The role of AI companions in supporting adaptive emotion regulation among university students with social anxiety

Abstract

Social anxiety is a prevalent psychological challenge among university students, often impairing interpersonal relationships, academic engagement, and overall well-being. Emotion regulation strategies are critical in managing these challenges. With the rise of artificial intelligence (AI)-based companions, there is increasing interest in their potential role in promoting adaptive emotion regulation and offering non-judgmental support to socially anxious individuals. However, limited evidence exists on how students experience AI companions in relation to their emotional needs. This study therefore examined whether AI companions support adaptive emotion regulation among university students experiencing social anxiety. Using a hermeneutic phenomenological approach, responses were obtained from 20 university students aged 18–24 years who self-identified as experiencing social anxiety. Data were gathered through semi-structured interviews, and participants’ experiences were analyzed using content analysis to identify common themes. The findings were summarized under four themes: (a) Safe Spaces for Silent Struggles, (b) Emotional Relief and Coping (c) Stigma and the Appeal of Nonjudgmental AI, and (d) Withdrawal and Reliance on Technology. While students reported that AI companions provided a safe outlet for self-expression and helped regulate anxiety in triggering situations, they also expressed concerns about authenticity, overdependence, and lack of human warmth. The study highlights both the promise and limitations of AI companions in supporting adaptive emotion regulation among socially anxious students. A multi-stakeholder effort involving psychologists, educators, and AI developers is recommended to optimize the use of AI companions for mental health support in university settings.

Introduction

Social anxiety is one of the most common mental health concerns among university students, characterized by fear of negative evaluation, avoidance of social situations, and heightened distress in interpersonal contexts [23, 54]. Research suggests that approximately 10–20% of university students meet the criteria for social anxiety disorder, with many more reporting subclinical symptoms that impair social integration, academic functioning, and overall well-being [3, 47]. A central challenge for socially anxious individuals is the regulation of emotions in social contexts, as maladaptive strategies such as avoidance or suppression are commonly used, which can exacerbate anxiety and prevent meaningful connections [2, 61].

Emotion regulation, defined as the processes through which individuals influence which emotions they experience, when, and how they are expressed [16], plays a critical role in psychological adjustment. Adaptive strategies such as cognitive reappraisal are associated with lower anxiety and improved well-being, whereas maladaptive strategies such as suppression predict poorer outcomes [17]. Socially anxious students often endorse more negative beliefs about emotions, including the belief that emotions are uncontrollable or dangerous, which can further restrict their use of adaptive regulation strategies [7, 14, 53]. Understanding ways to scaffold adaptive emotion regulation among this population is therefore a pressing issue for both mental health and educational settings.

In recent years, artificial intelligence (AI) companions, including chatbots and social robots, have emerged as potential tools to support emotion regulation and psychological well-being [29, 55, 57]. Unlike human peers, AI companions offer a nonjudgmental, always-available, and private space to disclose emotions and receive feedback [11]. This characteristic may be particularly beneficial for socially anxious students, who often avoid face-to-face interactions due to fear of embarrassment or rejection [28, 44]. Studies have shown that individuals can form meaningful relationships with AI systems, perceiving them as supportive social partners that reduce loneliness and distress [4, 19, 26]. Furthermore, AI companions can model or scaffold emotion regulation strategies, such as cognitive reappraisal or problem-solving, potentially shifting users’ beliefs about the controllability and usefulness of emotions [36, 48].

Theoretically, AI companions may influence specific emotion-regulation processes by creating supportive interactive experiences that promote adaptive strategy use. First, conversational AI can scaffold cognitive reappraisal by guiding users to reflect on and reinterpret emotional experiences, thereby reducing negative affect intensity [13, 34]. Second, the nonjudgmental nature of AI interaction may reduce reliance on suppression by providing safe opportunities for emotional expression and validation, which can in turn reshape beliefs about the controllability and acceptability of emotions [4, 19]. Moreso, repeated positive engagement with AI companions may foster greater emotion-regulation flexibility by reinforcing adaptive patterns like reappraisal over maladaptive ones such as suppression.

