An Interpretive Description Study of Clients’ Experiences of Integrating Artificial Intelligence-Based Mental Health Support With Face-to-Face Counseling

Introduction

Digital technologies have become increasingly embedded in the delivery, extension, and organization of psychological and counseling services, contributing to a broader shift in how individuals seek, receive, and sustain mental health support. Tele-mental health, mobile applications, online counseling, virtual environments, conversational agents, and artificial intelligence (AI)-enabled systems have expanded the range of options available to people who may experience geographical, temporal, financial, interpersonal, or psychological barriers to conventional services. Digital approaches have already demonstrated value in supporting continuity of care in contexts such as cancer-related supportive care, where remote modalities can extend access beyond traditional clinical settings (Soumerai et al., 2024), while technology-mediated counseling has also been used in educational contexts to enhance the accessibility and delivery of counseling services (Muhammad, 2024). Online group counseling has similarly shown that therapeutic and counseling processes can be meaningfully transferred into digital formats when intervention structure and participant needs are carefully considered (Zammitti et al., 2023). At the same time, accessibility remains a central concern in digital health because technological availability does not necessarily translate into equitable, usable, or psychologically acceptable services for all individuals (Koukopoulos & Danias, 2024). These developments have created a mental health landscape in which face-to-face counseling increasingly coexists with digital resources rather than functioning as the sole environment in which clients seek information, emotional support, psychological reflection, and behavioral guidance.

Within this broader digital transformation, AI has introduced a qualitatively different form of interaction because it enables systems not only to deliver pre-programmed content but also to respond dynamically to users’ language, questions, emotional concerns, and perceived needs. AI-driven mental health interventions now encompass a wide spectrum of applications, including screening, monitoring, psychoeducation, emotional support, prevention, clinical education, and conversational assistance (Ni & Jia, 2025). AI-enabled applications are also being designed around human-computer interaction principles to provide more integrated and personalized forms of mental health support (W. Zhang et al., 2026). Beyond conversational systems, AI has been explored for psychological assessment through approaches such as automated facial analysis, illustrating the growing role of algorithmic technologies in interpreting mental and emotional states (Mamcarz et al., 2025). Other applications include AI-supported trauma intervention (Y. J. Zhang et al., 2026), augmented-reality systems for remote psychological counseling (Bao et al., 2024), and interactive or therapeutic robots intended to support psychological and psychiatric care (Akyazı & BaŞTemur, 2024). Chatbots and virtual assistants have also been investigated in more specific behavioral health contexts, such as tobacco cessation counseling, where conversational systems may provide repeated guidance, reinforcement, and accessible support outside conventional clinical encounters (Deepika et al., 2025). Taken together, these developments indicate that AI is increasingly moving from an experimental technological adjunct toward a potentially routine component of the wider mental health support ecosystem.

The appeal of AI-based mental health support is closely related to several characteristics that differ from conventional counseling. AI systems can be available at any time, respond immediately, allow repeated questioning, and provide users with an opportunity to articulate distress without the immediate social consequences of speaking to another person. Such characteristics may be especially relevant for individuals who experience embarrassment, uncertainty, fear of judgment, or difficulty initiating help-seeking. Research on digital mental health has shown that attitudes toward online therapy, mental health applications, and chatbots are shaped by perceived usefulness, accessibility, trust, and willingness to engage with technology (Gbollie et al., 2023). More recent work examining university students’ perceptions of digital mental health services likewise suggests that intentions to use online therapy and mental health applications depend on how individuals evaluate the practicality, acceptability, and credibility of these options (Al-Mahrouqi et al., 2026). Acceptance of AI-based and digital mental health interventions has therefore become an important research focus, with emerging models specifically addressing the factors that influence willingness to adopt AI-enabled mental health technologies (Békés et al., 2025). Among counseling professionals themselves, acceptance and awareness may also vary according to professional and demographic characteristics, highlighting that successful AI integration depends not only on clients but also on the attitudes of practitioners and service systems (Ofem et al., 2025). Similarly, computer self-efficacy has been identified as an important consideration in attitudes toward tele-mental health among counselors, indicating that technological competence can influence readiness to engage with digitally mediated services (Lee et al., 2024).

For clients, however, technological acceptance is only one component of the experience. The psychologically distinctive feature of conversational AI is its capacity to simulate a responsive interpersonal exchange. Users can describe problems, receive reformulations, ask for clarification, request alternative perspectives, rehearse conversations, and engage in extended exchanges that may resemble some surface characteristics of counseling dialogue. This resemblance raises important questions regarding perceived empathy, alliance, trust, and therapeutic credibility. Research examining digital therapeutic alliances has shown that perceptions of relational quality can influence how helpful individuals find online mental health interactions (Huang et al., 2025). At the same time, perceptions of AI-generated mental health responses can change when users become aware of whether the response was produced by a human or an AI system, suggesting that judgments of empathy, usefulness, and authenticity are partly shaped by beliefs about the source of support (Jain et al., 2024). Consumers may also evaluate human, AI, and hybrid therapists through different psychological pathways when anticipating satisfaction with mental health services (Haderlie et al., 2026). These findings are particularly important because conversational fluency can create a subjective impression of understanding even when the system lacks the interpersonal history, embodied presence, ethical responsibility, and clinical judgment associated with a human counselor.

The relationship between AI and therapeutic processes therefore cannot be adequately understood solely by asking whether AI responses are accurate or whether users are satisfied with them. Psychotherapy and counseling depend on more than information delivery. Common therapeutic factors include alliance, trust, responsiveness, emotional engagement, validation, collaborative meaning-making, and the experience of being understood within an evolving interpersonal relationship. Discussions of AI-based psychotherapy have consequently emphasized the need to consider how common psychotherapeutic factors may or may not be reproduced in AI-mediated interaction (Giotakos, 2025). Systematic evaluation of text-based conversational agents has also revealed substantial variation in the metrics, methods, and contexts used to determine whether such agents are effective, safe, engaging, or clinically meaningful (Gong et al., 2026). These methodological variations make it difficult to reduce the value of AI mental health support to a single indicator such as symptom change or user satisfaction. In addition, reviews of AI chatbots for psychological health among health professionals show both potential usefulness and unresolved concerns regarding implementation, evidence quality, privacy, safety, and appropriateness (Baek et al., 2025). Generative AI in broader clinical practice similarly offers opportunities for efficiency, decision support, education, and communication while raising challenges related to reliability, ethical use, privacy, accountability, and professional oversight (Han et al., 2026).

