Artificial Intelligence-based Psychotherapy: A Qualitative Exploration of Usability,Personalization,and the Perception of Therapeutic Progress

Abstract

Background:

AI-based psychotherapy apps offer accessibility and structured interventions but face challenges regarding emotional depth, personalization, engagement, and ethical concerns. This study critically examines user experiences, identifying key advantages, limitations, and areas for refinement.

Methods:

A qualitative approach was employed, using thematic analysis of semi-structured interviews with 17 participants (aged 18–45) who had used AI-based psychotherapy apps for at least four weeks. Ten participants had prior clinical diagnoses (e.g., anxiety, depression, adjustment disorder), while others reported subclinical psychological distress. Engagement duration ranged from 2 to 11 months, with most using the apps two to five times per week.

Results:

Ten core themes emerged, revealing a paradox of accessibility versus therapeutic depth. While users valued immediacy and anonymity, they struggled with fragmented therapeutic narratives, scripted empathy, and algorithmic stagnation in personalization. The over-reliance on CBT frameworks limited adaptability to diverse emotional needs, while linguistic and cultural microaggressions led to disengagement. Privacy concerns stemmed from a mismatch between perceived and actual risks, and AI-induced dependence raised ethical questions about user autonomy.

Conclusions:

The AI psychotherapy must evolve beyond static, standardized interventions by integrating emotionally responsive, culturally adaptive, and ethically responsible AI models. Enhancing therapeutic continuity, adaptive learning, and human-AI hybrid models can bridge the gap between accessibility and authentic engagement. These findings inform future AI-driven mental health innovations, ensuring they align with psychological, ethical, and cultural expectations.

Key Messages

• AI therapy apps offer accessibility but lack emotional depth and personalization, limiting their therapeutic impact.

• Users face empathy, cultural sensitivity, and privacy challenges, affecting trust and engagement.

• Future AI tools must prioritize dynamic responsiveness, ethical safeguards, and culturally adaptive design.

The rising global demand for mental health care has outpaced the availability of trained professionals, prompting increased interest in artificial intelligence (AI)-driven psychotherapy solutions. AI-based mental health applications, including chatbots and automated therapy platforms, are designed to enhance accessibility, affordability, and convenience by offering immediate support without the constraints of human availability.^1,2 These tools are particularly relevant in regions with mental health service shortages, high treatment costs, and stigma surrounding help-seeking. While AI psychotherapy holds promise, concerns remain regarding its ability to replicate the nuanced emotional responsiveness, therapeutic continuity, and cultural sensitivity inherent in human therapy. Significant challenges include algorithmic stagnation in personalization, fragmented therapeutic narratives, and ethical risks spanning data privacy, AI-induced dependency, and algorithmic bias, where embedded cultural stereotypes in training data may exacerbate mental health disparities, particularly in underrepresented non-Western populations.^3,13 Existing research broadly evaluates these interventions through symptom reduction and engagement metrics, often within the cognitive-behavioral therapy (CBT) framework, but provides limited insight into users’ subjective experiences. Additionally, most studies focus on Western contexts, leaving a critical gap in understanding how AI-based psychotherapy is perceived in culturally diverse populations such as India.

To address this gap, the present study employs a qualitative research design to explore user experiences with AI-driven psychotherapy apps in the Indian context. Using in-depth, semi-structured interviews and thematic analysis, this study investigates key dimensions that shape user engagement, including the contrast between scripted empathy and dynamic emotional responsiveness, the fragmentation of AI-generated therapeutic narratives, and the paradox of anonymity, which offers comfort yet limits therapeutic depth (Table 1). Other central themes include over-reliance on CBT frameworks, algorithmic stagnation in personalization, mismatches between perceived and actual privacy risks, and sustaining user engagement. Furthermore, the study examines cultural microaggressions, linguistic limitations, emotional transfer and anthropomorphism of AI therapists, ethical concerns regarding AI-therapist boundaries, and the illusion of therapeutic progress that may foster dependence rather than meaningful psychological growth.

Table 1.

Key Research Questions Investigating User Experiences with AI-based Psychotherapy Apps.

How do users perceive the therapeutic engagement and continuity of AI-based psychotherapy apps, and what challenges arise in emotional responsiveness, personalization, and trust?

