Loneliness and AI chatbot dependence in young adults: the mediating role of perceived social support Authors: Özkan Uğuz, Satı Doğan, Dilay Güngör Published: 2026 ABSTRACT AbstractBackground and objectivesRapid advancements in artificial intelligence (AI) have facilitated the widespread integration of AI chatbots into everyday life, positioning them as alternative agents for social and emotional interaction. While such technologies may help regulate experiences of loneliness, stress, and social isolation, increasing reliance on AI-mediated interactions raises concerns regarding the emergence of maladaptive use patterns, including AI chatbot dependence. Despite growing attention to technology-related dependence, the underlying psychosocial mechanisms remain insufficiently examined. Accordingly, this study investigates the association between loneliness and AI chatbot dependence, with a particular focus on the mediating role of perceived social support among young adults.MethodsA quantitative research design was employed using a relational survey approach. Data were collected via an online questionnaire from 610 young adults (aged 18–30) residing in Türkiye. Measures included the UCLA Loneliness Scale–6, the AI Chatbot Dependence Scale, and the Perceived Social Support Scale. Data were analyzed using descriptive statistics, correlation analysis, and structural equation modeling (SEM) conducted in AMOS 24.0. The mediating effect was tested using a bootstrapping procedure.ResultsLoneliness was positively associated with AI chatbot dependence, whereas it was negatively associated with perceived social support; perceived social support, in turn, was negatively associated with AI chatbot dependence. Mediation analyses indicated that loneliness significantly predicted AI chatbot dependence both directly and indirectly, with perceived social support partially mediating this relationship. The proposed model demonstrated an acceptable fit to the data.ConclusionThese findings suggest that robust social support networks may function as a protective factor against emerging forms of AI chatbot dependence and highlight the importance of psychosocial interventions. Accordingly, coordinated efforts across universities, public institutions, local authorities, and mental health professionals are needed to develop evidence-based and sustainable interventions aimed at strengthening social connectedness and enhancing coping mechanisms for loneliness among young adults.Keywords: Loneliness, Perceived social support, Artificial intelligence, AI chatbot dependence, Young adults Introduction Artificial intelligence (AI) chatbots have proliferated rapidly in recent years, offering transformative potential for daily life and contributing to the design of a more efficient and innovative future [1, 2]. Academic evidence suggests that approximately 65% of researchers actively employ AI chatbots in research and educational activities [3]. In addition, more than half of hospitals in the United States are projected to integrate generative AI into electronic health record systems by the end of 2025 [4]. Recent studies in Türkiye indicate generally positive attitudes toward AI across diverse demographic and professional groups [5–8], highlighting a growing inclination toward the adoption and future use of AI chatbots [9]. Beyond facilitating access to information, AI chatbots are increasingly functioning as interactive tools capable of providing emotional support to users [10, 11]. Prolonged engagement with these technologies may foster emotional attachments, which can, over time, give rise to maladaptive usage patterns. In the literature, such patterns are conceptualized as AI chatbot dependence, typically delineated through two interrelated processes. Cognitive dependence refers to excessive reliance on chatbots for decision-making and problem-solving [12, 13], whereas emotional dependence encompasses the formation of parasocial bonds arising from chatbots’ unconditional acceptance, empathetic responses, and perceived social closeness [14, 15]. However, this conceptualization alone does not sufficiently explain why some individuals become more vulnerable to developing dependence than others. Despite this conceptual framework, the role of psychosocial factors in shaping these dependency processes remains insufficiently examined. Taken together, these observations highlight a critical gap in the literature. Existing research has predominantly focused on the roles of loneliness and social support in problematic internet use, yet mechanisms specific to AI chatbot dependence and potential cross-cultural variations remain underexplored. Although the literature on problematic internet use and social media dependence has expanded, relatively little empirical attention has been devoted to the psychosocial processes underlying reliance on AI-based conversational agents. In particular, the interplay between loneliness and perceived social support in shaping AI chatbot dependence remains poorly understood, especially in non-Western contexts. To address this gap, the present study investigates the mediating role of perceived social support in the relationship between loneliness and AI chatbot dependence among young adults in Türkiye. Theoretical background and hypothesis development Loneliness and AI chatbot dependence Loneliness is conceptualized not merely as physical