and their direction of association with behavioural intention. The Consensus-Based Checklist for Reporting of Survey Studies (CROSS) guidelines was used for reporting the study’s results.ResultsAmong the polling firm’s 100,000 panelists, 2740 eligible individuals were randomly chosen to receive an email invitation to the study. Of 465 panelists who opened the survey (i.e., unique visitors),199 were eligible and completed the online survey. The random forest model explained between 56% and 86% of the behavioural intention variance of using AI, with social influence demonstrating the highest predictive relevance as indicated by a 35% increase in mean-squared error once removed from the model. Among the nine variables considered, six demonstrated a positive association with behavioural intention. These variables included social influence, effort expectancy, performance expectancy, perceived trust, confidence in healthcare professionals’ advice for the use of AI-enabled technologies, and facilitating connditions. The variables perceived cost and technology anxiety indicated a negative association with behavioural intention.Discussion and implicationsOur extended UTAUT model identified factors associated with FCGs' intention to use AI. While all nine variables contributed, attitudes toward AI within caregivers’ social circles was the strongest predictor. Stakeholders from industry, government, and healthcare can enhance the adoption of AI-enabled technologies in older adult care by leveraging facilitators and addressing barriers experienced by caregivers. Supplementary InformationThe online version contains supplementary material available at 10.1186/s12877-025-06249-1.Keywords: Older adults' care, Informal caregiving, Attitudes, Artificial Intelligence , Digital innovation, Machine learning, Digital health Introduction The global population over age 60 is predicted to exceed 2.1 billion by 2050 [1]. In Canada, as of 2024, close to 19% of the population was over 65 years old [2]. Family or friends may assume responsibility as family caregivers (FCGs) for assisting in the care of older adults [3]. Artificial intelligence (AI) is a field of computer sciences and engineering that aims to create machines capable of performing tasks that typically require human intelligence, such as learning, reasoning, problem-solving, perception, and decision-making [4]. There is evidence on the rapidly expanding application of AI-enabled technologies in aiding the care of older adults. The potential for AI in older adult care can encompass, for example, clinical decision systems to support healthcare decision-making, wearables that monitor physical well-being and behaviors, and robots designed to provide socio-emotional support [5, 6]. A scoping review synthesized the literature on AI-enabled technologies to support FCGs, and the included studies were all either model development and validation or technical report [7], suggesting the infancy of AI-enabled technologies' application in the real world. Another scoping review summarized the adoption of AI-enabled robots by healthcare providers in long term care homes; however, its use is challenged by factors such as inadequate training and resources (e.g., technology infrastructure) [8]. In Canada, there are discussions around the application of AI-enabled technologies for healthcare [9]. These include priority areas for AI use [10] and barriers to integrating AI into care, such as the concerns for inequality and regulations of AI [11]. To our knowledge, there are no data on adoption rates of AI-enabled technologies among FCGs. This might be due to the maturity/readiness of technologies for older adult care being in early phases of development or testing [12]. Given the emergence of AI-enabled technologies within older adult care, it would be valuable to examine what impacts users’ intention to use such technologies. While several intervention studies using specific technologies have assessed users' acceptance of such assistance, and have identified high acceptance rates [13, 14], others have found that acceptance varies based on the technology in question [15, 16]. Implications of such research have been hard to establish because of small sample sizes, lack of specificity as to whether AI was specifically applied to technologies studied, and findings were more descriptive than analytical. Moreover, a recent review on AI-enabled technologies aimed to support informal caregivers to patients, found only 10 studies and briefly acknowledged the challenges of implementation [7], suggesting the infancy of AI innovations and the value of understanding what impacts FCGs' decision to adopt such technologies. To assess facilitators and barriers impacting acceptance of technologies, a technology acceptance framework has often been utilized [17]. One of the widely applied frameworks is the Unified Theory of Acceptance and Use of Technology (UTAUT) by Venkatesh et al. [18]. The UTAUT model uses a dependent variable, namely behavioural intention, four independent