These tensions become especially significant when AI is not used as a stand-alone intervention but alongside an existing therapeutic relationship. In real-world practice, clients may move between face-to-face sessions and AI interactions, using each form of support for different purposes. They may consult AI after a difficult counseling session, ask it to explain a psychological concept, use it to organize thoughts before the next appointment, rehearse a disclosure, generate coping strategies, or seek immediate conversational support during periods when the counselor is unavailable. Digital mental health resources may also influence help-seeking itself by helping individuals interpret their difficulties, evaluate whether professional support is needed, and prepare for contact with formal services (Ma et al., 2026). Among university students, a shift from traditional campus counseling toward generative-AI-based consultation has already been identified, suggesting that some users may increasingly position AI as an alternative or preliminary source of psychological support (Wang et al., 2026). This shift creates both opportunities and risks. AI may lower barriers to engagement and extend therapeutic reflection beyond scheduled appointments, but its convenience may also encourage users to delay professional consultation, avoid difficult interpersonal disclosures, or assign excessive authority to automated responses. The possibility of adverse effects when users become dependent on AI support or when access to such support is interrupted has also been conceptualized through emerging concerns about AI discontinuation effects in mental health applications (Kelly et al., 2026).

The role of perceived anonymity and reduced interpersonal exposure may be particularly important in explaining why clients choose to discuss certain experiences with AI before, or instead of, discussing them with a counselor. Research on anonymity in technologically mediated mental health environments suggests that reduced identifiability can alter users’ willingness to disclose and their experience of psychological safety (Akdag & Wahl, 2024). This may be relevant for topics involving shame, stigma, relationship conflict, sexuality, trauma, body image, self-evaluation, or other concerns that clients find difficult to express in person. Digital environments themselves can influence psychological well-being and self-perception, as illustrated by research on social media, body image, self-esteem, cultural expectations, and mental health (Merino et al., 2024). Broader understandings of well-being also emphasize that psychological health may involve meaning, values, existential concerns, and dimensions such as spiritual intelligence and subjective well-being that cannot always be reduced to symptom management (Stiliya & Pandey, 2022). Consequently, a client’s experience of AI support may depend not only on whether the system provides technically appropriate advice but also on whether it can accommodate complex personal meanings, relational histories, cultural assumptions, and values that become important in face-to-face counseling.

The need to understand integration rather than replacement is further supported by developments across specialized areas of psychological and behavioral health care. Informatics-based psychotherapeutic and psychiatric interventions have been applied in fields such as dermatology, where mental health and physical health concerns intersect and where digital interventions may supplement existing care pathways (Lamarre et al., 2026). AI and virtual counseling technologies are similarly being discussed as components of contemporary counseling practice rather than as isolated innovations (Ningrum et al., 2024). Online mental health communities are increasingly evaluated in terms of service quality, user expectations, and the characteristics required to make digital mental health experiences responsive and satisfactory (Yang & Li, 2026). These trends indicate that future mental health systems are likely to be hybrid, with users moving among face-to-face, telehealth, app-based, AI-mediated, community-based, and self-guided forms of support. In such systems, the central question is not simply whether AI should or should not be used, but how clients determine the role it should play in relation to professional care and how that role changes over time.

This issue is particularly salient in counseling because therapeutic work often extends psychologically beyond the counseling room. Clients continue to reflect on conversations, encounter emotionally activating situations, attempt therapeutic exercises, make interpersonal decisions, and reconsider their understanding of themselves between appointments. Digital tools may occupy this between-session space by providing continuity, reminders, reflection prompts, or immediate opportunities to verbalize distress. Studies of technologically mediated couples counseling illustrate how counseling relationships and practices are being reshaped within digital societies (Zhang & Tan, 2026). At the same time, the increasing sophistication of AI raises the possibility that clients may begin to compare the responsiveness of AI with that of human counselors, selectively disclose information to one but not the other, or use AI-generated interpretations to question or reinforce therapeutic formulations. Such comparisons may affect trust, therapeutic alliance, expectations of immediacy, and perceptions of what constitutes adequate psychological support. The resulting integration is therefore likely to be dynamic and negotiated rather than technically predetermined.

Despite the rapid expansion of AI mental health research, much of the available literature has focused on effectiveness, usability, technological acceptance, intervention design, service quality, or the performance of specific systems. These lines of research are essential, but they provide only a partial understanding of how AI becomes incorporated into the lived process of counseling. Clients may simultaneously experience AI as helpful and limited, private and uncertain, empowering and potentially dependency-forming. They may use it to facilitate face-to-face disclosure in one situation and to avoid such disclosure in another. AI may reinforce counseling recommendations, introduce conflicting interpretations, or alter expectations regarding the pace and responsiveness of therapeutic communication. What matters clinically is therefore not only the characteristics of the technology but the meaning clients assign to it within an existing therapeutic relationship. An interpretive qualitative approach is especially suited to examining these processes because it can move beyond cataloguing positive and negative attitudes toward identifying patterns in how clients evaluate AI, establish boundaries around its use, negotiate conflicting sources of psychological guidance, and determine what remains uniquely important about human counseling.

Interpretive Description offers a particularly appropriate methodological framework for this purpose because it is oriented toward generating clinically meaningful understandings of complex experiential phenomena rather than producing purely abstract theory or descriptive summaries. By examining the practical and relational meanings clients attach to AI use, such an approach can illuminate when AI is experienced as a bridge between sessions, when it becomes a rehearsal space for difficult disclosure, when it supports autonomy and self-reflection, and when it risks fragmenting care or displacing necessary interpersonal work. Understanding these distinctions is increasingly important as AI-based mental health tools become more accessible and as both clients and counselors must make decisions about how such systems should be positioned within routine psychological care.

Accordingly, the aim of this study was to explore and interpret clients’ experiences of integrating AI-based mental health support with face-to-face counseling in England.