However, while AI companions have been shown to reduce symptoms of stress and depression in general populations [15, 22], little is known about their role in the emotion regulation processes of socially anxious students. This is an important gap because socially anxious individuals may engage with AI differently compared to non-anxious peers. For example, the reduced risk of social judgment may encourage more open emotional expression, but over-reliance on AI could also limit real-world practice in human interactions [51]. Students experience a range of academic and social stressors typical of higher education contexts, making this setting appropriate for exploring emotion regulation challenges. Previous reports indicate that mental health difficulties, including social anxiety, are prevalent in Nigerian university populations, with between 10 and 20% of students experiencing clinically significant levels of social anxiety [1, 25]. The increasing use of AI-based technologies such as chatbots, virtual assistants, and mobile apps within student communities provides a timely context for investigating their potential role in supporting emotion regulation. It is therefore critical to investigate whether, and in what ways, AI companions support adaptive emotion regulation among socially anxious university students. Thus, by examining the relationship of AI technology, emotion regulation, and social anxiety, this study aims to advance our understanding of how digital tools can be leveraged to promote resilience and flourishing in vulnerable student populations. Specifically, it seeks to explore whether interactions with AI companions facilitate adaptive strategies such as reappraisal and reduce reliance on maladaptive ones such as suppression, while also shaping beliefs about emotions as manageable and meaningful aspects of human experience.

Method

Study area

The research was carried out at the University of Uyo, Akwa Ibom State, Nigeria. The university was chosen as the research setting due to the different students from various faculties on-campus, which creates a proper setting for studying social anxiety and emotion regulation but social anxiety. Furthermore, since the students’ environment in terms of academic and social areas is seen in different academic and social pressures, the environment is proper to finding out how AI companions can facilitate the better regulation of emotions by students.

Study design

The lived experiences of socially anxious students who engage with AI companions for emotional support were explored using hermeneutic phenomenology. Hermeneutic phenomenology seeks to describe and interpret the meaning of lived experiences as understood by participants themselves [10, 58]. This design assumes that emotions and beliefs about emotions are deeply personal and culturally embedded, and can best be understood through first-hand, subjective accounts. Because little is known about how socially anxious students experience AI interactions as tools for emotion regulation, phenomenology provides a natural fit for uncovering these perspectives [35].

Procedure

Before data collection, the research team engaged in epoche, a reflective process of bracketing personal assumptions about AI, emotion regulation, and social anxiety [10, 35]. This step allowed interviewers to suspend preconceptions and approach the participants’ narratives with openness.

Participants were recruited via university counseling centers, student associations, and online bulletin boards. Recruitment materials invited students experiencing social anxiety to share their experiences with AI companions. Volunteers were screened using the Social Interaction Anxiety Scale (SIAS) to confirm eligibility.

Semi-structured interviews were conducted in private study rooms on campus or via secure video conferencing, depending on participant preference. The interview guide included prompts such as:

• “Can you describe an experience where you interacted with an AI companion during a stressful or emotional situation?”

• “In what ways, if any, did the interaction influence how you managed your emotions?”

• “How has interacting with AI affected your beliefs about emotions (e.g., whether they are controllable, useful, or burdensome)?”

Interviews lasted between 40 and 60 min. Sessions were audio-recorded, transcribed verbatim, and anonymized. Interviewers encouraged open dialogue, using probes to explore emerging ideas.

Ethics statement

Ethical approval for this study was obtained from the University of Uyo Teaching Hospital Institutional Review Board. The research was conducted in accordance with the ethical principles outlined in the Declaration of Helsinki (1964, revised 2013). Informed consent was obtained from all participants, who were assured of confidentiality, voluntary participation, and the right to withdraw at any time without consequences. Contact information for counseling and mental health services was provided to participants following the interviews.