isolation but as the perceived insufficiency of meaningful social relationships [11], and it has been consistently identified as a risk factor for seeking alternative forms of social connection. According to the Compensatory Internet Use Theory (CIUT), individuals with unmet social or emotional needs engage with digital technologies not only for instrumental or recreational purposes but also to alleviate negative emotions such as loneliness and to mitigate psychological distress [16]. Within this framework, individuals with limited social resources are more likely to utilize online tools as a form of emotional compensation. AI chatbots provide personalized, response-oriented, and emotion-centered interactions, rendering them particularly appealing to individuals experiencing loneliness [17, 18]. Empirical evidence indicates that lonely individuals—especially those with limited social support—engage more frequently with chatbots and may develop cognitive and emotional dependence through the formation of parasocial or empathetic bonds [12]. Repeated interactions can reinforce such dependence, potentially substituting for real-life social interactions over time and increasing the risk of maladaptive or addictive-like use patterns. Among adolescents, psychosocial factors contributing to loneliness include a lack of meaningful social connections, social anxiety, and experiences of social exclusion [19, 20]. In these contexts, AI chatbots may function as effective “digital coping tools,” providing immediate attention, guidance, and emotional feedback to address unmet social needs [21, 22]. Consequently, heightened loneliness may drive more intensive engagement with AI chatbots, thereby creating conditions conducive to potential dependence. From this perspective, the following hypothesis is proposed: H1. Loneliness positively influences AI chatbot dependence. H1. Loneliness positively influences AI chatbot dependence. Loneliness and perceived social support Perceived social support refers to individuals’ subjective evaluation of the emotional, cognitive, and instrumental assistance received from significant others [23]. The relationship between loneliness and perceived social support is frequently examined within the framework of the Buffering Hypothesis (BH) in social and health psychology literature. According to this perspective, social support functions as a critical protective resource that mitigates the psychological impact of stressful life events and negative emotional experiences [24]. Specifically, higher levels of perceived social support are associated with attenuation of stress effects and adverse psychological states [25]. The literature indicates that individuals experiencing loneliness tend to evaluate their social relationships more negatively and perceive available support as more limited than it objectively is [26]. Additionally, such individuals may develop cognitive tendencies, including expectations of social rejection, negative interpretations of social interactions, and interpersonal mistrust, which further diminish perceptions of social support within their environment [27, 28]. Consequently, loneliness is likely to reduce perceived social support. Based on this theoretical and empirical foundation, the following hypothesis is proposed: H2. Loneliness negatively influences perceived social support. H2. Loneliness negatively influences perceived social support. Perceived social support and AI chatbot dependence High levels of perceived social support function as a protective resource, facilitating coping with stress and negative emotional experiences while promoting the fulfillment of emotional needs through offline social connections [29]. Within this framework, robust offline social networks reduce the necessity to compensate for social deficits via online tools, thereby decreasing the risk of problematic engagement and AI chatbot dependence [30, 31]. Conversely, limited perceived social support may drive individuals to seek social and emotional fulfillment through AI chatbots. Platforms such as ChatGPT and Replika provide personalized feedback and parasocial interactions, rendering them attractive sources of emotional support for lonely users [32, 33]. Prolonged engagement with these digital tools may increase cognitive and emotional reliance, potentially contributing to the development of dependence. Accordingly, the following hypothesis is proposed: H3. Perceived social support negatively influences AI chatbot dependence. H3. Perceived social support negatively influences AI chatbot dependence. Mediating role of perceived social support Integrating the BH and the CIUT suggests that perceived social support may serve as a key mediator in the relationship between loneliness and AI chatbot dependence. According to the BH, high levels of perceived social support provide individuals with a sense of protection when coping with stress and negative social emotions, thereby reducing the likelihood of developing maladaptive behavioral patterns. In this context, support from offline social networks mitigates the impact of negative emotional states and lowers the risk of excessive engagement with technology. In line with CIUT, individuals may use digital tools to compensate for unmet social needs. Consequently, highly lonely individuals are more likely to seek online interactions, including AI chatbots, to fulfill