Methods and Materials

Study Design and Participants

This study employed a qualitative research design based on Interpretive Description to explore clients’ experiences of integrating artificial intelligence (AI)-based mental health support with conventional face-to-face counseling. Interpretive Description was selected because the purpose of the study was not merely to describe participants’ experiences but to generate clinically meaningful interpretations of how clients perceived, negotiated, and incorporated AI-based mental health tools within an ongoing therapeutic relationship. The approach was particularly appropriate for examining the complex interaction between digital mental health support and interpersonal counseling, including the perceived benefits, limitations, tensions, and changes in therapeutic engagement associated with using both forms of support. The study was conducted in England and included 24 adult clients who had experience of receiving face-to-face psychological counseling while concurrently using at least one AI-based mental health support application or conversational system. Participants were recruited from counseling and psychological support services located in England using purposive sampling. Maximum variation principles were considered during recruitment to obtain a range of experiences in terms of age, gender, reasons for seeking counseling, duration of counseling, frequency of AI use, and type of AI-based mental health support used. Eligibility criteria included being 18 years of age or older, residing in England, having attended at least four face-to-face counseling sessions during the previous 12 months, having independently used an AI-based mental health support tool for at least four weeks during the same period, being able to describe experiences of integrating the two forms of support, and having sufficient English proficiency to participate in an in-depth interview. AI-based mental health support was broadly defined as the use of an artificial intelligence-enabled conversational agent, chatbot, digital mental health assistant, or other interactive AI system for purposes such as emotional support, self-reflection, mood management, coping guidance, psychoeducation, or preparation for counseling sessions. Individuals whose use of AI was limited exclusively to administrative functions, such as appointment scheduling, were excluded. Recruitment continued alongside data analysis until sufficient information power and conceptual depth had been achieved and additional interviews were no longer making a meaningful contribution to the developing interpretive account. The final sample of 24 participants was therefore considered adequate for addressing the study aim while preserving the depth and contextual sensitivity required by Interpretive Description. Before participation, all individuals received information regarding the study purpose, voluntary participation, confidentiality, data management, and their right to withdraw without affecting access to counseling services. Written informed consent was obtained from all participants before data collection.

Measures

Data were collected using a brief demographic and AI-use information form followed by individual semi-structured, in-depth interviews. The demographic form was designed to provide contextual information relevant to interpretation of participants’ experiences and included age, gender, employment status, previous experience with psychological counseling, approximate duration and frequency of current or recent face-to-face counseling, type and duration of AI-based mental health support use, frequency of interaction with the AI system, and the primary purposes for which AI support was used. These contextual data were used descriptively rather than as variables for statistical comparison. The principal data collection instrument was a semi-structured interview guide developed specifically for the study in accordance with the research aim and the applied orientation of Interpretive Description. The interview guide consisted primarily of open-ended questions and flexible prompts intended to encourage participants to provide detailed accounts of how AI became incorporated into their mental health support practices. Interviews explored participants’ initial motivations and expectations regarding AI-based support, circumstances in which they preferred to use AI rather than communicate with their counselor, perceived similarities and differences between AI-based interactions and face-to-face counseling, ways in which AI influenced preparation for counseling sessions, disclosure of sensitive experiences, emotional processing, coping between sessions, implementation of therapeutic strategies, and perceptions of the continuity of psychological support. Participants were also encouraged to describe circumstances in which AI advice conflicted with, reinforced, supplemented, or altered what occurred in counseling. Particular attention was given to perceptions of trust, emotional safety, privacy, anonymity, accessibility, immediacy, personalization, empathy, judgment, dependence, and the perceived credibility of AI-generated responses. Interviews additionally examined whether participants discussed their AI use with their counselors, how counselors responded when such use was disclosed, and how the integration of AI affected the therapeutic alliance and participants’ sense of autonomy in managing their mental health.

The interview guide was used flexibly so that questions could be reordered, expanded, or followed by additional probes according to each participant’s experience. Examples of probing areas included asking participants to describe a specific occasion when AI support was particularly useful or unhelpful, how they decided whether to trust an AI-generated suggestion, what they did when AI guidance appeared inconsistent with counseling, and whether their relationship with their counselor changed after incorporating AI into their mental health support. Participants were encouraged to provide concrete examples rather than general evaluations whenever possible. Interviews were conducted individually in a private setting or through a secure online videoconferencing platform, according to participant preference. Each interview lasted approximately 50 to 80 minutes. With participants’ permission, all interviews were digitally audio-recorded and subsequently transcribed verbatim. The interviewer also prepared field notes immediately after each interview to record contextual observations, preliminary interpretations, notable emotional responses, emerging concepts, and issues requiring exploration in subsequent interviews. Consistent with the iterative nature of Interpretive Description, data collection and analysis occurred concurrently. Consequently, the interview guide was progressively refined as conceptual patterns developed, allowing later interviews to investigate emerging issues in greater depth. Reflexive notes were also maintained throughout data collection to document the researchers’ assumptions, interpretive decisions, and potential influences on the research process.

Data analysis

Data were analyzed using the principles of Interpretive Description through an iterative, inductive, and comparative analytic process. Analysis began after the first interview and proceeded concurrently with subsequent data collection so that emerging insights could inform later interviews and areas requiring greater conceptual clarification could be explored. The researchers initially became immersed in the data by listening to the audio recordings, reviewing transcripts in their entirety, reading field notes, and repeatedly engaging with participants’ accounts before formal coding. Rather than fragmenting the data immediately into isolated units, each transcript was first considered as a complete account in order to retain the contextual meaning of the participant’s experience and to identify the overall pattern through which AI-based support was situated within face-to-face counseling. During repeated readings, segments of text relevant to the research question were identified and assigned preliminary codes representing experiences, perceptions, decisions, emotional responses, behaviors, expectations, and perceived consequences of combining AI support with counseling. Coding remained deliberately broad and provisional during the early phase of analysis to avoid imposing predetermined theoretical categories on participants’ experiences.

As analysis progressed, codes were constantly compared within and across participants to identify similarities, variations, contradictions, exceptional cases, and relationships among experiences. Particular attention was given to differences between situations in which AI was perceived as complementing counseling and situations in which it was experienced as competing with, disrupting, or complicating the therapeutic process. Codes sharing conceptual similarities were progressively grouped into broader patterns and categories. The analytic process moved beyond simple description by asking what the emerging patterns suggested about how clients used AI within their broader mental health care, why particular forms of integration appeared helpful or problematic, under what circumstances participants relied more heavily on AI or on their counselor, and what consequences these choices had for disclosure, self-management, therapeutic engagement, and perceived continuity of care. Categories were repeatedly examined against the complete transcripts to ensure that interpretations remained grounded in participants’ accounts and did not become detached from the clinical and personal contexts in which experiences occurred. Alternative explanations and negative cases were actively considered, particularly when participants reported experiences that challenged dominant patterns within the dataset.