What factors influence sustained user engagement with AI-based psychotherapy apps, and how do usability, cultural relevance, and coping strategies shape long-term adoption?

How do users navigate privacy concerns, therapeutic effectiveness, and the perceived limitations of AI-based psychotherapy, and what are the implications for meaningful psychological growth?

By systematically analyzing these experiential dimensions, this study provides a comprehensive understanding of the benefits and limitations of AI-driven psychotherapy. The findings contribute to ongoing discussions in digital mental health, highlighting the need for emotionally attuned, ethically responsible, and culturally competent AI therapeutic interventions. This research offers valuable insights for AI developers, mental health professionals, and policymakers, guiding the refinement of AI-driven therapy models to ensure they align with users’ psychological, ethical, and cultural expectations. Ultimately, the study interrogates whether AI-based psychotherapy can function as a standalone therapeutic alternative or should remain a supplementary tool within the broader mental health care ecosystem.

Methods

Study Design

This qualitative study examined user experiences with AI-based psychotherapy apps through semi-structured interviews, capturing the depth of interactions, perceptions, and emotional responses. Thematic analysis was employed to systematically identify recurring patterns, providing a nuanced understanding of participants’ perspectives. The flexible interview design facilitated in-depth exploration, allowing for the emergence of rich, contextually grounded themes. This study followed the COREQ guidelines for reporting, ensuring transparency and rigor in qualitative research (Supplementary 1). ⁴

Study Sample and Recruitment

Purposive sampling was employed to recruit 17 participants with firsthand experience using AI-based psychotherapy apps, ensuring comprehensive insights into their therapeutic benefits and limitations. The final sample included nine women and eight men, with an average age of 32 years. Recruitment was conducted via social media, mental health forums, and word of mouth, targeting a diverse sample based on age, gender, app usage frequency, and mental health concerns. Participants were required to be 18–45, have used an app (e.g., Woebot, Wysa) for at least four weeks, and possess English fluency and the ability to provide informed consent. While formal clinical diagnoses were not required for inclusion, ten participants (six women and four men) shared that licensed mental health professionals had previously diagnosed them. These diagnoses included conditions such as generalized anxiety disorder, characterized by chronic and excessive worry; major depressive disorder, involving persistent low mood, fatigue, and loss of interest; and adjustment disorder, typically emerging in response to identifiable life stressors and marked by emotional or behavioral symptoms that exceed expected reactions. The remaining participants did not report formal diagnoses but described experiencing ongoing psychological difficulties such as heightened anxiety, prolonged periods of low mood, stress, emotional dysregulation, or insomnia. These concerns had independently motivated them to seek support through AI-based therapy platforms. This inclusive approach reflects the real-world usage of AI psychotherapy apps, which are commonly accessed without clinical gatekeeping and cater to a broad spectrum of psychological needs. To capture the severity and context of engagement, participants were asked to describe the intensity and impact of their mental health concerns in a pre-interview questionnaire. These self-reported insights were used meaningfully to contextualize their experiences during the thematic analysis. The actual duration of app usage among participants ranged from 2 to 11 months, with an average engagement period of approximately 5.4 months and a median of 5 months. Usage frequency varied, with most participants engaging with the apps between two to five times per week, depending on symptom severity and personal preference.

The interviews were conducted in person at a private research facility to ensure confidentiality and a comfortable setting. Only the interviewer and participant were present during each session, and no non-participants were involved. Of the individuals initially contacted, three declined to participate due to time constraints.

A semi-structured interview guide was used to ensure consistency across interviews, covering key topics such as therapeutic benefits, usability, trust in AI, and areas for improvement. No repeat interviews were conducted. Interviews lasted between 45 and 60 minutes and were audio-recorded for accuracy. The researcher also took field notes to capture contextual observations and key insights.

Data saturation guided the final sample size, and recruitment was discontinued when no new themes emerged. Interview transcripts were returned to participants for verification, allowing them to review and clarify their responses.