these needs. Empirical evidence indicates that low perceived social support mediates the relationship between loneliness and problematic technology use, such that as perceived social support decreases, loneliness exerts a stronger influence on excessive engagement with digital tools [34]. Based on these theoretical considerations, perceived social support is expected to function as a critical mediator in the link between loneliness and AI chatbot dependence. High levels of perceived social support may attenuate the positive effect of loneliness on AI chatbot dependence, thereby reducing the risk of maladaptive AI use. Accordingly, the following hypothesis is proposed: H4. Perceived social support mediates the relationship between loneliness and AI chatbot dependence. H4. Perceived social support mediates the relationship between loneliness and AI chatbot dependence. The theoretical framework is presented in Fig. 1. Research model Method Research model This study was conducted within the framework of a quantitative research approach and employed a relational survey design to examine the direction and strength of associations among the variables [35]. Given that the primary objective was to explore the relationships among the study variables, this design was considered appropriate. Data were collected at a single time point, resulting in a cross-sectional research design. Accordingly, the findings represent associative patterns among the variables and do not allow for causal interpretations. Population and sample of the study The target population comprised young adults residing in Türkiye. This demographic was selected due to its high level of engagement with digital technologies. Indeed, social media and internet engagement increase from adolescence into young adulthood, with particularly high levels of online interaction observed during this developmental period [36]. The required sample size was estimated using G*Power 3.1.9.7, assuming an effect size of 0.27 and a statistical power of 0.80, yielding a minimum requirement of 434 participants. Participants were recruited via a non-probability convenience sampling approach, based on voluntary online participation. This strategy was deemed appropriate given the exploratory nature of the study and the objective of targeting individuals who actively use AI-based chatbots in daily life. Data were collected over a two-month period through an online survey platform. A total of 655 respondents initially completed the survey. Following data screening, 45 cases were excluded due to excessive missing responses (i.e., more than 50% incomplete). The final analytical sample comprised 610 participants who met the inclusion criteria and provided complete data suitable for statistical analysis. The inclusion criteria were defined a priori as follows: i.Aged between 18 and 30 years, Aged between 18 and 30 years, ii.Residing in Türkiye at the time of data collection, Residing in Türkiye at the time of data collection, iii.Having used AI chatbots (e.g., ChatGPT, Replika, Gemini) at least once per week during the previous three months, and. Having used AI chatbots (e.g., ChatGPT, Replika, Gemini) at least once per week during the previous three months, and. iv.Providing informed and voluntary consent to participate in the study. Providing informed and voluntary consent to participate in the study. Instruments In this study, data were collected using the UCLA Loneliness Scale–6, the AI Chatbot Dependence Scale, and the Perceived Social Support Questionnaire. UCLA loneliness scale-6 The scale developed to measure adults’ levels of loneliness was originally created by Neto [37], and its Turkish adaptation and validity–reliability study were conducted by Sarıçam [38]. The scale has a unidimensional structure and consists of six items (e.g., “I feel isolated from other people”), with the second item reverse-coded. Items are rated on a Likert-type scale ranging from 1 (not at all appropriate) to 4 (completely appropriate), yielding a total score range of 6–24, with higher scores indicating greater levels of loneliness. While the Cronbach’s alpha coefficient of the Turkish version was reported as 0.77 [38]. In the present study, the scale exhibited excellent internal consistency, with a Cronbach’s alpha of 0.93. AI chatbot dependence scale The scale developed by Zhang et al. [39] and adapted into Turkish by Yıldız Durak and Balıkçı [40] is an assessment tool designed to measure individuals’ levels of dependence on AI-based chatbots. The scale has a unidimensional structure and consists of eight items. Each item (e.g., “Compared to other people or things, I prefer to spend my time with AI chatbots”) is rated on a seven-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree), yielding a total score range of 8–56, with higher scores indicating greater levels of AI chatbot dependence. In the Turkish adaptation study, the internal consistency coefficient was reported as 0.90, demonstrating that the scale is a reliable measurement tool [40]. In the present study, Cronbach’s alpha was 0.98, indicating excellent internal consistency. Perceived social support questionnaire The scale is a measurement tool developed to assess individuals’ levels of perceived social support from their social environment. It was originally developed by Petersen et al. [41] and subsequently adapted to the Turkish cultural context