Through successive cycles of comparison, condensation, and conceptual development, preliminary categories were refined into higher-order themes that represented the most clinically relevant dimensions of participants’ experiences. Relationships among themes were examined to develop an integrated interpretation of how AI-based mental health support functioned alongside face-to-face counseling rather than treating the two forms of support as independent experiences. Analytic memos were written throughout this process to document conceptual developments, comparisons between cases, questions raised by contradictory data, and decisions concerning the boundaries and meanings of emerging themes. A reflexive record was maintained to examine how the researchers’ professional assumptions regarding counseling, digital mental health technologies, and artificial intelligence might influence interpretation. To enhance analytic credibility, emerging codes, categories, and themes were reviewed and discussed among members of the research team, with disagreements resolved through return to the original transcripts and discussion of alternative interpretations rather than through numerical coding agreement. Particular emphasis was placed on interpretive coherence, clinical relevance, and representation of variation across participants. Analysis continued until the final thematic structure provided a sufficiently rich and coherent explanation of participants’ experiences and additional data no longer substantially altered the developing interpretation. The final findings were therefore constructed as an interpretive account of the patterns, variations, and practical meanings through which clients integrated AI-based mental health support with face-to-face counseling in their everyday management of psychological difficulties.

Findings and Results

The final sample consisted of 24 clients living in England who had experience of using AI-based mental health support alongside face-to-face counseling. Participants ranged in age from 21 to 56 years, with a mean age of 34.8 years. Fifteen participants identified as women, eight as men, and one as non-binary. Fourteen participants were employed full-time, four were employed part-time, three were university students, and three were not in paid employment at the time of the interviews. Participants had engaged in face-to-face counseling for periods ranging from approximately 2 to 14 months, with most having attended counseling for at least 4 months before the interview. Seventeen participants generally attended counseling weekly, while seven attended approximately every two weeks. Participants reported seeking counseling for a range of concerns, and several described more than one reason for entering therapy. Anxiety and persistent worry were the most commonly reported concerns, followed by low mood or depressive symptoms, occupational or academic stress, relationship difficulties, trauma-related distress, low self-esteem, and difficulties with emotional regulation. The participants had used AI-based mental health support for between 2 and 12 months. Thirteen primarily used general-purpose conversational AI systems for mental health-related reflection and support, seven mainly used dedicated AI-enabled mental health chatbots, and four reported regularly using both types of systems. Nine participants described using AI-based support almost every day, ten used it several times per week, and five used it less regularly, usually in response to specific emotional difficulties or counseling-related needs. Fifteen participants had discussed their AI use with their counselor, whereas nine had not disclosed it or had mentioned it only indirectly. Across the interviews, AI was not generally described as an independent replacement for therapy. Instead, participants portrayed it as a flexible, immediately available resource that occupied the psychological and temporal space between formal counseling encounters, although the meaning and consequences of this integration varied considerably.

Analysis generated six interrelated themes that captured how participants experienced the integration of AI-based mental health support with face-to-face counseling. These themes were: AI as an in-between-session holding space; AI as a rehearsal and translation space for counseling; greater control and reduced interpersonal exposure; recognition of the emotional and contextual limits of AI; active negotiation of trust, accuracy, and safety; and reshaping rather than replacing the therapeutic relationship. The themes did not represent mutually exclusive categories. Most participants contributed to several themes, and individual experiences often contained both positive and negative dimensions. Participant counts are therefore presented to indicate the distribution of experiences across the dataset rather than to assign quantitative importance to particular themes.

Table 1 demonstrates that participants did not experience AI-based mental health support as a single or uniform intervention. Instead, AI acquired meaning through its relationship with the timing, emotional demands, and interpersonal characteristics of face-to-face counseling. The most widely represented theme concerned recognition of the emotional and contextual limits of AI, reported by 22 participants, indicating that even participants who described substantial benefits remained aware of important differences between machine-generated support and human counseling. Twenty-one participants described AI as an in-between-session holding space, particularly when difficult emotions emerged outside scheduled counseling hours. This function was repeatedly framed as one of accessibility rather than equivalence: participants valued being able to externalize thoughts immediately, but they generally distinguished this from the deeper relational work undertaken with a counselor. Nineteen participants described using AI as a rehearsal or translation space, suggesting that one of the most clinically meaningful contributions of AI was its capacity to help clients convert vague, confusing, or embarrassing experiences into material that could later be discussed in counseling. The theme of greater control and reduced interpersonal exposure was especially evident among participants who found immediate interpersonal disclosure difficult. Being able to type, edit, pause, reconsider, or discontinue a conversation reduced the perceived social consequences of disclosure and sometimes allowed participants to approach material they had previously avoided. However, this sense of safety coexisted with uncertainty. Twenty participants described actively evaluating AI responses, comparing them with previous counseling advice, testing whether suggestions felt appropriate, or disregarding responses that appeared overly confident, generic, or inconsistent with their experience. Overall, the thematic structure suggested that participants’ integration of AI was characterized by a continuous balancing process: the accessibility, privacy, and cognitive usefulness of AI were valued, while relational depth, contextual understanding, professional accountability, and emotional attunement remained strongly associated with face-to-face counseling.

A prominent finding concerned the way AI was used at different points surrounding the counseling session itself. Participants rarely described AI use as temporally random. Instead, the function of AI changed according to whether it was used before counseling, immediately after counseling, during the interval between sessions, or during episodes of heightened distress. Before sessions, AI was commonly used to organize thoughts, identify topics worth raising, and rehearse difficult statements. After sessions, it was frequently used to review what had happened, clarify therapeutic concepts, or continue reflection after the limited counseling hour had ended. Between sessions, participants used it to maintain awareness of goals and coping strategies. During emotionally intense moments, its most valued feature was immediate availability, although participants simultaneously acknowledged that such use could become problematic if AI was relied upon in situations requiring human or professional intervention.

