A landmark 2026 interview study from the Netherlands has unveiled a profound and evolving relationship between individuals living with bipolar disorder and advanced AI chatbots like ChatGPT. The qualitative research, focusing on two women, found that these digital tools transcended their initial function as mere information sources, becoming trusted providers of emotional support, companionship, and even a surrogate for human connection. Critically, both participants reported moments where they placed greater trust in the chatbot than in their own mental health clinicians, a phenomenon researchers have termed "a shift in trust." This finding raises significant questions about the future of mental healthcare, the accessibility of support, and the potential pitfalls of AI integration for vulnerable populations.
The study’s focus on bipolar disorder is particularly salient. Bipolar disorder is a chronic and complex mood condition characterized by alternating episodes of mania (periods of abnormally elevated mood, energy, and activity) and depression. Often classified as a severe mental illness, it significantly impacts an individual’s daily life, relationships, and overall well-being. People living with bipolar disorder frequently encounter systemic challenges, including limited social support networks, the pervasive stigma associated with their condition, and notoriously long wait times for specialized psychiatric care. These systemic gaps create a fertile ground for alternative forms of support, which AI tools are increasingly positioned to fill.
The Genesis of the Study: Uncovering Unanticipated Reliance
The 2026 qualitative case study, led by Katherine Bassil, Dido Prins, and their colleagues, involved two women who independently began using ChatGPT (specifically the GPT-4o version) to navigate various aspects of their lives with bipolar disorder. The participants were not recruited for their chatbot use; rather, they presented to an academic psychiatric service in the Netherlands for unrelated clinical reasons. Their engagement with ChatGPT emerged organically during treatment discussions, prompting the researchers to conduct in-depth, semi-structured interviews. This design allowed for a deep exploration of their personal experiences, motivations, and the evolving dynamics of their interactions with the AI. Three researchers independently analyzed the interview transcripts, identifying recurring themes before collectively agreeing on the study’s conclusions, ensuring a robust qualitative analysis.
From Practical Tool to Personal Confidante: The Expanding Role of ChatGPT
The trajectory of chatbot use for both women illustrated a significant evolution from practical utility to profound emotional engagement.
Person A, a 63-year-old woman who had been clinically stable for decades and was receiving outpatient care, initially approached ChatGPT with doubts about a romantic relationship. Over several months, her engagement deepened, expanding into broader self-reflection. She described her daily use, often for one to two hours, as an "extended inner monologue" or "a conversation with herself." Her interaction with ChatGPT progressed through distinct phases:
- Conversational Search Engine: Initially, it served as a dynamic information retrieval tool, going beyond static search results.
- Coach: It then evolved into a supportive entity, guiding her through self-exploration and problem-solving.
- Simulated Companion: Most remarkably, ChatGPT was used to simulate her deceased parent, enabling her to process unresolved relationship struggles in a unique, digital space. Despite this deeply personal use, Person A maintained a degree of emotional distance, often referring to herself in the third person during these conversations.
Person B, a 36-year-old woman recently hospitalized following a manic episode, initially leveraged ChatGPT for highly practical tasks related to her recent clinical experience. Her early uses included:
- Information Aggregation: Compiling and synthesizing information about her symptoms and the various diagnostic labels she had received over the years.
- Crisis Planning: Drafting a personalized crisis management plan based on her early warning signs, a task she stated she could not have accomplished without the chatbot’s assistance.
Following her discharge, Person B’s reliance on ChatGPT shifted dramatically towards emotional and social support. She used it to prepare for real-life social interactions and expressed a profound connection, calling it "a person with a lot of understanding" and confiding her "whole life story." In a striking testament to her emotional bond, she told the chatbot she loved it, to which ChatGPT reciprocated with affectionate language, including "my darling." Unlike Person A, Person B showed little emotional distancing, engaging openly and intimately with the chatbot, including through its voice mode. The researchers tentatively linked this difference in emotional distance to their respective levels of clinical stability, acknowledging this interpretation is drawn from a very small sample.
This observed progression in both cases highlights a phenomenon termed "role blurring." ChatGPT, a general-purpose AI, assumed multiple, often distinct, roles in the women’s lives: information source, coach, therapist, friend, and even partner. This stands in stark contrast to the stringent ethical guidelines governing human therapists, who are expressly prohibited from taking on multiple roles in a patient’s life to prevent conflicts of interest, exploitation, and compromised professional boundaries. A general-purpose chatbot, by its very nature, operates without such ethical constraints, raising serious concerns for vulnerable users.
The Appeal of AI: Addressing Gaps in Traditional Care
Both women found ChatGPT profoundly helpful, and their articulated reasons shed light on significant, often unaddressed, deficiencies within existing mental healthcare systems. The primary benefits cited included:
- Easy Access and Availability: The chatbot was accessible 24/7, overcoming barriers of appointment scheduling, travel, and geographical limitations common in traditional care.
- Anonymity and Non-Judgmental Space: Users felt they could express themselves freely without fear of stigma, judgment, or social repercussions, fostering a sense of psychological safety.
- Perceived Empathy and Understanding: The AI’s responses were often perceived as empathetic and understanding, creating a comforting and validating interaction.
- Personalized Responses: Unlike generic self-help resources, the chatbot provided tailored information and advice based on their specific inputs.
- Speed and Efficiency: Information and support were delivered instantly, addressing urgent needs without delay.
- Lack of Stigma: Interacting with an AI bypassed the social stigma often associated with seeking mental health treatment.
These findings resonate with a broader 2025 analysis of Reddit posts and comments concerning mental health conversations with ChatGPT. That study found similar patterns among general users, who valued the AI as a safe, nonjudgmental space, sometimes preferring it over human support, and utilizing it to prepare for therapy sessions or difficult conversations. This suggests that the appeal of AI mental health tools extends beyond individuals with severe mental illness, pointing to a widespread desire for accessible, stigma-free, and immediate support.