by Türk et al. [42]. The scale has a unidimensional structure and consists of three items, allowing for brief and practical administration. Items (e.g., “When I am ill, I can ask my friends and family without hesitation to take care of things that are important to me”) are rated on a five-point Likert scale ranging from 1 (not at all true) to 5 (completely true), yielding a total score range of 3–15. Higher scores indicate higher levels of perceived social support, whereas lower scores reflect insufficient perceived social support. While the Cronbach’s alpha coefficient of the Turkish version was reported as 0.82 [42]. The scale demonstrated excellent internal consistency in the present study, with a Cronbach’s alpha of 0.91. Data collection Data were collected online between September and December 2025. Participants were recruited through announcements disseminated across multiple online channels, including social media platforms (e.g., Instagram, X, and Facebook) and university student communities. The announcements briefly described the study objectives, eligibility criteria, and the voluntary nature of participation. Interested individuals were invited to access the survey via a Google Forms link or by scanning a QR code provided in the announcements. Prior to accessing the questionnaire, participants were presented with an electronic informed consent form detailing the study objectives, data confidentiality procedures, and the right to withdraw at any time without penalty. Only participants who provided consent were permitted to proceed. To prevent duplicate participation, participants were asked to provide their name and surname solely for verification purposes; these identifiers were stored separately from survey responses, excluded from the analytical dataset, and permanently deleted prior to analysis. All responses included in the analysis were fully anonymized. The average completion time for the questionnaire was approximately 15 min. Data analysis Data were analyzed using SPSS 26.0 and AMOS 24.0. Initially, common method bias (CMB) was assessed via Harman’s single-factor test. Recognizing the limitations of this approach, several procedural remedies were implemented during data collection, including ensuring participant anonymity, employing varied Likert-type response formats, and counterbalancing the order of questionnaire sections. Nonetheless, as the data were self-reported, the potential for common method variance cannot be entirely ruled out [43]. Descriptive analyses involved calculation of frequencies and percentages for categorical variables, and means and standard deviations for continuous variables. Associations among variables were examined using Pearson’s correlation coefficients, with the magnitude of correlations interpreted according to established guidelines in the literature [44]. Structural Equation Modeling (SEM) was employed to test the proposed mediation model. SEM allows for the simultaneous estimation of direct and indirect relationships among variables within a single framework and facilitates evaluation of overall model fit between the theoretical model and the observed data [45–47]. Compared with traditional regression approaches, SEM offers methodological advantages for testing mediation effects and estimating indirect paths. Given that the primary objective of the present study was to examine both the direct association between loneliness and AI chatbot dependence and the indirect mechanism via perceived social support, SEM was deemed an appropriate analytical approach. Because the Perceived Social Support Questionnaire comprises only three items, analyses were conducted using observed composite scores rather than latent constructs to avoid potential identification issues [48]. Model fit was evaluated according to the following criteria: χ²/df ≤ 5, NFI, IFI, CFI, and TLI ≥ 0.90, and RMSEA ≤ 0.08 [49, 50]. The significance of mediation effects was tested using the Sobel test and a bootstrap procedure with 5,000 resamples, and a significance level of p < 0.05 was adopted for all analyses. Additionally, a post-hoc power analysis was performed to determine whether the final sample size provided sufficient statistical power to detect both direct and indirect effects within the model. The results indicated adequate power to detect moderate path coefficients (β = 0.30), supporting the statistical robustness and reliability of the relationships reported in the structural model. Ethical approval This study was conducted in accordance with the ethical principles of the Declaration of Helsinki and was approved by the Ege University Faculty of Medicine Medical Research Ethics Committee (Approval date: October 16, 2025; Decision No: 25-10.1T/58). Informed consent was obtained from all individual participants included in the study. Results Participants’ characteristics The mean age of the participants was 23.84 ± 3.67 years, with more than half of the sample falling within the 18–24 age range (55.4%). Regarding gender distribution, 54.6% of the participants were female and 45.4% were male. The majority of participants were single (79.5%), and in terms of educational attainment, most held an undergraduate degree (41.0%) or an associate degree (24.6%). The average daily internet use was 8.54 ± 5.07 h, with 57.2% of participants reporting spending more than 7 h online per day. With respect to