As shown in Table 2, the most common use of AI occurred between face-to-face sessions, with 21 participants describing some form of ongoing reflection, coping support, or maintenance of therapeutic focus during this interval. This finding suggested that AI was particularly valued for addressing a structural limitation of conventional counseling: therapeutic contact occurred at scheduled times, whereas emotional difficulties and the need to process events occurred unpredictably. Eighteen participants also used AI to support implementation of strategies discussed during counseling. Rather than simply asking AI for new advice, these participants frequently asked it to explain, simplify, personalize, or break down therapeutic tasks that had already emerged within counseling. Seventeen participants used AI before sessions, and several explained that entering counseling with more clearly formulated thoughts allowed them to use limited session time more effectively. This preparatory function appeared particularly important for participants who became anxious when speaking spontaneously or who reported that they often forgot significant experiences once they were seated with the counselor. Fifteen participants used AI after counseling sessions to continue processing material that had felt unfinished, emotionally intense, or difficult to retain. However, the data also indicated an important boundary around accessibility. Sixteen participants had interacted with AI during periods of heightened distress, and although immediate availability was frequently described as calming, participants were more cautious about this use than about routine reflective use. Several described learning that conversational availability could create a misleading sense that adequate support was being received, even when their needs exceeded what they believed an AI system could safely address. The temporal analysis therefore indicated that the perceived value of AI was greatest when it extended, organized, or supported therapeutic work, whereas uncertainty increased when AI moved closer to functions requiring clinical judgment, relational containment, or crisis-sensitive decision making.

Variation across participants also revealed distinct patterns in the way AI and counseling were combined. These patterns were not treated as fixed participant types because some clients moved between them over time. Instead, they represented dominant modes of integration evident in participants’ accounts. The most common pattern was complementary integration, in which AI was clearly positioned as a support around counseling rather than an alternative to it. A second pattern involved selective parallel use, where participants maintained a clearer separation between what they discussed with AI and what they discussed with the counselor. A smaller group described periods of substitutive drift, in which the convenience of AI led them to postpone or reduce interpersonal disclosure. Finally, some participants experienced conflict between AI-generated guidance and the counselor’s perspective, requiring them to decide which source they trusted and sometimes to repair confusion within subsequent counseling sessions.

Table 3 shows that complementary integration was the dominant pattern, characterizing 10 participants whose accounts consistently positioned AI as an adjunct to counseling. In these cases, AI was most useful when it supported activities surrounding the therapeutic relationship rather than attempting to reproduce it. Participants in this group frequently carried AI-assisted reflections into sessions, asked counselors to help evaluate ideas generated by AI, or used AI to revisit counseling material during the week. Seven participants primarily demonstrated selective parallel use. They valued both forms of support but maintained relatively distinct boundaries between them. For example, some reserved emotionally significant or historically complex issues for their counselor while using AI for everyday stress, decision making, or cognitive organization. Others did the reverse initially, discussing highly embarrassing experiences with AI before deciding whether they were prepared to disclose them to a counselor. Four participants demonstrated periods of substitutive drift. Their accounts were particularly important because they showed that the same qualities that made AI appealing—immediacy, convenience, absence of visible judgment, and unlimited conversational availability—could also facilitate avoidance. These participants sometimes chose AI because it was easier than raising difficult topics in therapy, thereby reducing opportunities to work directly with discomfort, vulnerability, and interpersonal trust. Three participants primarily described conflict-and-repair experiences in which contradictory suggestions from AI and the counselor created confusion. Importantly, these conflicts did not necessarily produce disengagement. In some cases, bringing the disagreement into counseling generated productive discussion about assumptions, alternative interpretations, and the limitations of generic advice. Taken together, these patterns suggested that the consequences of AI use depended less on the presence or absence of AI itself than on the position it occupied within the client’s broader therapeutic ecology. Integration appeared most constructive when AI remained transparent, critically evaluated, and connected to rather than insulated from face-to-face therapeutic work.

The interpretive model represented in Figure 1 conceptualizes integration as a dynamic, client-led process rather than a simple combination of two treatment modalities. Participants’ accounts suggested that the process commonly began with an immediate psychological or practical need, such as distress occurring outside counseling hours, difficulty identifying emotions, uncertainty about what to discuss in the next session, a desire to rehearse disclosure, or the need to make sense of a recent counseling conversation. AI was then accessed because of its availability, privacy, conversational responsiveness, and low interpersonal demands. The initial response generated by AI was not automatically accepted. Instead, participants engaged in an appraisal process in which they evaluated whether the response felt emotionally relevant, logically plausible, consistent with their previous experience, and compatible with what they had learned in counseling. When AI output was perceived as useful, it could facilitate emotional labeling, reflection, behavioral planning, preparation for disclosure, or continuation of therapeutic work between sessions. When it was perceived as generic, inaccurate, overly reassuring, contradictory, or insufficiently sensitive to context, participants either modified their questions, rejected the advice, sought external information, or returned to the counselor for clarification. The model therefore places verification and judgment at the center of integration. It further indicates that the pathway from AI interaction back into face-to-face counseling was influenced by several conditions, particularly the client’s level of trust in the counselor, comfort discussing AI use, severity and complexity of the issue, perceived privacy of the AI platform, and the degree to which the counselor was experienced as open rather than dismissive toward technology-mediated support. Constructive integration was most evident when AI helped clients formulate experiences that were subsequently explored in counseling, when it reinforced rather than displaced individualized therapeutic work, and when clients maintained awareness that conversational fluency did not necessarily indicate clinical understanding. Less constructive integration occurred when the convenience of AI encouraged avoidance of difficult relational work, when participants treated AI-generated statements as authoritative without critical evaluation, or when important AI-mediated experiences remained concealed from the counselor. The central interpretive finding was therefore that AI functioned neither simply as a therapeutic substitute nor merely as a technological tool. For participants, it became an intermediary psychological space located between private self-reflection and interpersonal therapy. Its value depended on how clients negotiated accessibility against relational depth, anonymity against accountability, immediate responsiveness against contextual understanding, and autonomy against the need for professional and human support.