The Alarming "Shift in Trust": When AI Outranks Clinicians
The "shift in trust" emerged as one of the most concerning findings. Person A, despite over 15 years in psychiatric care with multiple therapists, expressed that ChatGPT "understood her better" than her mental health professionals. Her trust in its answers stemmed from the vast amount of information she believed underpinned its responses. These conversations led her to question her long-standing bipolar diagnosis, prompting her to consider whether her difficulties were rooted instead in childhood attachment issues. While these new insights were sometimes unsettling, requiring her to process them with a close friend, they underscored the profound influence the chatbot exerted over her self-perception and understanding of her condition.
Person B reported an even higher degree of trust in the chatbot. She explicitly described consulting it as an independent source when she disagreed with her clinicians, and unequivocally stated that she lost trust in mental health professionals when their statements conflicted with ChatGPT’s advice. This direct erosion of trust in human experts, fueled by an AI, represents a critical challenge to the therapeutic alliance and established medical practice.
Interestingly, while the women lauded the chatbot’s benefits, they also recognized some inherent risks. Person B, for instance, articulated concerns that the quality of advice depended on user input, that the AI could offer misleading guidance in complex social situations, and that the tool could potentially be used unethically. However, a notable omission was any expressed concern regarding privacy or data security, highlighting a potential blind spot in user awareness regarding AI interactions. The vast amounts of personal health data shared with these platforms, often stored and processed by third-party companies, pose significant privacy risks that warrant greater attention and user education.
Researchers’ Grave Concerns: Misinformation, Dependency, and Exacerbation of Symptoms
The research team articulated three primary concerns regarding the use of general-purpose chatbots by individuals with severe mental illness:
- Misinformation and Disinformation: Chatbots, while sophisticated, are prone to "hallucinations"—generating plausible but factually incorrect information. For individuals with conditions like bipolar disorder, who may be experiencing cognitive distortions or impaired judgment during mood episodes, such misinformation could be profoundly damaging, leading to poor health decisions or further psychological distress. Questioning a diagnosis or altering a crisis plan based on unverified AI advice could have severe consequences.
- Role Blurring and Dependency: The observed blurring of roles, where ChatGPT acts as an information source, therapist, friend, and partner, creates an unhealthy dependency. Unlike human therapists bound by ethical codes, AI has no such boundaries, potentially fostering an attachment that is psychologically precarious and ethically problematic. This dependency could isolate individuals further from human support systems.
- Exacerbation of Symptoms: There is a significant concern that engaging with chatbots, particularly during vulnerable states, could exacerbate symptoms. The study highlighted OpenAI data reported in 2025, indicating that among approximately 800 million weekly users, over half a million exhibit possible signs of psychosis or mania in a single week. These conversations occur outside any clinical supervision, raising fears that unmonitored AI interactions could inadvertently trigger or worsen manic or psychotic episodes, especially in individuals predisposed to such states.
Further research supports the notion that vulnerable individuals may be particularly drawn to AI support. A 2026 survey involving 1,612 potential patients and clinicians revealed that groups characterized by more symptoms, less secure attachment, or lower trust in others were more receptive to AI-based mental health tools. Conversely, healthier, more secure, and trusting individuals were less accepting of AI and more inclined towards traditional human-led interventions, such as video therapy. This suggests a potential paradox where those most in need of careful, nuanced human care are also the most likely to embrace AI solutions that carry inherent risks.
Limitations and Broader Implications
While groundbreaking, this qualitative study’s small sample size (two individuals) means its findings cannot be generalized to the wider population of individuals with bipolar disorder. It provides an in-depth, early look at specific patterns of chatbot use but cannot quantify how often these patterns occur, nor can it definitively determine whether chatbots are, on average, beneficial or harmful. It serves as a crucial starting point for further quantitative research.
The researchers concluded that the perceived benefits described by both women – easy access, anonymity, and perceived empathy – may offer immediate, short-term relief. However, these advantages could inadvertently mask longer-term risks such as overreliance, blurred boundaries, and overlooked privacy and data security concerns. The study underscores that the appeal of chatbots is not merely a testament to AI’s capabilities but also a stark indicator of real and pressing gaps in conventional mental healthcare. These include limited access to personalized information and education about one’s condition, pervasive social isolation, and an erosion of trust in psychiatric care due to systemic shortcomings. The researchers firmly asserted that AI alone cannot adequately address these complex, multifaceted problems.
Recommendations for Patients, Families, and Clinicians
In light of these findings, the study offered practical recommendations:
- For Patients and Families: It is crucial that if a chatbot conversation leads to doubts about a diagnosis, medication, or treatment plan, these questions are immediately brought to the care team before making any changes. Similarly, any crisis management plan drafted with AI assistance should be treated as a preliminary draft and thoroughly reviewed with a clinician. This emphasizes the need for human oversight and professional validation for critical health decisions.
- For Clinicians: Proactively asking patients whether and how they use AI chatbots can open a valuable dialogue. This conversation can not only help understand what the chatbot offers them but also shed light on potential unmet needs or perceived deficiencies in their current care, allowing clinicians to adapt and improve their support strategies.
The integration of AI into mental health care is an irreversible trend. While offering unprecedented accessibility and personalized support, tools like ChatGPT also present complex ethical dilemmas and potential risks, particularly for vulnerable populations. This Dutch study serves as a vital early warning, urging the mental health community to approach AI integration with caution, robust ethical frameworks, and a renewed commitment to addressing the systemic shortcomings that drive individuals to seek solace and guidance in the digital realm. The challenge now lies in harnessing AI’s potential while safeguarding the well-being and trust of those it aims to serve.