variables (performance expectancy, effort expectancy, social influence, and facilitating conditions), and four moderator variables (gender, age, experience, and voluntariness of use) [18]. A limited number of studies employing UTAUT have examined FCGs' technology acceptance [19–21], and they lack clarity about whether the technology was enabled or enhanced by AI. Within Canada studies employing UTAUT have been focused on exploring technology acceptance by healthcare providers [22, 23] and older adults [24]. There are few studies exploring FCGs' acceptance of technologies for older adult care in Canada and it is often unclear whether the technologies were AI-enabled [25]. More Canadian-based research is needed to examine FCGs’ intention to use AI-enabled technologies and what impacts such a decision. FCGs often manage and perform various types of daily living and social-related tasks for older care recipients [26]. As such, they should be considered to be key users and stakeholders in technology acceptance studies, and this lack of attention to FCGs in technology-related studies has been raised [27]. In Canada the majority of FCGs are over 45 years old [28]. Likewise, in Quebec, those aged 45–64 represent the highest proportion of FCGs in the province, accounting for 30.6% of FCGs — more than other age groups [29]. FCGs between the ages of 15 to 44 and over 65 are less prevalent, and they are more likely to be caring for their parents or spouse/partner, respectively [29], rather than an aging parent. In contrast, FCGs aged 45–64 are more likely to provide care to their parents [29]. Therefore, considering the caregiving burden of middle-aged FCGs and current research gaps, we aimed to adapt UTAUT to explore middle-aged FCGs’ intention of using AI-enabled technologies for older adult care, as well as factors predicting such intention. Methods Study design We conducted a cross-sectional online survey with FCGs in the Canadian province of Quebec using a Web panel. The Consensus-Based Checklist for Reporting of Survey Studies (CROSS) guidelines [30] was used for reporting the study’s results (See Supplementary Material 1). Ethics approval was obtained from the Research Ethics Board of McGill University’s Faculty of Medicine and Health Sciences (A09-B88-22 A). Participants and data collection Eligible FCGs for this study were: (1) aged 45–64; (2) residing in Quebec, Canada; (3) providing care for at least one older adult (65+); (4) having access to a computer or smartphone with internet connectivity; and, (5) having proficiency in reading and comprehending English or French. To recruit such eligible participants and conduct our online survey, we used a Canadian firm specializing in research and polling by maintaining an ongoing large database of potential participants. Once our survey was programmed, it was pre-tested with 39 random participants from the firm’s participant base to assess its clarity and comprehension. Following subsequent small modifications to address clarity issues, an email was sent to prospective participants with a one-use survey link, or the survey could be accessed by logging into the firm’s accounts. Once this was done participants were asked to declare in which language they wished to complete the survey (i.e., English or French). They were then directed to read a consent form and requested to confirm their consent to participate voluntarily. This was followed by a single question that addressed their eligibility, and those eligible proceeded with the survey. The recruiting firm had a pre-established incentive structure whereby points were accumulated by participants and later traded in for rewards. This reward system was at arm’s length from us, and we provided no compensation to the participants. The survey was designed to collect descriptive demographic and caregiving-related information, along with closed-ended items to quantitatively measure participants’ intention to use AI-enabled technologies in their caregiving activities, and factors predicting their intention. The questions appeared in the same order for all participants, and no randomization took place. (See Supplementary Material 2 and 3). Before answering the closed-items questions, participants were asked to read a short definition of AI along with general examples of AI-enabled technologies, such as AI-enabled chatbots [31–36]. They were presented with an additional (yes/no) question that probed for their experience with AI-enabled technology which further inquired about the type of AI-enabled technologies used, and one four-point scale query assessing their level of knowledge about AI (See Supplementary Material 2 and 3). Questionnaire and measures The model we designed for our study included the original UTAUT variables i.e., performance expectancy, effort expectancy, social influence, and facilitating conditions [18]. Variables outlined by Pal et al. [37] were also added i.e., technology anxiety, perceived trust, perceived cost, and expert advice. We modified the variable ‘expert advice' to