Supervised by Denis Persichini.

Introduction

The effectiveness of artificial intelligence (AI) chatbots is becoming an increasingly relevant topic within psychology and psychotherapy research. It is particularly important for young people, who may interact frequently with AI while also experiencing mental health difficulties. The key concepts discussed throughout this research paper include: referrals from AI-based support to human therapy and how these referrals affect help-seeking; the severity of mental health problems and its influence on young people’s preference for AI-based or human support; and the role of self-stigma, which refers to the internalisation of negative beliefs about seeking psychological help.

Existing research suggests that AI-based mental health support and human therapy may complement one another rather than functioning as completely separate forms of support. However, it remains unclear how this relationship will develop in the future, particularly as the use of AI continues to increase. Understanding this is important because it may provide a foundation for future research into how AI-based mental health support could influence help-seeking behaviour among young people.

The research question addressed in this article is: “how does AI-based mental health support, compared with human therapy, affect young people’s trust and willingness to seek help?”. Based on the evidence reviewed, it can be argued that AI-based mental health support may influence young people’s trust and willingness to seek help through the way referrals to human support are communicated, the role of stigma and the perceived severity of their mental health difficulties. However, this relationship may also depend on other factors, such as accessibility, and the available evidence does not necessarily demonstrate a causal relationship.

1. How Referrals from AI Support to Human Therapy May Shape Trust and Help-seeking

1.1 AI Referrals: Support or Dismissal?

AI-based mental health support provides young people with an accessible way to discuss personal concerns and receive immediate responses. Although AI cannot experience human empathy, its responses may still be perceived as supportive and non-judgemental. Kuhail et al. (2025) found that mental health professionals struggled to distinguish short AI conversations from human therapy transcripts and rated the AI conversations as being of higher quality on average. However, trust may also depend on how AI responds when professional care is needed. A referral to a human therapist may increase trust if it demonstrates that the AI recognises its own limitations, takes the user’s concerns seriously and prioritises their safety. In contrast, an impersonal or poorly explained referral may make the young person feel dismissed. Therefore, when AI chatbots refer young people to human therapists, the way in which the referral is communicated may influence whether users feel dismissed or reassured by the support they have received. This, in turn, may affect their willingness to continue seeking help.

1.2 Referrals as a Supportive Bridge to Human Therapy

An AI referral may increase trust when young people interpret it as a sign that the system recognises its limitations and takes their safety seriously. Participants in a study by Petersson et al. (2025) believed that AI could act as a guide or digital companion, helping them navigate the help-seeking process and connect with a healthcare professional when necessary. Kretzschmar et al. (2019) help to explain why this transition may be useful, suggesting that the privacy and accessibility of chatbots may make it easier for young people to discuss mental health concerns and gradually become more open to professional support. Together, these studies suggest that AI and human therapy do not always need to be viewed as competing forms of support. If a referral acknowledges the young person’s feelings, explains why professional care is needed and provides a clear next step, AI may function as a supportive bridge to human therapy. However, both studies mainly examine young people’s perspectives rather than their actual help-seeking behaviour, meaning that they cannot establish whether AI referrals actually increase trust or lead users to engage with professional therapy.

1.3 When Referrals May Feel Dismissive

An AI referral may reduce trust when it is delivered as an abrupt or generic response that does not acknowledge the user’s feelings or circumstances. Tang et al. (2026) found that some adult users experienced AI refusals and recommendations to seek professional help as invalidating or dismissive, particularly when the suggested services were difficult to access because of privacy concerns or limited availability. Some participants felt that these responses ended the support abruptly instead of helping them move towards human care. Similarly, Ni and Yang (2025) propose that abruptly ending an emotionally sensitive conversation may create feelings of rejection or shame and potentially discourage further help-seeking. They suggest that AI should instead acknowledge the user’s emotions, explain its limitations, and provide a gradual and supportive transition to appropriate care.

Together, these sources suggest that the referral itself may not be the main problem; rather, its effect may depend on how it is communicated and whether the recommended support is realistically accessible. A poorly explained referral could therefore weaken young people’s trust in AI and reduce their willingness to continue seeking help. However, Tang et al. (2026) studied adults rather than young people, while Ni and Yang (2025) present a conceptual framework rather than direct empirical evidence. Therefore, these possible effects on young people still require further research. Nevertheless, the available evidence suggests that referrals may be experienced as dismissive rather than supportive when they are communicated without emotional acknowledgement, a clear explanation or an accessible next step.

1.4 The Dual Role of Referrals

Taken together, the evidence suggests that referrals from AI to human therapy can play two contrasting roles in the help-seeking process. Petersson et al. (2025) and Kretzschmar et al. (2019) present AI as an accessible first step that may help users become more open to professional support. From this perspective, a referral may strengthen trust by showing that the system recognises its limitations and takes the user’s needs seriously. In contrast, Tang et al. (2026) found that some users experienced referrals as dismissive when they were generic, sudden or disconnected from the user’s lived circumstances. Ni and Yang (2025) similarly suggest that ending an emotionally sensitive conversation without a smooth and supportive transition may create feelings of rejection and discourage further help-seeking.