Across all themes, the findings indicated that participants evaluated AI and face-to-face counseling according to different but overlapping criteria. AI was most strongly valued for immediacy, availability, privacy, repeatability, and the ability to interact without concern about tiring, disappointing, or being judged by another person. Face-to-face counseling, by contrast, was valued for emotional attunement, continuity of personal knowledge, professional accountability, sensitivity to non-verbal communication, capacity to challenge avoidance, and the experience of being recognized by another human being. Participants generally did not expect AI to fully reproduce these relational characteristics. Instead, difficulties arose when the conversational style of AI created expectations of understanding that exceeded what participants later felt the system could actually provide. Several participants explained that an AI response could appear highly empathic at a linguistic level while still failing to grasp the significance of a particular relationship, history, trauma, or pattern that their counselor had come to understand over time. This distinction between the appearance of empathic responsiveness and the experience of being personally understood formed an important boundary in participants’ accounts.

The findings further suggested that AI sometimes altered the threshold for disclosure. Participants who struggled with embarrassment, shame, fear of judgment, or uncertainty about whether an experience was “important enough” to discuss often used AI as an initial disclosure environment. Putting the experience into words appeared to make subsequent disclosure to the counselor more manageable. In this sense, AI occasionally functioned as a transitional communicative space between private thought and interpersonal expression. However, the same mechanism could operate in the opposite direction. When disclosure to AI provided sufficient short-term emotional relief, a minority of participants felt less urgency to raise the matter in counseling. The issue could therefore remain outside the therapeutic relationship despite having clinical relevance. This finding illustrates an important tension within the integrated model: reducing the immediate emotional cost of disclosure could either facilitate deeper therapeutic engagement or become a means of avoiding it.

Participants also described changes in their sense of agency. Many valued the opportunity to engage with psychological ideas independently rather than waiting passively for the next counseling appointment. They could ask repeated questions, request alternative explanations, seek examples, or explore the same issue from several perspectives without being constrained by session time. This contributed to a sense of active involvement in their own psychological care. Nevertheless, agency was strongest when accompanied by critical evaluation. Participants who developed clearer boundaries around what AI could and could not provide described feeling more confident in deciding which issues could be explored independently and which required discussion with a counselor. By contrast, uncertainty was more evident when participants relied heavily on the apparent certainty or fluency of AI responses. Thus, the data suggested that client autonomy in AI-supported mental health care was not simply a product of access to information; it depended on the capacity to evaluate, contextualize, and appropriately position that information within a broader therapeutic process.

Finally, disclosure of AI use to counselors emerged as an important contextual factor shaping the quality of integration. Participants who had openly discussed their use of AI generally described greater continuity between digital reflection and counseling. In these cases, AI-generated ideas could be examined collaboratively, inaccurate assumptions could be corrected, and useful reflections could be incorporated into therapeutic work. Participants who had not disclosed their AI use gave several reasons, including believing it was irrelevant, fearing that the counselor might disapprove, feeling embarrassed about needing support between sessions, or worrying that the counselor would interpret AI use as a lack of confidence in therapy. For some, this created two partially disconnected therapeutic narratives: one developed in dialogue with AI and another presented to the counselor. The findings therefore indicated that the integration of AI-based mental health support with face-to-face counseling was shaped not only by technological characteristics but also by openness within the therapeutic relationship. Participants’ experiences were most coherent when the two sources of support could be critically connected, while secrecy, substitution, or unexamined trust increased the likelihood of fragmentation. Overall, clients portrayed AI as potentially valuable when it occupied a clearly bounded and reflective role around counseling, but they consistently identified human relational engagement as the element that gave deeper context, accountability, and personal meaning to the therapeutic process.

Discussion

The present study explored clients’ experiences of integrating AI-based mental health support with face-to-face counseling and identified a pattern of use that was neither wholly substitutive nor wholly separate from conventional therapy. Participants generally positioned AI as an adjunctive resource that occupied the temporal and psychological space around counseling sessions. Six interrelated themes were identified: AI as an in-between-session holding space; AI as a rehearsal and translation space for counseling; greater control and reduced interpersonal exposure; recognition of the emotional and contextual limits of AI; active negotiation of trust, accuracy, and safety; and reshaping rather than replacing the therapeutic relationship. The findings further showed that AI was used differently before, after, and between counseling sessions and that participants followed several distinct integration patterns, including complementary integration, selective parallel use, substitutive drift, and conflict-and-repair. Taken together, these findings suggest that the clinical significance of AI cannot be understood solely in terms of whether clients use it, but rather in terms of how it is positioned within the broader therapeutic process and how clients evaluate its role relative to professional human support.

A central finding was that AI frequently functioned as an in-between-session holding space. Participants valued the immediate availability of AI at times when their counselor was unavailable, particularly when emotional distress, uncertainty, or the need for reflection arose outside scheduled appointments. This finding is consistent with the broader expansion of digital mental health services, which has been driven partly by the capacity of digital systems to overcome temporal and accessibility constraints associated with conventional service delivery (Koukopoulos & Danias, 2024). Similar benefits of continuity and extended access have been reported in telehealth-based supportive care, where remote support can complement rather than necessarily replace existing clinical relationships (Soumerai et al., 2024). Digital mental health resources have also been shown to influence help-seeking by allowing individuals to engage with psychological concerns before or between formal encounters with professionals (Ma et al., 2026). The present findings extend this literature by suggesting that clients may experience AI not simply as another information source, but as a psychologically meaningful transitional environment in which immediate distress can be externalized until it can be revisited in counseling. This may explain why participants emphasized continuity more strongly than substitution. The value of AI was often derived from its ability to fill intervals in care while the counselor remained the principal source of deeper therapeutic interpretation.

The finding that AI served as a rehearsal and translation space was also particularly important. Participants described using AI to identify emotions, organize complex experiences, formulate questions, and rehearse disclosures that they later intended to bring into counseling. This suggests that conversational AI can function as a pre-therapeutic or para-therapeutic linguistic scaffold, helping clients transform diffuse emotional experiences into words that can be shared in a human therapeutic relationship. Research on digital mental health has already demonstrated that technology can enhance counseling-related processes by improving access to psychological information and supporting structured engagement with personal concerns (Muhammad, 2024). Studies of online counseling and digitally mediated support likewise indicate that people can engage in meaningful reflective processes outside face-to-face environments (Zammitti et al., 2023). More specifically, the ability of digital mental health resources to encourage help-seeking suggests that technology may reduce some of the cognitive and emotional barriers that precede professional engagement (Ma et al., 2026). The present study adds an important relational dimension to these findings. AI was not merely used to decide whether help was needed; it also helped some clients prepare themselves psychologically and linguistically to make better use of help they were already receiving.