obtain broader clarity by dividing it into “confidence in the source of advice for care (i.e., healthcare professional vs. AI-enabled technology)” and “confidence in healthcare professionals’ advice for the use of AI-enabled technology”. As such a total of nine variables were included in our study’s model; their conceptualization is summarized in Table 1; and their conceived relationship to the dependent variable, behavioural intention, is shown in Fig. 1. Table 1UTAUT variables’ conceptualizationUTAUT variableConceptualizationBehavioural intentionFamily caregivers’ intention to use AI-enabled technologies in older adult care.Performance expectancy“The degree to which an individual believes that using the system will help him or her to attain gains in job performance” [18, p447].Effort expectancy“The degree of ease associated with the use of the system” [18, p450].Social influence“The degree to which an individual perceives that important others believe he or she should use the new system” [18, p451].Facilitating conditions“The degree to which an individual believes that an organizational and technical infrastructure exists to support [the] use of the system” [18, p453].Technology anxietyThe anxiety that an individual experiences when thinking of or using a technology [37, 38].Perceived trustThe trust that an individual has towards the technology in keeping the data secure [37].Perceived costThe financial investment that individuals believe is justified by the value they receive from technology [37].Confidence in the source of advice for care (healthcare professional vs. AI-enabled technology)Family caregivers’ confidence in the information/advice they received about their care recipients from a healthcare professional versus AI-enabled technologies.Confidence in healthcare professionals’ advice for the use of AI-enabled technologyFamily caregivers’ confidence in the healthcare professional’s advice regarding using AI-enabled technologies for care. UTAUT variables’ conceptualization Extended and adapted unified theory of acceptance and use of technology model The closed-ended items used to quantitatively measure participants’ intention, and factors predicting such intention were developed by adapting an extended UTAUT measurement instrument by Pal et al. [37], which had 34-itmes, and two items were not relevant to our study and were therefore removed. The remaining 32 closed-ended items were scored on a 5-point scale ranging from strongly disagree to strongly agree with the last option being "I don't know". Behavioral intention was measured using three items, and the nine predictor variables were measured using different numbers of items, ranging from one to six. For example, performance expectancy had six items, while social influence had three items (See Supplementary Material 4). The polling firm provided us with all anonymized survey responses, and we stored them within a password-protected file and laptop. Sample size A sample size of 200 was projected as adequate to allow for the estimation of population proportions within the 95% confidence intervals of width ± 7% or less. Such sample size is also suitable for fitting multiple linear regression models with up to 20 predictor variables [39] and was used for analysis using a random forest model (with predefined hyperparameters and features) to provide variable importance measures [40]. Data analysis and handling missing data The data cleaning, pre-processing, and analysis were performed using the statistical software package R and the graphical user interface RStudio (version 4.2.2) [41]. The code repositories are accessible on the Rahimi lab’s GitHub (https://github.com/rahimi-s-lab/AI-Caregivers-Survey.git), and further details can be obtained from the corresponding author. For preprocessing the five-point scale responses, they were assigned numerical values, with the option “I don’t know” being coded as a missing value, rather than a ‘neutral’ category since the latter implies some familiarity with the topic being asked [42], and this could not be assumed. We reversed-coded four out of the 32 items. For each participant, we generated an overall mean variable score based on the responses to each item. If a participant did not respond to any of the items used to measure a variable, the mean score was treated as missing data and excluded from the analysis. We employed two types of analysis: (1) descriptive statistics to outline the sample’s characteristics and behavioural intention scores and (2) supervised machine learning, i.e., random forests, to identifying factors enabling prediction of FCGs’ behavioural intention. The random forest method involves aggregating individual classification or regression trees, a process called ensembled learning, which creates a forest to minimize errors and enhance prediction accuracy [43, 44]. In this study, we applied default hyperparameters to develop the random forest model, which is outlined in the R package [45]. Random forests can establish variable importance measures, by