Why Chatting with AI Feels Easier than Talking with a Therapist: Social Psychology Explanations for the Rise of Therapy Chatbots

Artificial intelligence (AI) is the new great technology of our time, and it is impacting every field of work, including mental health. Increasingly, people are turning to AI for mental health support. Hundreds of “therapy” chatbots can be found online and on mobile applications. In this unprecedented time of technological change, therapists should remain informed of these trends, since AI usage and related dilemmas concerning ethics will only continue to increase.

Emerging research highlights both the benefits and harms of using AI technology as a pseudo-therapist. As with any powerful tool, the possibilities and risk factors are substantial. But the question remains: why would people prefer to chat with a bot instead of connecting with a human therapist? In this article, we will employ several theories from the field of social psychology to examine the factors driving people toward AI therapy and the ethics and dangers that accompany it.

Social Exchange Theory

The first theory from social psychology that we will examine is social exchange theory, which suggests that the value of our social interactions can be measured by subtracting the costs from the benefits (profit = benefits – costs) (Homans, 1958). It posits that our goal is to seek social situations, just like in economic transactions, that have the lowest costs and highest benefits. While social costs and benefits vary across individuals, for many people experiencing distress, AI therapy represents a lower-cost approach than psychotherapy conducted by human therapists. This includes lower logistical costs, since AI is nearly free, requires no commuting or schedule interruptions, and is available 24/7. Perhaps more importantly, AI represents lower emotional costs, since it is private and does not require vulnerability with another person. Unless prompted to push back, AI rarely induces discomfort, providing high degrees of acceptance and affirmation beyond what most humans would offer in a healthy therapeutic alliance.

However, individual users appreciating the benefits of AI technology may fail to consider the ethics and costs of AI, such as receiving inaccurate or misleading information, sidestepping core issues, feeling affirmed in problematic decisions, or even reinforcing self-harm or suicidal intentions (Chatterjee, 2025). Combined, these factors can lead to the perception that AI therapy is preferable compared to a therapeutic alliance with a human clinician. As of last year, 13% of all adults and 22% of 18–21-year-olds use AI for mental health advice, and nearly all (97%) found AI’s advice helpful (McBain et al., 2025). Social exchange theory can help therapists respond effectively to the increasing use of AI for mental health support by shifting conversations to the overlooked costs of AI and the overlooked benefits of therapy with human clinicians. Informed individuals can then conduct more accurate cost-benefit analyses. 

Self-Determination Theory

Reasons within the realm of social psychology for why people use therapy chatbots include influences on human motivation (Hudson, 1981). People tend to be motivated by their needs for competence, autonomy, and relatedness (Ryan & Deci, 2000). Self-determination theory (SDT) helps clarify how motivation driven by external factors and outcomes (extrinsic) and motivation driven by the enjoyment of an activity for its own sake (intrinsic) may support or limit our ability to feel competent, autonomous, and connected to others (Ryan & Deci, 2000). Contexts that support these three needs are more likely to lead to increased intrinsically motivated actions and overall well-being (Deci & Ryan, 2000). 

The tenets of SDT help us understand why our clients, or potential clients, turn so readily to large language models (LLMs) to address their emotional and mental distress. AI interactions can help individuals meet all three basic needs, although important caveats related to ethics remain (Bergdahl et al., 2023). For example, during a chat with AI, an individual may feel enhanced personal autonomy because they were able to seek help on their own schedule and in the privacy of their home. While they may feel positive about taking action to address their distress, this sense of autonomy may be inaccurate because it is difficult to ascertain the influence of AI when the “interaction” is more constrained and aseptic than any real-world interaction.

Additionally, individuals can experience an inaccurate sense of competence when using AI technology to access vast information online about mental health considerations. While this information can be helpful, such instant access to information may foster dependence rather than competence. A more important downside, of course, is that AI does not have the clinical experience of a therapist to sift through and provide the most relevant and accurate information to each client.

A good therapist has a strong foundational understanding of theoretical approaches, evidence-based interventions, and a network of other professionals to consult with as needed. For example, a recent client experienced diminished trust in their partner after watching a TikTok video on the topic of partner forgetfulness. The client’s subsequent chat on Claude validated the erosion of trust. As a clinician, my role is to consider information in the context of my client’s cultural factors, past experiences, goals, values, our therapeutic alliance, and their non-verbal communication, among other factors in determining how to best help my client explore their concerns. Despite access to vast information, AI cannot form this type of collaborative and iterative therapeutic relationship to offer therapy uniquely tailored to the individual.