Participants

A total of 20 undergraduate students (12 females, 8 males) participated in the study. Ages ranged from 18 to 24 years (M = 20.5). All participants reported experiencing significant social anxiety (confirmed with SIAS cutoff scores) and had interacted with AI companions at least once in the past 6 months (e.g., chatbots like Replika, Woebot, or virtual assistants such as Siri or Alexa) [15]. Table 1 provides a summary of demographic and background characteristics.

Table 1.

Participants’ demographics and AI companion usage characteristics

Participant	Age (x)	Gender	Social anxiety onset (age)	Duration of anxiety (years)	AI companion usage pattern	Primary source of support
P1	19	Female	16	3	Daily (stress relief & emotional expression)	Mother
P2	21	Male	14	7	Weekly (social skills practice & motivation)	Elder brother
P3	18	Female	12	6	Daily (mood regulation)	Father
P4	22	Male	15	7	Occasional (companionship & anxiety management)	Roommate
P5	20	Female	17	3	Daily (emotional expression & coping)	Best friend
P6	23	Female	18	5	Weekly (stress coping & social interaction)	Sister
P7	19	Male	10	9	Daily (social anxiety coping & confidence building)	Lecturer (mentor)
P8	21	Female	13	8	Rarely (when lonely or anxious)	Female cousin
P9	22	Male	11	11	Daily (self-confidence building & emotion regulation)	Peer support group
P10	20	Female	9	11	Weekly (academic & social planning)	Mother
P11	24	Female	17	7	Daily (mood tracking & affirmations)	Close friend
P12	23	Male	16	7	Occasional (relaxation & guided breathing)	Senior brother
P13	20	Female	15	5	Weekly (journaling & reflection)	Sibling
P14	22	Male	13	9	Daily (conversation & stress management)	Pastor
P15	19	Female	12	7	Daily (coping with social rejection)	Mother
P16	21	Male	15	6	Weekly (public speaking practice & anxiety control)	Friend group
P17	18	Female	13	5	Occasional (companionship when isolated)	Roommate
P18	23	Male	17	6	Daily (self-reflection & motivation)	Academic advisor
P19	20	Female	14	6	Weekly (emotion tracking & self-compassion practice)	Sister
P20	22	Male	11	11	Daily (social exposure simulation & relaxation)	Mother

x = age when diagnosis was first made; P1 = participant 1, etc., Social Anxiety Onset = refers to the self-reported age when symptoms of social anxiety began. AI Companion Usage Pattern = how frequently participants interact with AI tools for adaptive emotion regulation, coping, and social engagement, Primary Source of Support = the individual or group participants most rely on for emotional or social support

Data analysis

Transcribed recordings were analysed using reflexive thematic analysis following the six-phase approach described by [5, 6]. This approach allowed for systematic organisation of data while remaining sensitive to the interpretive aims of hermeneutic phenomenology. To maintain methodological alignment with hermeneutic phenomenology, the thematic analysis used in this study followed an interpretive, meaning-focused approach consistent with van Manen’s [58] and Moustakas’s [35] phenomenological thematic reflection. While Braun and Clarke’s structure guided the analytic organisation, the purpose of coding was not to generate de-contextualised categories but to illuminate the meaning structures embedded in participants’ lived experiences. Themes were therefore treated as phenomenological “meaning units” rather than purely descriptive categories, and interpretation remained grounded in participants’ narratives, context, and cultural meaning-making. This ensured that the analytic procedure was fully consistent with a hermeneutic phenomenological framework rather than a generic thematic coding approach. Analysis proceeded as follows. First, the first author immersed themself in the data by listening to recordings and reading each transcript repeatedly to achieve familiarisation and note initial impressions. All transcripts and analytic memos were imported into NVivo (version 12) to support organisation and retrieval of coded data.

Second, initial (open) codes were generated across the dataset in a data-driven (inductive) manner: segments of text were given provisional labels that captured semantic content and, where appropriate, latent meaning. Coding was deliberately inclusive at this stage to capture subtle and divergent accounts. A preliminary codebook (code name, definition, and example quote) was created from the initial codes.