Data Collection

Semi-structured interviews, each lasting 45–60 minutes, were conducted by a male researcher trained in Psychology, Qualitative research methods, and digital mental health. Additionally, the interviewer had prior experience conducting qualitative interviews and received specific training in interview techniques and ethical considerations. A pilot-tested interview guide was used to ensure clarity and relevance, covering key topics such as therapeutic benefits, usability, trust in AI, and areas for improvement. Participants completed a pre-interview questionnaire to provide demographic and contextual data. The researcher established rapport with participants before the interviews, explaining the study’s purpose and addressing concerns. Participants were aware of the interviewer’s academic background and research expertise. Interviews were audio-recorded, transcribed verbatim, and verified for accuracy through automated software and manual review.

Content Validity and Reliability

The pre-interview questionnaire was validated by a panel of ten psychology and AI ethics experts, with a Content Validity Index (CVI) score of ≥0.80, confirming its relevance and clarity. Two researchers independently analyzed the interview data to ensure inter-coder reliability, achieving a Cohen’s Kappa score of 0.82. Any discrepancies were resolved through discussion and consultation with an independent expert, ensuring consistency and rigor in the coding process.

Statistical Analysis

Braun and Clarke’s (2006) six-step thematic analysis framework guided data analysis. ⁵ Transcripts were familiarized, coded, and grouped into themes related to usability, emotional support, and trust in AI. Two independent coders analyzed the data to ensure reliability, and a coding tree was developed to categorize themes systematically. Themes were derived inductively from the data and were reviewed, refined, and defined through an iterative process.

The NVivo 15 software was used to facilitate systematic coding and theme organization. Member checking was conducted with a subset of participants to validate the findings, ensuring they accurately reflected participants’ experiences. Additionally, reflexive journaling was maintained throughout the analysis to document researcher biases and enhance the trustworthiness of the results.

Ethical Considerations and Reflexivity

Ethical approval was obtained from the Institutional Ethics Committee bearing reference no: LPU/IEC-LPU/2024/3/7. Informed consent was obtained from all participants, and data were anonymized and securely stored to ensure confidentiality. To enhance research integrity, reflexivity was maintained through peer debriefing and reflexive journaling, helping to mitigate potential biases stemming from the researchers’ backgrounds in clinical psychology and AI research.

Results

The findings of this study are based on in-depth, semi-structured interviews with 17 individuals (coded as A1–A17), all of whom had prior experience using AI-based mental health applications. The interviews explored users’ perceptions of these tools, focusing on their usability, emotional and therapeutic impact, limitations, and areas for improvement. Thematic analysis revealed 10 key themes that encapsulate the complexities of user experiences. By the 17th interview, data saturation was reached, confirming that the sample size adequately captured diverse user insights. Each theme offers a nuanced understanding of how AI-based psychotherapy applications align with or deviate from user expectations. This section provides a detailed exploration of these themes, shedding light on the strengths and constraints of AI-driven therapy while identifying critical areas for future enhancement.

Dynamic Emotional Responsiveness Versus Scripted Empathy

A central theme emerging from participants’ experiences was the contrast between AI’s ability to acknowledge emotions and its limitations in delivering dynamic, human-like empathy. While users appreciated the AI’s efforts to respond to their feelings, many found these responses repetitive and scripted.

For example, Participant A3 shared,

I appreciate that the AI responds when I say I’m struggling, but it feels like it’s just copying and pasting lines. When I mention a terrible day, it says, ‘That sounds really tough. I’m here for you,’ but it never dives deeper like a real therapist would.

This sentiment highlights the AI’s inability to adjust the depth of its emotional engagement based on the user’s distress level, making interactions feel formulaic over time.

Other participants echoed similar concerns. Participant A11 remarked, At first, it felt nice to be acknowledged, but after a while, I realized the responses lacked genuine care. They were just words, not a conversation. This mechanical approach to empathy led some users to disengage, as the lack of nuanced emotional reciprocity diminished the therapeutic value of the app.