The theme of greater control and reduced interpersonal exposure further clarifies why participants found AI useful for sensitive topics. Many reported that communicating with AI involved less fear of judgment because there was no visible human reaction, no concern about burdening another person, and no requirement to respond immediately to another individual’s emotional presence. These experiences are consistent with work examining anonymity in technology-mediated mental health settings, which suggests that reduced identifiability can affect disclosure, perceived safety, and willingness to engage with sensitive concerns (Akdag & Wahl, 2024). Similar issues are relevant across digital environments in which self-presentation, self-evaluation, and psychological vulnerability are shaped by perceived social exposure (Merino et al., 2024). In the present study, however, reduced interpersonal exposure had a dual effect. For some clients, it lowered the threshold for eventual disclosure in counseling; for others, it risked becoming a way to avoid difficult interpersonal work. This duality is clinically significant because the absence of judgment can be both facilitative and avoidant. AI may create a low-risk entry point into difficult material, but the therapeutic value of that material may depend on whether it ultimately becomes connected to the relational processes of counseling.

Participants’ strong recognition of the emotional and contextual limits of AI was one of the most widely represented findings. Even those who used AI frequently distinguished between conversational responsiveness and genuine relational understanding. They described AI responses as sometimes repetitive, generic, overly reassuring, or insufficiently attuned to personal history. This finding aligns with current debates about AI-based psychotherapy, especially the question of whether common therapeutic factors such as alliance, empathy, responsiveness, and collaborative meaning-making can be meaningfully reproduced by artificial systems (Giotakos, 2025). Evidence concerning digital therapeutic alliance indicates that perceived relational quality can influence how helpful users find online mental health interactions (Huang et al., 2025); however, source awareness also affects judgments of mental health responses, with perceptions changing when individuals know whether content was produced by a human or AI (Jain et al., 2024). Research comparing expected satisfaction with human, AI, and hybrid therapists similarly shows that users evaluate these modes through different psychological pathways (Haderlie et al., 2026). The current findings suggest that clients can experience AI as emotionally useful without equating it with a counselor. They may appreciate immediate validation and structured reflection while still assigning unique value to a human therapist’s continuity of knowledge, capacity to perceive non-verbal cues, ethical accountability, and ability to understand experiences within a longer personal and relational history.

This distinction is also supported by the broader literature on the evaluation of conversational agents. Text-based mental health systems are currently assessed through heterogeneous metrics and usage contexts, and there remains no single standard by which relational or therapeutic quality can be judged (Gong et al., 2026). Scoping work on AI chatbots has likewise identified promise alongside unresolved concerns related to implementation, safety, privacy, and the adequacy of evidence (Baek et al., 2025). More generally, reviews of generative AI in clinical practice have emphasized that algorithmic efficiency and responsiveness must be balanced against reliability, accountability, ethical responsibility, and context sensitivity (Han et al., 2026). These concerns were reflected directly in participants’ accounts. They frequently differentiated between responses that sounded appropriate and responses that they considered trustworthy enough to act upon. This indicates that conversational fluency itself should not be treated as equivalent to therapeutic competence.

The theme of active negotiation of trust, accuracy, and safety further demonstrates that participants were not passive recipients of AI-generated guidance. Most described strategies for evaluating responses, including comparing AI suggestions with previous counseling discussions, rejecting advice that felt inconsistent, rephrasing questions, or returning to the counselor for clarification. This finding resonates with emerging research on acceptance of AI-based mental health interventions, which shows that uptake depends on perceived usefulness, trust, credibility, and broader expectations about digital care (Békés et al., 2025). Studies of university students’ intentions to use digital mental health services similarly indicate that acceptance depends on more than simple technological availability (Al-Mahrouqi et al., 2026; Gbollie et al., 2023). Professional acceptance is also relevant. Counselors’ and practitioners’ awareness, confidence, and attitudes toward AI may shape whether clients feel able to disclose their use of these technologies (Lee et al., 2024; Ofem et al., 2025). In the present study, openness between client and counselor appeared to facilitate safer integration because AI-generated ideas could be examined, corrected, or contextualized within therapy. By contrast, nondisclosure created the possibility of two partially disconnected therapeutic narratives.

The findings on complementary integration, selective parallel use, substitutive drift, and conflict-and-repair provide a more differentiated understanding of how AI enters counseling. Complementary integration was the most common pattern and occurred when AI was used to prepare for, reinforce, or extend counseling without displacing the counselor’s role. This pattern is consistent with the emerging concept of hybrid mental health care, in which digital and human services operate as interconnected rather than mutually exclusive modalities. AI-enabled systems are increasingly being developed for multiple functions across screening, support, monitoring, prevention, and education (Ni & Jia, 2025), while design-oriented work has emphasized the potential for human-computer interaction to support holistic mental health applications (W. Zhang et al., 2026). AI-assisted trauma intervention has likewise been proposed as a way of extending psychological support within broader systems of social assistance (Y. J. Zhang et al., 2026). These developments support the interpretation that the greatest value of AI may lie in augmentation rather than replacement.

Selective parallel use reflected a different form of integration. Some participants deliberately assigned different psychological functions to AI and counseling, using AI for everyday stress, decision making, or cognitive organization while reserving emotionally complex or historically rooted material for face-to-face therapy. This division is understandable given the broad range of specialized applications now being developed, including therapeutic robots (Akyazı & BaŞTemur, 2024), augmented-reality counseling systems (Bao et al., 2024), behavioral health chatbots (Deepika et al., 2025), and informatics-based psychotherapeutic interventions in specialized clinical settings (Lamarre et al., 2026). Such technological diversity encourages users to conceptualize digital systems as task-specific resources rather than comprehensive therapeutic substitutes. However, the present findings indicate that parallel use may also fragment care if significant AI-mediated experiences remain unknown to the counselor.