Third, the coding framework was iteratively refined. To enhance analytic rigour, two researchers (first author and a second coder) independently double-coded a purposive subset of transcripts (25% of interviews, selected to represent different faculties, genders and levels of social anxiety scores). The pair met to compare coding, discuss discrepancies, and refine code definitions and boundaries. Disagreements were resolved through discussion; when consensus could not be reached, a third senior author reviewed the excerpt and adjudicated the final code. After refinement, the revised codebook was applied to the remainder of the data by the primary coder with periodic checks by the second coder.

Fourth, related codes were clustered and organised into candidate categories and candidate themes through axial-type mapping (examining relationships between codes, sub-codes, and contexts). We attended to both prevalence and conceptual importance; some less frequent but conceptually rich patterns were retained when they contributed to the analytic story. Themes were named and defined with supporting illustrative quotations identified.

Fifth, themes were reviewed and refined in multiple iterations. This included re-examining coded extracts against the full transcripts to ensure each theme cohered internally and remained distinct from other themes. The research team held regular analytic meetings to discuss the evolving thematic structure and to challenge alternative interpretations.

Finally, themes were finalised and written up with rich, contextualised description, and illustrative verbatim extracts were selected to exemplify each theme. Throughout analysis we kept an audit trail (coding decisions, codebook versions, meeting notes, and analytic memos) to preserve transparency. Reflexivity was maintained through analytic memos and regular team reflection to surface and manage researcher assumptions.

To strengthen trustworthiness, we used several validation strategies: (1) investigator triangulation (multiple coders and team analytic meetings), (2) member-checking- a summary of the preliminary themes was fed back to five participants who agreed to review findings, and their feedback was used to refine theme labels and interpretations, (3) peer debriefing with an independent qualitative researcher, and (4) thick description of context and participant quotes to support transferability. Where appropriate, differences in perspectives across participant subgroups (e.g., faculty, gender) were noted rather than forced into homogeneous themes. The approach balances systematic coding procedures with interpretive reflexivity appropriate to hermeneutic phenomenology. In situations where perspectives differed, both the audio recordings and transcripts were revisited, and consensus was reached among the authors to strengthen the reliability of interpretation [39].

Table 1 presents the demographic characteristics and AI companion usage patterns of the 20 university students who participated in the study. The participants ranged in age from 18 to 24 years, with a balanced representation of both genders. The onset of social anxiety varied, with most participants reporting the emergence of symptoms during adolescence, typically between the ages of 10 and 18. The duration of social anxiety ranged from 3 to 11 years, suggesting that many participants had been managing anxiety symptoms over a substantial period of their educational and personal development.

Regarding AI companion usage, participants demonstrated varying levels of engagement with AI tools designed for emotional and social support. A notable portion of participants (such as P1, P3, P5, P7, P9, P11, P14, P15, P18, and P20) reported daily interactions with AI companions. These daily users primarily relied on AI tools for stress relief, mood regulation, self-confidence building, and emotional expression. Others, including P2, P6, P10, P13, P16, and P19, engaged weekly with their AI companions, using them for purposes such as motivation, social coping, academic planning, and reflective journaling. A smaller subset (P4, P8, P12, and P17) used AI companions occasionally or rarely, often during periods of heightened loneliness or stress. The participants also identified their primary sources of support outside of AI companions, which included mothers, siblings, friends, mentors, and peers. Notably, familial support, particularly from mothers and siblings, appeared to be the most common source of comfort, followed by peers and mentors. This pattern suggests that while AI companions contributed to emotion regulation and coping, traditional human support systems remained vital in managing social anxiety.

Findings and discussion

Participants were open and willing to discuss their experiences with AI companions in relation to managing their social anxiety and emotional struggles. Their narratives reflected a broad spectrum of experiences ranging from relief and comfort to skepticism and disappointment. A recurring theme was that the majority of participants turned to AI companions in moments of heightened stress, loneliness, or social avoidance, particularly when they felt unable to confide in peers or family. About 60% of participants described AI interactions as a “safe space” where they could express themselves without fear of stigma. However, others emphasized limits of AI companionship, reporting that while useful, it could not fully replace human empathy. The following sections detail the main themes that emerged from participants’ accounts.