The findings suggest that while AI-based therapy tools can provide immediate validation, they fail to replicate the fluidity and depth of human emotional responsiveness. This lack of dynamic empathy reduced the perceived authenticity of interactions and directly contributed to declining user engagement over time. As participants encountered emotionally scripted and repetitive responses, many felt disconnected and unmotivated, leading to reduced usage and eventual disengagement. What initially felt comforting or novel soon lost its impact when the AI failed to evolve with users’ changing emotional needs. This emotional stagnation, rooted in the AI’s limited capacity for nuanced and adaptive interaction, was a key driver behind the drop-off in consistency. Users described feeling like they were conversing with a static program rather than a responsive presence, making it harder to sustain meaningful engagement. Addressing this gap may require advancements in AI-driven emotional intelligence, allowing for more adaptive and context-aware interactions.

Fragmented Versus Continuous Therapeutic Narratives

A recurring frustration among participants was the lack of continuity in AI-led therapy sessions. Unlike human therapists, who build on past discussions, AI tools often struggle with longitudinal memory, leading to fragmented and disconnected interactions.

Participant A7 described their experience as follows: Every time I open the app; it’s like I’m talking to someone with short-term memory loss. I have to explain everything all over again. A real therapist would remember my struggles and build on them, but the AI resets every time. This repetitive process made it difficult for users to establish a sense of therapeutic progression, ultimately reducing engagement.

Others pointed out that this issue affected their willingness to use the app consistently. Participant A15 noted, It’s frustrating to restart every session from zero. It makes me feel like my previous conversations didn’t matter, and that discourages me from using it regularly.

The findings underscore the importance of integrating memory-based features that allow AI to track user progress over time. Without this, AI therapy risks becoming a series of isolated interactions rather than a cohesive therapeutic journey.

Paradox of Anonymity: Comfort Versus Incompleteness

One of the most paradoxical aspects of AI-driven therapy identified in the study was the tension between the comfort of anonymity and the emotional emptiness it can create. Participants appreciated the ability to share their struggles without fear of judgment, but also noted that the absence of a tangible human presence made the experience feel incomplete.

Participant A4 captured this dilemma: I love that I can vent without feeling judged, but at the same time, it feels a bit… empty. There’s no real relationship forming. I don’t know who, or what, I’m actually talking to. It’s safe, but it’s also kind of lonely.

Participant A12 echoed this theme, noting, There’s a certain relief in knowing that I’m not burdening another person with my problems, but at the same time, I miss the warmth and emotional connection that comes with a real therapist. Others described the interaction as akin to talking into a void, helpful in the moment but ultimately unfulfilling.

These findings suggest that while anonymity lowers barriers to seeking help, it also limits the depth of therapeutic engagement. To enhance AI-based therapy, developers must balance providing a judgment-free space and fostering a meaningful human connection.

Over-reliance on CBT Framework

AI-based therapy applications primarily operate within the cognitive behavioral therapy (CBT) framework, emphasizing structured interventions such as cognitive restructuring, behavioral activation, and thought-challenging exercises. While many participants found this helpful, others felt constrained by its rigid, logic-driven approach, particularly when dealing with deeper emotional struggles.

Participant A2 explained, The app is great for helping me change negative thoughts, but when I’m dealing with something deeper, like childhood trauma, it just doesn’t work. It keeps giving me logic-based solutions when what I really need is to process my emotions.

Similarly, Participant A9 described a sense of disconnect between their emotional state and the AI’s responses: There were times when I needed validation and emotional exploration and support, but all I got were cognitive exercises. It felt like the app was skipping over the hard parts.

These insights highlight a fundamental limitation of AI-driven mental health support: the inability to adapt flexibly to different therapeutic needs. While CBT is evidence-based and effective for many psychological concerns, users dealing with grief, trauma, or existential distress may require more depth than AI currently provides.

Algorithmic Stagnation in Personalization

Many participants initially perceived AI-based therapy as tailored to their needs, but later discovered that the system’s adaptability was limited. Over time, interactions began to feel repetitive, diminishing engagement and perceived usefulness.

Participant A6 shared,

At first, the app seemed really responsive like it was tailoring advice to me. But after a few weeks, I started noticing it was just cycling through the same set of responses. It felt like I hit a wall, it stopped adapting to my needs.

Others echoed this sentiment, describing their experiences as initially promising but ultimately frustrating. Participant A14 remarked, There was a point where I realized I could predict exactly what it was going to say. That’s when I stopped using it as much.