The pattern of substitutive drift deserves particular attention. A minority of participants described situations in which AI became easier to use than counseling precisely because it demanded less emotional exposure. In these cases, convenience, constant availability, and reduced interpersonal discomfort could lead clients to postpone bringing difficult topics into therapy. A similar shift has been observed among university students moving from campus counselors toward generative-AI-based consultation (Wang et al., 2026). While such shifts may reflect dissatisfaction with access barriers or a desire for immediacy, they also raise concerns about over-reliance. The emerging concept of AI discontinuation effects further suggests that dependence on AI-based mental health support may have consequences when access is interrupted or when the user has come to rely heavily on the system for emotional regulation (Kelly et al., 2026). The present study therefore suggests that convenience should not automatically be interpreted as therapeutic benefit. A system may reduce immediate distress while simultaneously reducing opportunities for relational work that would otherwise occur in counseling.

Conflict-and-repair represented another clinically meaningful pattern. Some participants encountered AI-generated interpretations that contradicted their counselor’s perspective, leading to uncertainty about which source to trust. Yet these conflicts were not always detrimental. When brought back into counseling, they could stimulate clarification, critical reflection, and more explicit discussion of therapeutic assumptions. This finding can be understood in light of research on online mental health service quality, which emphasizes the importance of responsiveness, credibility, and alignment with user expectations (Yang & Li, 2026). It is also relevant to the finding that perceived source matters in the evaluation of AI-generated mental health content (Jain et al., 2024). Rather than treating disagreement between AI and counselors as inherently problematic, the present findings suggest that such disagreement may become therapeutically useful when it is transparent and processed collaboratively.

The study also contributes to a broader understanding of digital transformation in counseling practice. Digital societies are already reshaping how counseling is delivered and experienced, including in areas such as couples counseling (Zhang & Tan, 2026) and virtual counseling services (Ningrum et al., 2024). As AI systems become more sophisticated, clients may increasingly expect mental health support to be immediate, interactive, personalized, and continuously available. Such expectations may influence how face-to-face counseling itself is evaluated. At the same time, psychological well-being cannot be reduced to instantaneous problem-solving or information delivery. Human well-being includes relational, existential, cultural, and meaning-oriented dimensions, some of which are reflected in research linking broader forms of intelligence and subjective well-being (Stiliya & Pandey, 2022). The current findings reinforce the importance of retaining a broad conception of therapeutic value. Participants repeatedly emphasized that what distinguished counseling was not merely better information but being known, challenged, understood, and held accountable within a continuing relationship.

Conclusion

Overall, the present findings support a model in which AI functions as an intermediary psychological space between private self-reflection and interpersonal therapy. Its benefits were most apparent when it enhanced preparation, continuity, reflection, and implementation of therapeutic work. Its limitations became more prominent when clients required contextual understanding, complex clinical judgment, relational challenge, or emotionally nuanced interpretation. The resulting integration was shaped by clients’ trust in AI, trust in counselors, comfort discussing technology use, severity of the issue, and willingness to critically evaluate automated guidance. Accordingly, the clinical question is not whether AI should be regarded as inherently helpful or harmful, but under what conditions it supports rather than disrupts therapeutic processes. The findings suggest that AI is most constructive when it is critically evaluated, transparently incorporated, and clearly bounded within a broader human-centered counseling relationship.

Limitations & Suggestions

The present study has several limitations. First, the sample consisted of 24 clients living in England who had experience with both face-to-face counseling and AI-based mental health support, and the findings therefore may not reflect the experiences of clients in other countries, cultural contexts, health systems, or forms of psychotherapy. The purposive qualitative design was appropriate for generating depth but was not intended to produce statistically generalizable conclusions. Participants also differed in the type, frequency, and duration of AI use, and these differences may have shaped their experiences in ways that could not be fully separated analytically. The study relied on retrospective self-report, meaning that participants’ accounts may have been influenced by memory, current attitudes toward AI, or their relationship with their counselor at the time of interview. In addition, the study focused on clients’ perspectives and did not directly examine counselors’ interpretations of the same integration processes. Finally, AI technologies evolve rapidly, and participants may have interacted with systems that differ substantially in functionality, safeguards, and conversational capabilities from systems available in the future.

Future research should examine AI–counseling integration across larger and more diverse populations and should compare different age groups, clinical concerns, cultural settings, and counseling modalities. Longitudinal qualitative research would be particularly valuable for examining how patterns of complementary use, parallel use, substitution, and conflict change over the course of therapy. Future studies could also investigate dyadic perspectives by interviewing clients and their counselors separately or together in order to identify areas of convergence and disagreement regarding the role of AI. Mixed-method research could examine whether specific forms of AI use are associated with therapeutic alliance, treatment engagement, disclosure, symptom change, dropout, or help-seeking behavior. Research should also distinguish between general-purpose generative AI, dedicated mental health chatbots, clinician-supervised tools, and other AI-enabled systems rather than treating all forms of AI support as equivalent. Greater attention should additionally be paid to high-risk situations, privacy concerns, misinformation, dependency, and the effects of abrupt discontinuation of frequently used AI systems.

In practice, counselors should consider AI use as a potentially relevant part of clients’ contemporary mental health environment and create opportunities for clients to discuss such use without assuming that it necessarily reflects dissatisfaction with therapy. Routine, nonjudgmental inquiry about whether clients use AI for emotional support, psychoeducation, reflection, or preparation may help bring otherwise hidden material into the therapeutic process. Practitioners can help clients distinguish between low-risk reflective uses of AI and situations that require professional judgment, crisis support, or individualized clinical assessment. Counselors may also encourage clients to critically evaluate AI-generated advice, compare it with established therapeutic goals, and bring confusing or contradictory outputs into sessions for discussion. Clear boundaries should be established around privacy, dependency, crisis situations, and the limits of automated systems. Most importantly, AI should be positioned as a supplementary resource that can support continuity and self-reflection while preserving the central role of human relational engagement, professional accountability, and individualized therapeutic understanding.

Acknowledgments

We would like to express our appreciation and gratitude to all those who cooperated in carrying out this study.

Declaration of Interest

The authors of this article declared no conflict of interest.

Ethical Considerations

The study protocol adhered to the principles outlined in the Helsinki Declaration, which provides guidelines for ethical research involving human participants.

Transparency of Data

In accordance with the principles of transparency and open research, we declare that all data and materials used in this study are available upon request.

Funding

This research was carried out independently with personal funding and without the financial support of any governmental or private institution or organization.

Authors’ Contributions

All authors equally contributed in this article.

Declaration

In order to correct and improve the academic writing of our paper, we have used the language model ChatGPT.