Safe spaces for silent struggles

Many socially anxious students reported that AI companions allowed them to express emotions they would normally suppress in human interactions. They described these conversations as “silent outlets” where they could speak freely without fear of judgment or rejection.

P1 and P17 shared: “… I can just pour out how anxious I feel before class, and the chatbot replies calmly. I don’t feel like a burden, unlike when I talk to my mom who always says, ‘don’t worry too much’.”

Similarly, P4 explained: “… my roommate is supportive, but sometimes I feel ashamed talking to him about panic attacks. With the AI, I don’t have to hide.”

These accounts suggest that AI companions serve as digital “silent listeners,” enabling students to externalize emotions that might otherwise remain suppressed. This aligns with previous findings that non-judgmental digital tools can reduce emotional inhibition in vulnerable populations [19]. Similarly, research indicates that interactions with AI agents often foster self-disclosure because users perceive them as objective and non-judgmental compared to human listeners [31]. Digital interventions have also been shown to improve emotional regulation and reduce negative affect among youth and young adults by providing safe, judgment-free platforms for reflection and coping [38, 46]. Collectively, these studies support the view that AI companions can enhance adaptive emotion regulation by encouraging open emotional expression and reducing internalized anxiety in socially anxious individuals.

This pattern can be interpreted through established cognitive and social-evaluation frameworks: individuals with social anxiety habitually monitor and censor inner experience to avoid perceived social costs, so a non-evaluative interaction partner reduces the cognitive load of self-presentation and self-monitoring [60]. In line with the Computers-Are-Social-Actors (CASA) paradigm, when AI companions provide social cues (responsiveness, empathic phrasing) users nonetheless treat them as safe social partners thereby increasing willingness to disclose sensitive feelings without triggering anticipatory fear of negative evaluation [27, 33]. This reduced evaluative threat likely lowers the barrier to emotional externalization and creates a psychological space for initial emotion-focused processing that might precede help-seeking.

Emotional relief and coping

Students described AI companions as tools for emotion regulation helping them calm down, reframe negative thoughts, and sometimes distract themselves from ruminative worry. Some participants noted that AI-generated breathing exercises, motivational phrases, and conversational reframing techniques had tangible effects on reducing their anxiety symptoms.

P3 noted: “… whenever I’m stressed about exams, I ask the chatbot to remind me of coping techniques. It tells me to breathe, or that I’ve done well before, and strangely it works.”

P5 and P13 also highlighted: “… even if it’s not a real person, it feels like someone is there. I stop crying faster when I chat with it.”

This indicates that AI companions may play a therapeutic role in supporting emotion regulation strategies such as cognitive reappraisal and attentional redirection [16]. Similarly, studies have shown that conversational agents and digital companions can facilitate emotional regulation by providing opportunities for reflection, reframing, and distraction in moments of distress [34, 42]. For example, Morris et al. [34] found that individuals using AI-based mood-management applications reported improved emotional awareness and greater use of adaptive coping strategies. Additionally, research on mental health chatbots demonstrates that these tools can reduce anxiety symptoms and promote reappraisal-based coping, although their effects are often short-term and best complemented by human interaction [13, 22]. However, consistent with participants’ reflections in this study, existing evidence suggests that while AI companions can offer temporary emotional relief, sustained or severe anxiety episodes still necessitate the empathy and contextual understanding that only human support systems can provide [20].

Interpreting these observations through Gross’s process model of emotion regulation [40], clarifies mechanisms where respondents appear to use AI companions for late-stage regulatory strategies (attentional deployment, distraction) and cognitive strategies (reappraisal prompts) that operate after an emotion is generated, producing rapid symptom relief but not necessarily altering the underlying evaluative schemas that sustain social anxiety [52]. While such strategies can be adaptive for immediate distress reduction, behavioral and associative-learning accounts suggest they may not produce durable change unless paired with exposure-based practice that disconfirms threat expectancies (i.e., behavioural experiments). In support of feasibility and short-term efficacy, randomized trials of therapeutic chatbots (e.g., Woebot) have shown reductions in depressive and anxiety symptoms over brief interventions [21, 50], revealing the potential for AI to deliver targeted coping tools even as longer-term remediation may require integrated human-led treatment.