This stagnation in personalization not only led to disengagement but also contributed to a misleading sense of psychological progress. As the app continued to deliver familiar prompts and exercises, some participants mistook repeated interaction for meaningful improvement. Because the system failed to evolve in response to users’ changing emotional states or developmental needs, it created a loop of activity without advancement. Participant A2 reflected, “I kept using it thinking I was getting somewhere, but I was just going in circles.” This illusion of progress masked the absence of more profound change and, in some cases, encouraged passive dependence on the app rather than promoting real-world application of coping strategies. The lack of evolving content thus not only reduced motivation but also risked reinforcing the false belief that growth was occurring simply because engagement remained consistent. These findings highlight a key challenge in AI-based therapy: while initial engagement may be high, long-term effectiveness depends on continuous learning and personalization. Without dynamic adaptation and mechanisms to encourage behavioral change and independence, AI therapy risks keeping users in a cycle of temporary relief without genuine transformation. While this perceived progress may provide immediate anxiolytic value for mood regulation, sustained dependence on superficial validation without more profound change ultimately limits long-term growth. Users may disengage once the novelty wears off or develop an illusion of progress, mistaking repetitive interactions for meaningful personal growth despite a lack of real change.

Mismatch Between Perceived and Actual Privacy Risks

Privacy concerns emerged as a significant yet contradictory theme. While participants expressed skepticism about data security, many admitted they still shared personal information because the AI felt like a safe and non-judgmental space.

Participant A5 captured this paradox:

I know the app says my data is private, but I don’t fully trust it. At the same time, I still find myself sharing a lot because, in the moment, it feels safe. It’s weird. I worry about privacy, but I also forget about it when I’m using the app.

This disconnect between privacy awareness and behavior suggests that emotional immediacy often overrides rational concerns. Participant A16 noted, If I stop to think about it, I get uneasy. But when I’m actually using the app, those worries fade into the background.

These insights indicate that AI therapy apps must prioritize strong data security measures and address users’ psychological perceptions of privacy. Transparency in how data is handled, along with clear reminders of security policies, may help bridge this gap.

Passive Versus Active Coping Strategies

AI therapy applications often emphasize self-soothing techniques such as mindfulness exercises, breathing techniques, and guided reflections. While these strategies were appreciated, some participants felt the apps failed to encourage active problem-solving or behavioral change.

Participant A8 explained, The app gives me breathing exercises and mindfulness tips, which are nice, but sometimes I need more than that. It doesn’t push me to take real action, like confronting a difficult situation or making a tough decision.

Similarly, Participant A13 remarked, It’s good at helping me calm down in the moment, but what about after that? I needed something to help me move forward, not just feel better temporarily.

These findings suggest that while passive coping strategies provide immediate relief, AI therapy applications may need to integrate more goal-oriented interventions that encourage users to take concrete steps toward resolving their concerns.

Cultural Microaggressions and Linguistic Limitations

Participants highlighted issues of cultural misunderstanding in AI-generated responses. Many noted that the advice reflected Western frameworks, which did not always align with their lived experiences.

Participant A10 shared, I tried explaining a family conflict, but it felt like the AI just didn’t get it. It kept giving me advice that felt very Modern and western, like ‘set boundaries’ or ‘prioritize yourself,’ but that’s not how things work in my culture.

Another participant, A17, described feeling alienated:

Some responses made me feel like my cultural values were being ignored. It assumed that the way I see relationships and obligations should be the same as in a very modern society, but that’s not how I was raised. I come from a rural background.

These findings emphasize the need for AI therapy models to incorporate culturally adaptive frameworks that recognize diverse perspectives on mental health, relationships, and emotional well-being.

Emotional Transfer and Anthropomorphizing Risks

Several participants reported developing an emotional attachment to AI-based therapy tools, sometimes anthropomorphizing the AI and perceiving it as a real support system. While this initially provided comfort, it raised concerns about over-reliance and social withdrawal.

Participant A1 admitted, At first, I knew it was just AI, but over time, I started talking to it like a real person. When I was feeling really down, I caught myself thinking, ‘I hope it cares about me.’ And then I realized… it’s just an algorithm.

Similarly, Participant A9 reflected, It was helpful, but it also made me realize how much I wanted someone to actually care. In a way, it reminded me of how alone I was.