AI-based chatbot usage frequency, 38.0% of participants reported using chatbots several times per week. Regarding usage purposes, the most common reason was academic/educational use (39.2%), followed by use for emotional support (23.8%). Testing of common method bias CMB can distort the observed relationships among constructs when data are collected using a single measurement method [41]. Because this study relied on self-report instruments, the potential presence of CMB was systematically evaluated. In this context, the full collinearity test proposed by Kock [42] was applied, and the results showed that the variance inflation factor (VIF) values for loneliness, AI chatbot dependence, and perceived social support were all below the recommended threshold of 3.3, indicating the absence of a serious common method bias. To further substantiate this finding, Harman’s single-factor test was conducted, revealing that the first factor explained less than 40% of the total variance, thereby providing additional evidence that CMB was not a significant concern in the present study (Table 1). Following the assessment of common method bias, descriptive statistics were examined. Table 1Full collinearity testingLonelinessAl-Chatbot DependencePerceived Social Support2.182.671.89 Full collinearity testing Descriptive statistics and normality As presented in Table 2, the mean scores and standard deviations of participants’ loneliness, AI chatbot dependence, and perceived social support were within acceptable ranges. The distributional properties of the variables were further examined through skewness and kurtosis values, all of which fell within the acceptable threshold of ± 2.0, indicating that the data were approximately normally distributed. Accordingly, the use of maximum likelihood estimation in subsequent analyses was deemed appropriate. To further explore the relationships among variables, correlation analyses were conducted. Table 2Descriptive statistics of variablesMinMaxMeanSDSkewnessKurtosisLoneliness62415.033.19-1.11-2.58AI chatbot dependence85631.088.25-1.360.95Perceived social support3159.402.130.810.73 Descriptive statistics of variables Correlation check Figure 2 depicts the relationships among participants’ age, daily internet use, and the primary study variables. Correlation analyses indicated a moderate, statistically significant negative association between age and daily internet use (r = − 0.44, p < 0.001), suggesting that younger participants reported longer daily internet engagement. Loneliness and AI chatbot dependence were moderately to strongly negatively correlated with perceived social support (r = − 0.55, 95% CI [− 0.60, − 0.49] and r = − 0.61, 95% CI [− 0.66, − 0.56], respectively; p < 0.001), indicating that higher levels of loneliness and chatbot dependence were associated with lower perceived social support. Additionally, a strong positive correlation was observed between loneliness and AI chatbot dependence (r = 0.71, 95% CI [0.66, 0.74], p < 0.001). These findings provided the basis for subsequent SEM analyses. Correlation heatmap (※, p < 0.001; AIU, average ınternet use; ULS-6, loneliness; AICD, AI chatbot dependence; PSS, percevied social support) Mediator check Figure 3 illustrates the direct relationship between loneliness and AI chatbot dependence. SEM analyses indicated that loneliness exerted a significant positive effect on AI chatbot dependence (β = 0.74, 95% CI [0.69, 0.79], p < 0.001). The model demonstrated acceptable fit indices (χ²/df = 4.3, RMSEA = 0.074, CFI = 0.976, NFI = 0.969, IFI = 0.976, TLI = 0.971), indicating good alignment between the proposed model and the observed data. Loneliness accounted for approximately 55% of the variance in AI chatbot dependence (R² = 0.55). Yalnızlık ve AI-chatbot bağımlılığının yapısal eşitlik modeli için standartlaştırılmış katsayısı (* p < 0.001) Figure 4 illustrates the mediating role of perceived social support in the relationship between loneliness and AI chatbot dependence. SEM analyses indicated that loneliness exerted a significant negative effect on perceived social support (β = −0.60, 95% CI [− 0.65, − 0.55], p < 0.001). Perceived social support, in turn, was significantly negatively associated with AI chatbot dependence (β = −0.31, 95% CI [− 0.37, − 0.25], p < 0.001). When perceived social support was included in the model, the direct effect of loneliness on AI chatbot dependence decreased but remained statistically significant (β = 0.55, p < 0.01), indicating partial mediation. Algılanan sosyal destek, yalnızlık ve AI-chatbot bağımlılığının yapısal eşitlik modeli için standartlaştırılmış katsayıları (* p < 0.001) Bootstrap analyses supported the statistical significance of the mediating effect, as the 95% confidence intervals did not include zero (β = 0.46, 95% CI [0.33–0.59]). The Sobel test further confirmed the significance of the mediation effect (Table 3). These results indicate that perceived social support partially mediates the relationship between loneliness and AI chatbot dependence. The mediation model demonstrated good fit across all standard indices (χ²/df = 3.49, RMSEA = 0.064, CFI = 0.976, NFI = 0.967, IFI = 0.976, TLI = 0.972), indicating alignment between the proposed model and the observed data. Together, loneliness and perceived social support accounted for approximately 74% of the variance in AI chatbot dependence within the