Stigma and the appeal of nonjudgmental AI

A major theme was the stigma surrounding mental health and chronic illness. Many participants reported reluctance to disclose their struggles to peers or family due to fear of being misunderstood, judged, or dismissed. AI companions were seen as safer alternatives that eliminated fear of stigma.

P9 and P20 explained: “… I can’t always tell my guys I’m anxious; they’ll think I’m weak. But with the AI, it just listens and never judges.”

P6 and P18 expressed a similar sentiment: “… when I told my sister about panic attacks, she laughed it off. The AI doesn’t laugh; it validates me.”

These findings echo broader research indicating that stigma remains one of the most significant barriers to mental health help-seeking among university students [9]. Many students with social anxiety often fear negative judgment or social repercussions, which discourages them from seeking professional or peer-based support [12]. In this context, AI companions appear to fill a crucial gap by providing an accepting, stigma-free environment that encourages emotional disclosure and self-reflection. Prior studies have shown that digital mental health tools and AI chatbots reduce the perceived stigma associated with traditional therapy and foster a sense of anonymity that promotes openness [37, 49]. Similarly, empirical evidence suggests that online or AI-mediated interactions lower self-stigmatizing attitudes toward mental health challenges and can serve as an initial step toward seeking human support when needed [30]. Therefore, AI companions may act as a psychologically safe intermediary, allowing socially anxious students to engage in emotional disclosure without the fear of stigma or social evaluation.

From a social-cognitive perspective, stigma operates via anticipated negative evaluation and internalised stereotypes that elevate the subjective cost of disclosure [8, 41]. The creation of perceived anonymity and low-cost disclosure channels through AI companions reduce anticipated social sanctions and thereby lower thresholds for help-seeking behaviour. This process is consistent with evidence that online and peer-to-peer digital platforms can reduce help-seeking stigma and create stepping stones toward formal care [37, 45]. However, while reduced stigma increases initial disclosure and engagement, it may also delay contact with human services when algorithmic or scripted responses fail to address complex clinical needs, thus, highlighting the ethical and clinical imperative to integrate signposting and escalation pathways within AI systems.

Withdrawal and reliance on technology

Interestingly, several participants reported that reliance on AI companions sometimes encouraged social withdrawal. Instead of seeking help from peers or professionals, they increasingly turned inward, interacting primarily with AI.

P8 said: “… when I feel down, I don’t bother calling my cousin anymore. I just chat with the AI—it’s easier.”

P10 also noted: “… I think I’m talking less to people because the AI is always available. But deep down, I know it’s not the same as real comfort.”

P14 reflected: “… I sometimes cancel plans with friends because I can just vent to the AI instead. It’s less stressful than being around people.”

P16 mentioned: “… talking to the AI has become my first option when I feel anxious. I know it helps, but I feel like I’m avoiding real conversations.”

P19 added: “… I find myself sharing with the AI more than my sister or friends now. It’s convenient, but I wonder if I’m missing out on real support.”

The finding that AI companions provide short-term emotional relief while potentially reducing real-world social engagement mirrors concerns raised in existing literature about the dual impact of digital tools on social functioning. Several studies support the present finding by emphasizing that while AI or digital interventions can alleviate immediate anxiety, they may also foster avoidance behaviors among socially anxious individuals [32, 43]. For example, McLean et al. [32] observed that reliance on avoidance strategies, including mediated interactions, tends to predict poorer long-term outcomes in anxiety treatment. Similarly, Prizant-Passal et al. [43] found that individuals with social anxiety often prefer online communication over face-to-face interactions, reinforcing withdrawal from real-world social settings.