generating a metric called the percent increase in mean squared error (%IncMSE). When the variable is permuted from the model, a higher %IncMSE indicates that the variable had a high relative importance in predicting the variable of interest. To provide a bagged estimate of the mean %IncMSE, alongside a 95% confidence interval, we generated 1000 bootstrapped random forests. We chose 1000 trees, since our model’s performance showcased that at least 200 trees were needed for the error rate to converge (See Supplementary Material 5). We also constructed scatterplots to examine the direction of association between predicted behavioural intention and the respective levels of a predictor variable. We conducted further analysis to examine observed and predicted behavioural intention. This involved plotting two random forest models: a full model included the nine predictor variables and a reduced model which removed one variable at a time while retaining the other eight. We converted the %IncMSE to a four-point scale to be able to visualize on a scatterplot the degree of change when a variable is removed from the model. In this study, a variable was seen to have ‘moderate to strong’ predictive capability if the behavioural intention score between the full versus the reduced models shifted more than one point, for example, a shift from ‘agree’ to ‘strongly agree’. The interquartile range was used to synthesize the distribution of behavioural intention score shifts across the participants. Finally, confusion matrices were used to illustrate how many participants predicted behavioural intention changed. For the confusion matrices, the behavioural intention score was divided into either disagreement (≤ 2.5) or agreement (> 2.5). Results Participants’ flow Among the polling firm’s 100,000 panelists, 2740 people over 45 years old were randomly chosen to receive an email invitation to the study. Of 465 panelists who opened the survey (i.e., unique visitors), 197 were ineligible, 97 submitted incomplete surveys, and 201 were retained (See Supplementary Material 6). Based on 201 surveys, the response rate, i.e. the ratio of participants who accessed the survey in response to the email invitations [24], was 17%. The completion rate, denoting the ratio of participants who finished the survey to those who initially accessed it [46], was 43%. However, two participants were removed due to age ineligibility, leaving 199 participants for analysis. Sample’s characteristics The eligible pre-test participants (n = 39) shared similar demographic characteristics to the overall sample (n = 199) (See Supplementary Material 7). The characteristics of the latter group have a mean age of 56.7 (standard deviation [SD] = 5.49) and an average time lived in Canada for 55 years (SD = 9.07). The majority of the 199 participants completed the French survey (n = 173) and were women (n = 128). While a few completed junior college/CEGEP (n = 82) and were employed full-time (n = 88). Most of the participants were caring for one older adult (n = 167), commonly for a parent (n = 140). Eighty-three of 199 care recipients lived in their own homes. The mean number of years and hours per week spent as an FCG was 7.7 years (SD = 6.94) and 16 hours (SD = 19.5), respectively. Among caregiving tasks, household duties and psychosocial care (e.g., emotional support and companionship) were the top two tasks frequently performed. Of the 199 participants, 109 reported low AI knowledge (55%), while 16 indicated experience with using AI (8%). Behavioural intention for using AI-enabled technologies for caregiving Figure 2 is a stacked bar chart outlining the level of agreement/disagreement with items measuring behavioural intention. About 45% of participants were not sure if they would use AI in the future when providing older adult care. If AI was accessible, 62% of participants indicated agreement about their intent to use it. In terms of frequently investing in and using AI-enabled technologies for older adult care, 37% reported being uncertain about it. Responses to behavioural intention items Random forest analysis – variable importance The random forest method was used on complete cases (n = 115) and based on the ad hoc sensitivity analysis (See Supplementary Material 8), there were no meaningful differences in terms of demographic characteristics, caregiving experiences, and AI knowledge or experiences between complete cases and the entire sample (n = 199). The random forest model of the nine predictor variables showed a 56–83% variance explained in the boot-strapped related samples. The variable-specific means of estimated relative importance measures and 95% confidence intervals are presented in Fig. 3. The mean relative importance measures ranged from 15–34%. Social influence showed the highest importance, with a 34% IncMSE (95% confidence interval: 25–43%). To assess the statistical significance of the variable importance measures, permutation-based p-values were estimated using the rfPermute package in R. This approach involved