mediation model. Table 3Indirect effect of mediator modelPathCoefficient95%CILLULLoneliness → Perceived social support → AI chatbot dependence0.460.3330.590CI Confidence interval, LL Lower limit, UL Upper limit Indirect effect of mediator model CI Confidence interval, LL Lower limit, UL Upper limit Discussion This study investigated the relationship between loneliness and AI chatbot dependence among young adults and examined the mediating role of perceived social support in this relationship. The results indicated that loneliness was significantly and positively associated with AI chatbot dependence, whereas perceived social support exhibited significant negative associations with both loneliness and AI chatbot dependence. Furthermore, perceived social support partially mediated the relationship between loneliness and AI chatbot dependence. These findings suggest that higher levels of perceived social support may reduce the likelihood of developing AI chatbot dependence among young adults, indirectly attenuating the effect of loneliness and thereby functioning as a protective mechanism. To the best of current knowledge, this study represents one of the first empirical examinations of the mediating role of perceived social support in the relationship between loneliness and AI chatbot dependence. Accordingly, the results contribute to advancing the emerging literature on the psychosocial mechanisms underlying AI chatbot dependence. Perceived social support may function as a protective mechanism against technology-related dependence by attenuating the adverse effects of loneliness [51]. Such support can facilitate the development of stronger social connections in both online and offline contexts [52]. Within this framework, loneliness and AI chatbot dependence may directly influence one another, whereas perceived social support is positioned to modulate this relationship. To date, no empirical study has examined the mediating role of perceived social support in the relationship between loneliness and AI chatbot dependence among young adults, highlighting the present study’s contribution to the literature. Consistent with prior research, young adults with higher levels of loneliness exhibited greater levels of AI chatbot dependence [17, 53]. Loneliness is a prevalent experience among young adults. International meta-analyses and large-scale studies indicate that approximately 40% of young adults report experiencing loneliness frequently or very frequently [54]. In Türkiye, one study reported that around 25% of university students experience loneliness, whereas other samples have indicated rates as high as 54% [55]. In the present study, participants reported a mean loneliness score of 15.03 on a self-report scale with a maximum score of 24, reflecting above-average levels of loneliness. The prevalence of loneliness among young adults can be attributed to both individual and environmental factors. Major life transitions—including commencing university, leaving the family home, relocating to a new city, or entering the workforce—may disrupt or substantially alter existing social networks, resulting in weakened social relationships [56, 57]. Additionally, the absence of close and meaningful friendships, differences in communication skills, and challenges in adapting to social environments may diminish young adults’ sense of social connectedness, contributing to emotional loneliness [58]. Broader factors, such as socioeconomic status, cultural mismatch, life stress, and social comparison, may further exacerbate feelings of social disconnection [54]. The convergence of these multiple factors underscores that loneliness in young adulthood represents a common and complex psychosocial phenomenon. Consequently, individuals may increasingly turn to digital technologies to compensate for unmet social and emotional needs in offline contexts, in line with the CIUT. Although prior research suggests that AI-mediated interactions may temporarily alleviate feelings of loneliness under certain conditions [21, 59], this apparent benefit also presents a critical paradox. Interactions that initially serve as adaptive coping strategies may, over time, reinforce social withdrawal and partially substitute for real-world social interactions, thereby promoting addiction-like usage patterns [8, 60]. In this context, loneliness may function as a catalyst that triggers compensatory engagement with AI chatbots by providing short-term emotional relief [61, 62]; however, as such interactions intensify and acquire characteristics such as emotional dependence and the substitution of face-to-face interactions [18, 63, 64], they may paradoxically exacerbate loneliness over time, pointing to a self-reinforcing reciprocal feedback loop. Furthermore, loneliness and psychological distress have been identified as sequential mediators in digital addiction processes. For instance, a recent study examining the relationship between the “Joy of Missing Out” (JOMO) attitude and social media addiction demonstrated that loneliness and psychological distress functioned as serial mediators in the development of addictive behaviors [65]. This evidence suggests that loneliness can shape digital addiction behaviors through its impact on psychological distress, aligning conceptually with the present findings. Accordingly, the influence of loneliness on AI chatbot dependence should