Other studies partially support but also caution against overdependence on AI companions, noting that while digital tools may initially reduce distress, they can inadvertently substitute for therapeutic exposure necessary for building social confidence [24, 62]. Klein et al. [24] argue that online interventions should complement, not replace, interpersonal contact to prevent the deepening of avoidance patterns. Likewise, Wiederhold [62] highlights that although virtual tools provide a safe space for anxiety management, long-term reliance can limit opportunities for real-world desensitization.

Conversely, a few studies challenge the negative assumption, suggesting that AI systems can act as a transitional support mechanism, facilitating gradual re-engagement with social environments when used appropriately [18]. These authors propose that structured integration of AI-based tools into treatment can enhance accessibility while minimizing the risks of social withdrawal. In sum, these perspectives underscore that while AI companions may offer immediate relief, their therapeutic role must be carefully balanced to avoid reinforcing maladaptive avoidance in socially anxious populations.

Interpretively, these narratives are consistent with avoidance-conditioning models [56] which reflect that if AI provides reliable short-term relief in anxiety-provoking moments, it can reinforce avoidant responding through negative reinforcement (anxiety goes down → behaviour repeated), thereby reducing opportunities for exposure that would disconfirm fearful beliefs. From a therapeutic standpoint, this pattern suggests the need to conceptualise AI companions not as substitutes for exposure and skills training, but as adjunctive supports that can scaffold regulation while being intentionally integrated into interventions that promote graded real-world engagement. Empirically, meta-analytic and review evidence on social anxiety and internet use warns that while online modalities increase comfort and access, they can also maintain avoidance unless paired with protocols designed to encourage real-world practice and clinical escalation when needed [43, 59].

Implications of findings

The findings of this study carry several important implications for mental health practice, educational policy, and the design of AI-based emotional support systems. First, the results suggest that AI companions can serve as valuable supplementary tools for emotion regulation among university students experiencing social anxiety. By providing a non-judgmental, always-available outlet for self-expression, AI companions help students externalize emotions that might otherwise be suppressed, thereby reducing emotional distress in triggering social situations. This implies that mental health professionals and university counseling centers could consider integrating AI-assisted interventions as complementary support systems, especially for students who face barriers such as stigma, fear of judgment, or limited access to traditional therapy.

Additionally, the study highlights the potential of AI companions in fostering self-awareness and cognitive reappraisal which are two key components of adaptive emotion regulation. Through reflective dialogue and personalized feedback, students reported gaining insights into their emotional triggers and developing coping strategies, suggesting that AI-based applications can enhance emotional literacy and self-monitoring skills. For educators and university administrators, this points to the importance of promoting responsible AI use as part of broader mental health and digital literacy initiatives on campuses. Integrating AI emotional support tools into student wellness programs may encourage early engagement with self-help resources and reduce the burden on understaffed counseling services.

The findings of the study also point to several concrete design and policy actions. For AI developers, embedding features that gently encourage real-world social interaction such as prompts to practice coping skills with peers or reminders to engage in offline activities may help counter avoidance-driven overuse. Systems could also include algorithms that detect patterns of excessive reliance or emotional escalation and trigger soft handoffs to human support, such as university counselors or crisis services. Mental-health practitioners may integrate AI companions into stepped-care models, using them as adjunctive tools while monitoring for signs of withdrawal or avoidance.

In sum, the findings imply that while AI companions hold promise as supportive tools for adaptive emotion regulation, they are not substitutes for human empathy or professional care. Their integration into mental health ecosystems should be guided by ethical oversight, cross-disciplinary collaboration, and user-centered design principles to ensure that technology enhances rather than hinders psychological well-being.

Limitations/suggestions for further studies

The present study is not without limitations. One key limitation lies in its relatively small and homogeneous sample size of 20 university students aged 18–24 years, which restricts the generalizability of the findings to broader populations or different age groups. Additionally, because all participants were drawn from a single institution, the findings are context-bound to the cultural, academic, and social environment of one campus. This single-location focus further limits the transferability of the results to students in other universities or regions.