generating null distributions for the %IncMSE importance metric by repeatedly permuting the response variable (BIr.score) across 50 permutations for each of the 1000 bootstrap samples. The p-values indicate the probability of observing a predictor’s importance at least as extreme as the one obtained, assuming the null hypothesis of no association with the response is true. Across the bootstrap iterations, the mean p-values for the nine predictors ranged from 0.01 to 0.045, with social influence (SIr.score), technology anxiety (TAr.score) and effort expectancy (EEr.score) exhibiting the lowest mean p-value of 0.01, indicating statistical significance. The proportion of bootstrap iterations where each predictor’s p-value was ≤ 0.05 ranged from 0.74 to 1.00, with ‘social influence’, ‘effort expectancy’, ‘technology anxiety’, ‘performance expectancy’ and ‘source of advice for care’ being significant in 100% of iterations, reinforcing its importance. These p-values, combined with the confidence intervals presented in Fig. 3, provide evidence of the statistical reliability of the variable importance rankings. Variable Importance Measures: pooled (bagged) estimates with 95% confidence intervals and permutation-based p-values Figure 4 outlines the direction of association between the predicted behavioural intention and predictor variables (linear slope estimate based on the random forest predictions). We identified positive association for the six predictor variables: social influence (β ≈ +0.58), effort expectancy (β ≈ +0.50), performance expectancy (β ≈ +0.43), perceived trust (β ≈ +0.43), confidence in healthcare professionals’ advice for the use of AI-enabled technologies (β ≈ +0.42) and facilitating conditions (β ≈ +0.37). There were two negative associations: perceived cost (β ≈ −0.12) and technology anxiety (β ≈ −0.64). Meanwhile, a quadratic association (i.e., inverse U-line instead of monotonous positive or negative trend) was identified for the predictor variable, confidence in the source of advice for care (healthcare professional vs. AI-enabled technologies). Direction of Association: scatter density plots with smooth spline trend functions and pointwise 95% confidence bands Further analysis examined the observed and predicted behavioural intention and was plotted using two random forest models: full (green dots) and reduced (red dots) models (See Supplementary Material 9). For this analysis, 120 participants’ datasets were utilized, this included 115 complete cases plus 5 cases, whereby they had all variable scores present but only missed the behavioural intention score. Based on that figure, the model appears to have good accuracy, as there is alignment between the observed and predicted behavioural intention scores, closely following the black diagonal line, indicating perfect agreement. However, despite the low variance, our model showed a slight underestimation of lower behavioural intention scores and an overestimation of higher behavioural intention scores. Moreover, the participants’ changes in predicted behavioural intention from the full to the reduced model were between the interquartile range of −0.06 to + 0.06. The change in predicted behavioural intention was also evident when looking at the confusion matrices. Based on the matrices, regardless of the variable removed in the reduced models, only one to seven participants changed from agreement to disagreement and zero to four participants changed from disagreement to agreement. Removing one predictor variable at a time from the model, the change in behavioural intention did not considerably shift. This suggests that all predictor variables together have complementary value in predicting FCG’s behavioural intention to use AI-enabled technology. Discussion & Implication We aimed to investigate FCGs’ acceptance of AI-enabled technologies for older adult care, focusing on two main objectives: understanding their behavioural intention and analyzing the predictive capability of predictor variables in influencing behavioural intention. In our study, we observed the following three main findings: First, the study revealed that the average behavioural intention score was high, suggesting an overall acceptance of AI-enabled technologies among caregivers for the use of these technologies in providing care to their loved ones. This is in line with previous works on FCGs’ acceptance of digital health technologies [13, 47]. Yet, a detailed analysis of the responses to specific behavioural intention items showed varying levels of acceptance depending on the circumstance. For instance, about two-thirds of FCGs reported accepting AI-enabled technologies’ use if they have access to it, while almost half were uncertain about using AI-enabled technologies in the future. The uncertainty could stem from our sample of FCGs lacking experience and knowledge regarding AI. Also, misconceptions about AI may manifest through public discourses or by companies potentially sensationalizing AI capabilities [48, 49]. This emphasizes the need