be conceptualized not as a simple unidirectional effect but as a complex, potentially bidirectional process encompassing both regulatory and risk-enhancing components. Another finding of the present study is the statistically significant negative association between loneliness and perceived social support. This indicates that as individuals’ perceptions of available social support decrease, experiences of loneliness increase, consistent with well-established evidence in the literature [26, 66]. Previous research has also examined the effects of different forms of social media engagement on loneliness. Findings suggest that passive use behaviors, such as content consumption or browsing, may exacerbate loneliness through social comparison, whereas active participation behaviors, including content creation and sharing, may attenuate this effect [67]. These observations provide a framework for anticipating that young adults with higher levels of loneliness may engage in specific patterns of digital platform use that elevate the risk of AI chatbot dependence. Accordingly, the relationship between loneliness and digital dependence can be conceptualized more concretely and mechanistically through varying forms of social media interaction. From a theoretical perspective, loneliness arises not only from the absence of social relationships but also from the perceived discrepancy between desired and actual levels of social connectedness [11]. Within this context, social support constitutes a critical psychological resource, providing emotional reassurance, informational guidance, and practical assistance in coping with stress and maintaining psychological well-being. From the perspective of the BH, this relationship is explained by the proposition that social connectedness represents a fundamental psychosocial need [68]; when this need is insufficiently fulfilled, the likelihood of experiencing loneliness increases [69]. A key contribution of the present study is the examination of this relationship during late adolescence and early adulthood—a developmentally sensitive period characterized by ongoing identity formation and increasingly complex interpersonal relationships. During this stage, lonely young adults may encounter difficulties in effectively utilizing their existing support networks or adequately perceiving the support available, potentially due to limited social skills or more introverted personality traits [70, 71]. Moreover, subjective perceptions of social support are particularly salient in technology-mediated contexts. Even when objective support networks are present, individuals with low perceived support may increasingly turn to digital platforms or AI-mediated communication channels to satisfy their emotional and social needs. One of the key findings of the present study is that perceived social support exerts a significant indirect mediating effect in the relationship between loneliness and AI chatbot dependence. This result suggests that loneliness should not be conceptualized as a unidimensional factor directly eliciting addictive tendencies; rather, it should be understood within a more complex socio-cognitive framework. Experiences of loneliness may diminish individuals’ perceptions of support from existing social networks, thereby increasing the likelihood of seeking alternative sources of interaction, such as AI chatbots, to satisfy unmet emotional and social needs [72]. This two-stage mechanism aligns closely with the core assumptions of the CIUT, which emphasizes the role of unmet psychosocial needs in the development of problematic technology use. Simultaneously, the present findings provide empirical support for the BH. High levels of perceived social support attenuate the negative psychosocial consequences of loneliness, enabling individuals to fulfill emotional needs through healthier interpersonal interactions. Under such conditions, reliance on compensatory digital tools, including AI chatbots, may be reduced [73]. Overall, these results contribute to the understanding of AI chatbot dependence by integrating the CIUT and BH frameworks within a single theoretical model. Whereas previous research has predominantly examined loneliness and social support in the context of internet or social media addiction, the present study demonstrates that these mechanisms are equally applicable to AI-based communication technologies. This integration extends existing theoretical models of digital addiction to encompass AI-mediated interaction environments. Limitations and future research Despite the meaningful contributions of the present study, several limitations should be considered when interpreting the findings. First, the cross-sectional design precludes definitive conclusions regarding the directionality or causality of relationships among the study variables; thus, the results should be interpreted as indicative of associations rather than causal effects. Second, all data were collected via self-report instruments, which may introduce potential response biases, including social desirability and recall bias, as well as the possibility of common method variance. Although procedural and statistical precautions were implemented to mitigate these risks, measurement-related biases cannot be entirely excluded. Third, the use of convenience sampling and voluntary online participation may have limited the representativeness