Nevertheless, AI technology provides a veneer of relatedness through the familiar and conversational style programmed into most AI chatbots. Programmed text outputs denoting relatedness may feel like connection, mistaken for true intimacy. Many years of psychotherapy research have confirmed that a therapeutic alliance promotes client improvement (Stubbe, 2018). Despite AI algorithms using words denoting a friendly tone, there is no actual alliance. While individuals may feel an increase in relatedness, typing and then receiving machine-generated responses can reinforce superficiality. Using AI to meet our need for relatedness and connection can lead to what Dr. Dan Wendler has described as “rabbit starvation” (Wendler, 2026). While eating exclusively rabbit meat can immediately reduce the sensation of hunger and meet some of our nutritional needs, it is a low-quality protein, missing fat and other important nutrients, so ultimately, this diet leads to malnutrition. Meeting our need for relatedness primarily through interactions with AI technology is much the same and may result in increased loneliness and isolation.

Echo Chambers

The dangers of using AI technology for therapy go beyond inaccurate perceptions, short-term reasoning, psychological/intellectual dependency, and emotional malnutrition. One of the most enticing factors driving people to turn to chatbots is agreeableness to the point of sycophancy, which can result in adverse consequences (Yeung et al., 2025). To better understand the psychological effects of AI agreeableness, we can learn from another theory from social psychology: balance theory (Heider, 1946), which posits that we prefer to be around people who agree with us. When we feel down, having someone tell us that our opinion is correct not only feels affirming, but it also bonds us emotionally to the source of validation. This can draw people in, especially when they are emotionally vulnerable.

For example, Yeung et al. (2025) measured how AI would respond to a user telling it about a delusion. The researchers were looking to see if AI would challenge or encourage delusion. The research findings were unnerving: not only did the chatbots tend to perpetuate delusion, but they also offered safety interventions in only 40% of cases. Since AI cannot discern fact from fiction, it conveys the same confidence for both. Combined with a friendly tone, this can “supercharge vulnerabilities” in users who may tend to have difficulty discerning the truth (Griesser, 2025). 

AI sycophancy raises serious concerns about ethics as it has the danger of harming people who are vulnerable to a variety of maladaptive habits and beliefs. It can create an echo chamber of one, in which untrue beliefs are amplified instead of challenged until they get out of hand. There are multiple cases of this happening already (Wei, 2025). For example, people who use avoidance to seek relief from anxiety often experience short-term relief. While AI may encourage this behavior for its immediate merits, therapists can recognize this as a pattern that maintains anxiety and can challenge clients to decrease this behavior to seek long-term benefits.

People generally present to therapy because of longstanding habits and beliefs that don’t serve them well. Therapists are better equipped than AI to push back and encourage clients to change these harmful ways of being in addition to providing validation. Additionally, learning to repair relational ruptures, navigate disagreement, and tolerate the ambiguity of human relationships leads to essential client growth.

Conclusion

As with other tools clients may access, AI supports for mental health and therapy have their place. There are many explanations within the realm of social psychology for why more and more people are turning to AI for mental health supports. The perceived emotional and financial costs are smaller than talking to a human therapist. With access to online information, AI can give genuinely helpful advice. Further, users feel more control when using AI than when talking with a human. That being said, there are real dangers to using AI tools without the guidance of a professional therapist.

Therefore, therapists need to leverage their therapeutic alliance and openly inquire about and discuss AI use with their clients. They also need to inform them directly about AI limitations, including attachment bonds with AI personas. While we should not avoid AI because of its shortcomings, therapists need to take steps to support clients and the public. Psychotherapists may consider including questions about AI use in their intake paperwork, creating online posts and public commentaries, and advocating for state legislation regulating AI-generated mental health advice. APA (2024) and state licensing boards (e.g., Utah Department, 2025) have published professional/ethical guidelines, but these guidelines lack specifics and are being outpaced by the rapid development of AI. Many practitioners remain uninformed of specific complexities and may inadvertently contribute to harm, such as by overlooking adverse consequences of clients using AI as a replacement for human clinicians. At the very least, clinics and training programs need to specify and enforce policies to protect clients. Finding specific ways to mitigate the harm and maximize the benefits of AI solutions, without adhering to the narratives of the corporations that benefit financially from AI, is needed—and needed now.