for improved AI literacy among FCGs through educational initiatives [50] and transparency from AI companies and AI developers to reduce over-sensationalising or misrepresenting AI. Second, social influence emerged as the most important predictor variable relative to the other variables such as performance expectancy or perceived trust. Our finding was consistent with the findings from past studies [19, 51]. For example, a South Korean study [19] identified a strong association between social influence and intention to use digital technologies. Other work suggested otherwise, for example, a Canadian-based study of FCGs reported that social influence was not statistically significant in influencing behavioural intention to use a mobile application [25]. Additionally, the study found a non-significant finding related to performance expectancy and effort expectancy on intention to use the technology [25]. We assume that these differences could be explained by differences in technology type, analysis approach, study locales, cultural differences in caregiving, and cultural attitudes toward technology. Third, all nine predictor variables in our study contributed complementarily to explaining FCGs’ behavioural intention. Six of the variables were facilitators: performance expectancy, effort expectancy, social influence, facilitating conditions, perceived trust, and confidence in healthcare professionals’ advice for the use of AI-enabled technologies. Two were considered barriers: technology anxiety and perceived cost. Other studies on digital technology amongst other user groups have also supported the facilitating impact of, for example, performance expectancy [52] and technology anxiety as a barrier [19]. Finally, confidence in the source of advice for care (healthcare professional vs. AI-enabled technologies) appears to have a quadratic association with behavioural intention to use AI-enabled technologies. Given that this variable was adapted into our model and yielded such an unusual finding, we recommend that future studies should explore it further to better understand its association with intention. Our study contributes to the growing technology acceptance literature and potential practice in terms of AI adoption among FCGs. The study reinforces the relevance of the core UTAUT variables and the applicability of including and modifying other relevant variables to the model to help better refine our understanding of participants’ technology acceptance. Also, highlighting that social influence is the most important predictor, and identifying the facilitators and barriers to AI use among FCGs can help stakeholders to collaboratively develop initiatives to either enhance facilitators or mitigate barriers to FCGs’ acceptance of AI-enabled technologies. Technology acceptance facilitating interventions that outline a technology and pertinent considerations (e.g., effectiveness) through a video, have improved technology users’ performance expectancy [53] and anxiety about a technology [54]. Another approach that should be considered is a user-centred approach through co-design and participatory methods [55] with FCGs and older adults to ensure that AI-enabled technologies are developed to be user-friendly and potentially allow users to voice any concerns at the early stages of development. The co-design approach may help shape societal attitudes, which may in turn shape the influence that other people have on FCGs' decision to use AI-enabled technologies. At the level of government, there are ongoing policy efforts towards AI. For example, there is an AI and Data Act under Bill C-27, under consideration in the House of Commons, that aims to regulate the development and implementation of AI [56]. In addition, as generative AI (e.g., OpenAI's ChatGPT) is emerging, the Government of Canada has been suggesting policies and recommendations to navigate its potential consequences at the federal level, such as understanding how and what is being used to train the AI [57]. Also, with AI-enabled technologies cost being a barrier to intention, there should be financial programs in place, similar to how some provincial governments have financial programs for assistive products [58]. Finally, similar to the guidelines to help healthcare providers recommend mobile health applications [59], there should be ones that are designed by stakeholders from the AI industry, healthcare providers/organizations and patients/caregivers that outline recommendations on using AI-enabled technologies for caregiving. By understanding FCGs' acceptance, stakeholders can better adapt to improve adoption of AI-enabled technologies for older adult care. Consequently, the potential use of AI-enabled technologies by FCGs could have positive impacts on older adults’ well-being. Technologies, including robotics, could promote independence by monitoring older adults health and behavioral patterns as they age at home and can provide rehabilitation assistance [60], as well as offer activities to improve cognitive function [5]. Beyond the physical and cognitive impacts, AI-enabled technologies has the potential to promote emotional well-being and facilitate social interactions [5]. Although there are probable benefits, these should not be over-emphasized as there are ethical considerations [60], such as the potential of AI to dehumanize care [55]. Strengths, Limitations, and Future Directions This is the first study to assess FCGs’ acceptance of AI-enabled technologies for older adult care by extending the UTAUT and employing a random forest analysis to assess the factors predicting their acceptance. However, this study is not without limitations. Due to time constraints, a confirmatory factor analysis was not performed, but rigour was maintained by modifying and adapting a previous UTAUT measurement tool and pilot testing it. Further, our sample included both English and French-speaking participants in Quebec. As such, their attitudes and ideas about technology acceptance may be specific to this population. Additionally, by recruiting FCGs from a single polling firm, there is a potential for a sampling bias. Likewise, as we are aiming to understand their behavioural intention to use AI-enabled technologies, there is the potential for a social desirability response bias. Moreover, due to our unbalanced (i.e., not equally representative) sample in some demographic variables and AI knowledge and experience, variables including education, past experience using AI, age, gender, knowledge of AI, and employment did not appear to show any potential importance (See Supplementary Material 10). Future studies should consider purposive ensuring a balanced sample across various demographic characteristics to better assess its impact on behavioural intention. Given the cross-sectional nature of our study, we are not able to examine the potential temporal changes in FCGs’ behavioural intention to use AI-enabled technologies. Therefore, future work should focus on conducting longitudinal studies to assess for any evolving factors influencing FCGs’ sustained acceptance of AI-enabled technologies. Finally, future research could optimize the machine learning model by adjusting the hyperparameters to improve predictive accuracy. Conclusion The aging population places a strain on FCGs, and with the emergence of AI-enabled technologies, this study showed that social influence appears to be an important factor in contributing to FCGs’ intention to use AI. In addition, various facilitators and barriers were shown to contribute to FCGs overall acceptance of AI. To promote AI-enabled technologies use, in caregiving among FCGs stakeholders need to ensure better education, develop guidelines, and employ user-centric approaches. Supplementary Information Supplementary Materials. Acknowledgements SAR is Canada Research Chair (Tier II) in Artificial Intelligence and Advanced Digital Primary Health Care, received salary support from a Research Scholar Junior 1 Career Development Award from the Fonds de Recherche du Québec-Santé during part of this project, and her research program is supported by the Natural Sciences Research Council Discovery (grant 2020-05246). Abbreviations Artificial intelligence Family caregivers Unified Theory of Acceptance and Use of Technology Consensus-Based Checklist for Reporting of Survey Studies Authors’ contributions AY: Conceptualization, Methodology, Investigation, Formal analysis, Writing - Original Draft, Visualization, Funding acquisition. MY: Conceptualization, Methodology, Writing - Review and Editing, Supervision. TS: Methodology, Software; Formal analysis, Writing - Review and Editing. SL: Writing - Review and Editing. SR: Conceptualization, Methodology, Software, Formal analysis, Supervision, Funding acquisition, Writing - Review and Editing. Funding This study was funded by Canadian Institutes of Health Research (CIHR) Canada Graduate Scholarships, Innovation Fund of Roche Canada and NSERC DG (Grant 2020–05246). The funding bodies had no role nor influence into the conceptualization, design, data collection, analysis, decision to publish, nor preparation of the manuscript for submission. Data availability The dataset supporting the conclusions of this article can be obtained from the corresponding author. The code repositories are accessible on the Rahimi lab’s GitHub (https://github.com/rahimi-s-lab/AI-Caregivers-Survey.git). Declarations Ethics approval and consent to participate The study adheres to the Declaration of Helsinki and ethics approval was obtained from the Research Ethics Board of McGill University’s Faculty of Medicine and Health Sciences (A09-B88-22 A). Informed consent was obtained electronically from each participants prior to the start of the online survey. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Materials. Data Availability Statement The dataset supporting the conclusions of this article can be obtained from the corresponding author. The code repositories are accessible on the Rahimi lab’s GitHub (https://github.com/rahimi-s-lab/AI-Caregivers-Survey.git).