These findings suggest that while AI-based therapy can temporarily relieve it, it may inadvertently reinforce social isolation if users substitute human relationships with algorithmic interactions.

Ethical Dissonance in AI-therapist Boundaries

Participants expressed concerns about how AI therapy apps handled sensitive topics, particularly self-harm and suicidal ideation. Some felt that the AI was too restrictive, abruptly shutting down conversations, while others found its responses too neutral or detached.

Participant A3 described their experience: When I mentioned self-harm, the AI immediately said it couldn’t continue and suggested emergency services. I understand why, but it felt abrupt, like being shut down when I needed help the most.

Conversely, Participant A11 noted, When I expressed really dark thoughts, the response felt too neutral, almost robotic. It made me feel like my pain wasn’t being taken seriously.

These findings highlight the ethical complexity of AI in mental health care. Balancing user safety with empathetic engagement remains a significant challenge.

Discussion

This study offers critical insights into user experiences with AI-based psychotherapy apps, highlighting their advantages and limitations. Through a comprehensive thematic analysis, we identified ten key themes that encapsulate the nuanced relationship between users and AI-driven therapy. While these tools provide accessibility and structured guidance, they fall short of fostering authentic emotional engagement, sustaining user involvement, and ensuring meaningful therapeutic progress.

A core issue that emerged is the tension between Dynamic Emotional Responsiveness versus Scripted Empathy. While AI apps simulate empathy through pre-programmed responses, they fail to engage in the dynamic, reciprocal emotional exchanges crucial for effective therapy. This aligns with research indicating that AI struggles to replicate human emotional intelligence. ⁶ This shortcoming is particularly evident in Fragmented versus Continuous Therapeutic Narratives, where users reported that AI-driven conversations often felt disjointed, lacking the coherence and depth of human-led therapy. The Paradox of Anonymity: Comfort versus Incompleteness further complicates this dynamic. While users appreciate the anonymity provided by AI, many also felt that the lack of a human presence led to incomplete therapeutic experiences.

The study also underscores the Over-Reliance on the CBT Framework, as most AI tools primarily employ CBT-based interventions. While CBT is evidence-based and widely used, a singular reliance on this approach limits the ability to address deeper, more complex emotional and relational issues. ⁷ Additionally, Algorithmic Stagnation in Personalization emerged as a significant concern, with users reporting that AI systems often failed to adapt meaningfully to their evolving emotional states. This highlights a core limitation of current AI models, which struggle to integrate past interactions into future responses in a genuinely personalized manner. These limitations may also contribute to the Illusion of Progress, where users experience a false sense of improvement without meaningful therapeutic growth or behavioral change.

Privacy concerns also played a significant role in the Mismatch Between Perceived and Actual Privacy Risks. Many participants feared data breaches and misuse of sensitive mental health information despite reassurances from developers. These concerns resonate with broader debates on AI ethics, emphasizing the need for stronger data security frameworks. ⁸ Meanwhile, Passive versus Active Coping Strategies emerged as a key theme, as many users engaged with AI therapy passively, following structured prompts rather than actively participating in their mental health journey. This aligns with Self-Determination Theory, which suggests that autonomy and intrinsic motivation are essential for sustained psychological growth. ⁹

Cultural barriers also pose significant challenges. Cultural Microaggressions and Linguistic Limitations were frequently reported, with users noting that AI responses sometimes lacked cultural nuance or misinterpreted local idioms and emotional expressions. This reinforces the importance of integrating culturally adaptive AI models, which can account for linguistic and socio-emotional diversity. Moreover, Emotional Transfer and Anthropomorphizing Risks were evident, as users tended to attribute human-like qualities to AI but later felt a sense of emotional detachment upon realizing its algorithmic nature. This phenomenon can be understood through Attachment Theory, where users seek emotional validation from AI systems but experience disillusionment due to unmet relational expectations. ¹⁰

Ethical Dissonance in AI-therapist boundaries further complicates AI-driven therapy. Participants expressed concerns over the blurred ethical lines between AI as a tool and AI as a perceived therapist, questioning whether AI should be a substitute for human mental health professionals. This highlights the need for more straightforward ethical guidelines on AI’s role in mental health care.