of the sample, thereby constraining the generalizability of the findings. In addition, the analyses did not explicitly control for potentially relevant demographic or psychological variables, which could influence the observed relationships among loneliness, perceived social support, and AI chatbot dependence. Fourth, in the structural equation modeling analyses, the reliance on observed composite scores rather than latent constructs precluded the explicit modeling of measurement error. Although the employed instruments demonstrated high reliability, using observed variables limited the ability to account for measurement error fully. Finally, restricting the sample to young adults aged 18–30 residing in Türkiye limits the generalizability of the findings to other cultural or developmental contexts. This limitation is particularly relevant given that norms regarding social support, patterns of technology use, and experiences of loneliness may vary across cultures and age groups. Building on these limitations, future research should consider employing experimental or quasi-experimental designs to directly investigate the causal effects of interventions aimed at enhancing social support on AI chatbot dependence. Furthermore, extending the mediation framework to incorporate additional psychosocial variables-such as social anxiety, attachment styles, and digital literacy-would provide a more comprehensive understanding of the mechanisms underlying AI chatbot dependence. Integrating qualitative methodologies to examine user experiences in greater depth could further advance both the theoretical refinement and the practical applicability of research in this emerging field. Conclusions Overall, this study provides a theoretically grounded and empirically supported framework for understanding AI chatbot dependence. This study demonstrates a positive association between loneliness and AI chatbot dependence among young adults and identifies perceived social support as a significant mediating factor. The findings underscore the importance of interventions aimed at strengthening social connectedness to mitigate technology-related dependence. In this context, integrating critical digital literacy, self-control, and self-regulation skills into educational curricula, along with developing targeted awareness programs for specific user groups, is recommended. Furthermore, collaborative efforts among universities, local governments, public institutions, and mental health professionals are essential to design and implement programs that foster meaningful social relationships. Given the growing impact of AI-based applications on individuals’ emotional and social functioning, it is also crucial to develop ethical interfaces that limit anthropomorphism, excessive personalization, and reward-based design features, alongside establishing regulatory frameworks. Such a comprehensive approach can promote healthy, balanced, and safe engagement with AI technologies while reducing the risk of emotional dependence. Acknowledgements The authors would like to thank all participants for generously sharing their time and perspectives. Abbreviations Artificial Intelligence AI Chatbot Dependence Average Internet Use Analysis of Moment Structures Buffering Hypothesis Confidence Interval Compensatory Internet Use Theory Common Method Bias Comparative Fit Index Degrees of Freedom Incremental Fit Index Lower Limit Maximum Likelihood Normed Fit Index Perceived Social Support Root Mean Square Error of Approximation Structural Equation Modeling Standard Deviation Tucker–Lewis Index Upper Limit Loneliness Scale–6 Variance Inflation Factor Authors’ contributions ÖU: Conceptualization, Methodology, Formal analysis, Validation, Investigation, Data Curation, Writing – Original Draft, Writing – Review & Editing, Visualization, Supervision. SD: Conceptualization, Methodology, Writing – Original Draft, Writing – Review & Editing, Visualization, Project administration, Supervision. DG: Conceptualization, Methodology, Data Curation. Funding This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Data availability Due to ethical constraints and the need to protect participant confidentiality, the datasets generated and/or analyzed in this study are not publicly available but may be obtained from the corresponding author upon reasonable request. Declarations Ethics approval and consent to participate All procedures involving human participants were conducted in accordance with the ethical standards of the institutional and/or national research committees and with the Declaration of Helsinki and its later amendments or comparable ethical standards. Ethical approval for this study was obtained from the Ege University Faculty of Medicine Medical Research Ethics Committee (Approval date and number: October 16, 2025; 10.1T/58). All participants were fully informed about the purpose and procedures of the study and provided written electronic informed consent prior to participation. 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. Özkan Uğuz and Satı Doğan contributed equally to this work. References Associated Data Data Availability Statement Due to ethical constraints and the need to protect participant confidentiality, the datasets generated and/or analyzed in this study are not publicly available but may be obtained from the corresponding author upon reasonable request.