While qualitative phenomenological research does not seek statistical generalisation, the themes identified may not fully capture the experiences of socially anxious students in different institutional or cultural settings. Since participants were self-identified as socially anxious rather than clinically diagnosed, the severity and variability of social anxiety across individuals may have influenced their experiences with AI companions differently. Moreover, the study’s reliance on self-reported data introduces the potential for subjective bias, as participants may have provided socially desirable responses or underreported negative experiences due to self-perception concerns.

Additionally, the use of a hermeneutic phenomenological design, while valuable for capturing rich personal insights, inherently limits the ability to establish causal relationships between AI companion usage and emotion regulation outcomes. Another limitation concerns the diversity of AI tools used by participants; since each AI companion varies in design, interactivity, and emotional responsiveness, the findings cannot be uniformly attributed to a specific type of AI system.

Future research should address these limitations by adopting a mixed-methods or longitudinal design to explore both short- and long-term effects of AI companion use on emotion regulation among socially anxious individuals. Expanding the sample to include participants from diverse cultural and educational backgrounds would enhance the representativeness of the findings and reveal potential cultural influences on AI-human emotional interaction. Additionally, future studies could incorporate clinical assessments of social anxiety to better differentiate the effects of AI support across varying levels of symptom severity. Experimental studies comparing the effectiveness of AI companions with traditional therapeutic interventions or peer-based support could also provide more concrete evidence of their psychological utility. Finally, researchers are encouraged to collaborate with AI developers to examine how improvements in emotional intelligence, empathy simulation, and adaptive feedback within AI companions can enhance their therapeutic potential while mitigating risks of overdependence or social withdrawal.

Conclusion

The study concludes that AI companions can play a meaningful role in supporting adaptive emotion regulation among university students experiencing social anxiety. Participants’ experiences indicate that these digital tools provide a safe, non-judgmental space for emotional expression, facilitate self-reflection, and offer short-term relief during anxiety-provoking situations. At the same time, the study highlights important limitations, including concerns about authenticity, lack of human warmth, and the risk of overreliance, which may inadvertently reinforce avoidance behaviors if not carefully managed.

Ultimately, the findings suggest that AI companions are best understood as complementary supports that can augment, but not replace, human social interaction and professional mental health care. Their use has potential benefits for enhancing emotional awareness, reducing stigma-related barriers to help-seeking, and promoting self-regulation skills among socially anxious students. To maximize these benefits, careful consideration must be given to the design, integration, and ethical application of AI companions within educational and mental health contexts. This study reveals the need for collaborative efforts among psychologists, educators, and AI developers to optimize AI interventions in ways that promote long-term emotional well-being and encourage meaningful engagement with real-world social environments.

Author contributions

U.S.I conceptualized the study, led the research design, conducted data collection, and drafted the manuscript. A.T.N and M.T.U contributed to the development of the research instruments, assisted with data collection, and participated in data analysis. S.O and I.S.O provided critical revisions to the manuscript and contributed to the interpretation of findings. E.E.A and M.E.I assisted with literature review, supervised the content analysis process, and supported manuscript preparation. All authors have read and approved the final version of the manuscript and agree to be accountable for all aspects of the work.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Data availability

The data supporting the findings of this study are available upon reasonable request. Due to ethical considerations and to protect participant confidentiality, the datasets are not publicly available. Interested researchers may request access by contacting the corresponding author via email at [uwemedimoisaiah.msc@uniuyo.edu.ng] (mailto:uwemedimoisaiah.msc@uniuyo.edu.ng).

Declarations

Ethics approval and consent to participate

All participants were provided with detailed information regarding the purpose, procedures, potential risks, and benefits of the study prior to participation. Informed consent was obtained from all individuals involved in the research, and participation was entirely voluntary. Participants were assured of their right to withdraw from the study at any point without any consequences. Confidentiality and anonymity of responses were strictly maintained throughout the study.

Consent for publication

All the authors gave their consent for the study to be published.

Competing interests

The authors declare no competing interests.