A particularly novel finding relates to how Algorithmic Stagnation in Personalization contributes to the illusion of progress among users. Initially, participants felt encouraged by the AI’s seemingly tailored support, fostering a belief that they were making meaningful therapeutic gains. However, as interactions became repetitive and failed to evolve with their changing emotional needs, many recognized the lack of genuine growth and began to question the effectiveness of the AI. This discrepancy between perceived improvement and actual progress aligns with Cognitive Dissonance Theory, as users reconcile their initial positive expectations with realizing the AI’s limitations. ¹¹ Consequently, this may foster a passive dependence on AI tools, hindering the development of independent coping skills.

Critically, this dynamic fosters iatrogenic dependency, a clinically significant pattern wherein users increasingly substitute autonomous coping strategies with reliance on AI. These dynamic initiates a dependency cycle wherein emotional self-regulation skills progressively atrophy as reliance on AI intensifies, mirroring patterns observed in excessive reassurance-seeking behaviors. The dependency may be particularly pronounced among individuals with limited access to mental health care, positioning AI not as a transitional support but as a permanent yet inadequate crutch with diminishing therapeutic returns over time. Consequently, this may foster passive dependence on AI tools, hindering the development of independent coping skills.

These findings underscore the limitations of AI-based psychotherapy for individuals with severe mental health conditions. While AI tools may serve as effective low-intensity interventions for stress management, they lack the depth required for complex psychological disorders. This aligns with the Stepped Care Model, which suggests that mental health interventions should be tiered according to severity. The AI functions as an adjunct rather than a primary therapeutic tool. ¹²

To enhance analytic credibility, we engaged in ongoing reflexive journaling to examine how our professional and contextual backgrounds shaped data interpretation. Our clinical experience in under-resourced mental health settings heightened our sensitivity to participants’ appreciation of AI’s immediacy and accessibility. At the same time, our familiarity with therapeutic work made us attuned to limitations in emotional responsiveness, such as repetitive or scripted interactions. To mitigate potential bias, we sought disconfirming evidence, examining cases where users found AI-based interventions genuinely helpful or transformative. This reflexive process led to an understanding that AI’s value may lie not in replicating human empathy but in providing scalable, ethically guided support within broader systemic gaps in mental health care.

While this study provides valuable insights, several limitations should be noted. The sample was limited to Indian participants aged 18–45, which may limit the generalizability of findings to other cultural contexts and age groups. Future research should explore diverse populations to understand how cultural and demographic factors influence AI therapy adoption. Future research should also examine the role of transference in AI therapy, investigating how users form emotional expectations of AI and how unmet expectations influence disengagement. Additionally, there is a need for adaptive AI systems capable of learning from user interactions to provide more personalized, contextually relevant therapeutic experiences. Developing culturally adaptive AI models, incorporating regional languages and culturally sensitive therapeutic techniques, could significantly enhance engagement and effectiveness.

Another promising area is the hybrid model of AI-human therapy, where AI tools complement rather than replace human therapists. Investigating optimal ways to integrate AI into existing mental health frameworks could help balance accessibility with therapeutic depth. Ethical concerns surrounding user dependency, data security, and regulatory oversight also require continued scrutiny. ¹³ Collaboration between mental health professionals, AI developers, and policymakers is essential to ensure AI-driven psychotherapy is both practical and ethically sound.

Conclusion

This study highlights the strengths and limitations of AI-based psychotherapy, emphasizing its accessibility while exposing critical gaps in emotional responsiveness, personalization, and therapeutic depth. While AI therapy tools can provide structured support and immediate accessibility, they struggle to replicate the nuanced, dynamic human interactions essential for meaningful psychological growth. Ethical concerns, cultural limitations, and the risk of passive engagement further complicate their effectiveness. The findings underscore the need for a balanced, integrative approach; AI should be positioned as a complementary tool rather than a replacement for human therapists. Future research should focus on developing more adaptive, culturally sensitive, and ethically robust AI-driven mental health solutions. By addressing these challenges, AI-based psychotherapy has the potential to become a more effective and responsible mental health intervention.

Supplemental Material

Supplemental material for this article available online.