Kuhail et al. (2025) add a further layer to this picture, as therapists in their study could not reliably distinguish AI-generated dialogue from human therapy and often rated it as higher quality. This suggests that qualities such as warmth and tone may influence whether an interaction feels supportive, although the study did not specifically examine referrals.

Therefore, the referral itself cannot be considered inherently supportive or dismissive. Its effect may depend on whether the user experiences it as continued guidance or as the end of support. A well-communicated referral may maintain trust in AI while encouraging engagement with human therapy, whereas a poorly communicated one may weaken trust in both forms of support. However, none of these studies directly measured whether an AI referral changed users’ actual help-seeking behaviour. Instead, they relied on self-reported perceptions, personal accounts, expert judgement or theoretical reasoning. The relationship proposed here should therefore be viewed as plausible rather than established.

Overall, these findings suggest that comparisons between AI and human therapy should also consider how effectively users are supported during the transition between the two forms of care.

1.5 Implications for Trust and Help-Seeking

Overall, AI-based mental health support and human therapy should not always be viewed as competing forms of care. AI may provide a more accessible and less intimidating first step, while human therapists can offer professional judgement and a deeper understanding of complex emotional concerns. Referrals may connect these two forms of support, but their effect on young people’s trust and willingness to seek help may depend on how the transition is experienced.

A supportive referral that acknowledges the young person’s feelings, explains the need for professional care and provides an accessible next step may strengthen trust and encourage further help-seeking. In contrast, a sudden or impersonal referral may be perceived as dismissive and could discourage the user from seeking further support. Therefore, the comparison between AI and human therapy is not only about which form of support young people prefer, but also about how effectively AI can guide users towards appropriate human care when additional support is needed.

This suggests that AI systems may benefit from being designed to acknowledge users’ emotions, explain their limitations clearly and provide supportive and practical guidance during transitions to human care, rather than ending support abruptly. However, existing research has not directly measured young people’s behaviour following AI referrals. Future studies should therefore track actual help-seeking outcomes and engagement with professional services, rather than relying on self-reported perceptions alone, to better understand the real-world effects of AI referrals on trust and willingness to seek help.

2. Young People’s Preference for AI or Human Therapy Depending on the Severity of Mental Health Problems

2.1 AI as an accessible option for more severe issues

Mental health chatbots have the potential to serve as an accessible entry point to mental health support for young people. Many factors can contribute to young people not seeking professional support, including stigma, low awareness and limited availability of services. Koulouri et al. (2022) conducted a study of the acceptability of mental health chatbots among young adults and found that stigma, limited awareness and resources, and “unfavourable perceptions of mental health chatbots” were significant barriers to engagement with mental health services. The researchers concluded that mental health chatbots were generally an “acceptable mental health intervention for young adults,” although several limitations and reservations were also identified.

This suggests that chatbots may provide a useful option for young people who are reluctant or unable to access professional support immediately. For someone experiencing day-to-day stress or feelings of isolation, interacting with an easily accessible chatbot may feel more manageable than arranging an appointment with a therapist. However, it is important to note that Koulouri et al. (2022) did not specifically examine whether preferences differed according to the severity of young people’s mental health concerns. The researchers only briefly noted that young adults had “mixed perceptions of mental health chatbots”. Therefore, while the study highlights the potential accessibility of chatbots, it does not confirm that young people with less severe concerns would actually prefer them over traditional mental health services.

A similar point can be made regarding the findings of Rackoff et al. (2025). The researchers surveyed 428 university students in the United States about their attitudes towards traditional mental health services and chatbot-delivered services. The study found that although 49% of participants had used a chatbot for some purpose, only 5% had used one specifically for mental health reasons. Furthermore, participants reported significantly more positive attitudes towards traditional services than towards chatbot services, although cost, time and stigma were perceived as fewer barriers to chatbot use. As with Koulouri et al. (2022), these findings suggest that chatbots may offer a more accessible first step for some young people, but they do not indicate that young people necessarily view chatbots as a satisfactory alternative to professional support across different levels of severity.

Taken together, both studies highlight the potential value of mental health chatbots as an accessible first step towards engaging with mental health support. However, they also suggest that accessibility alone is unlikely to make chatbots the preferred option overall, particularly when compared with traditional services. Rackoff et al. (2025), for example, found that participants generally held more positive attitudes towards traditional mental health services despite perceiving chatbots as involving fewer barriers such as cost, time and stigma. This suggests that reducing barriers may increase the appeal of chatbot-based support without necessarily replacing young people’s preference for human professional care.

2.2 Why human support might be preferred for more severe issues

The value of easily accessible but potentially less comprehensive services such as chatbots may differ when young people experience more severe mental health concerns that require greater levels of support. Traditional human-led services can offer professional support that young people may view as more appropriate for a serious situation.

Rackoff et al. (2025) found that although the majority of participants had positive attitudes towards both types of service, those attitudes were significantly more positive for traditional services. Furthermore, only 5% of participants had ever used a chatbot for mental health reasons, and 8% of those with probable depression or generalised anxiety disorder reported doing so. The researchers note that the use of chatbots for mental health purposes “was rare overall and did not vary greatly across levels of need”. This suggests that greater mental health need was not clearly associated with greater or lower use of chatbots within this sample. It is worth noting that Rackoff et al. (2025) do not establish a causal link between severity and preference for traditional services over chatbots. The study used a cross-sectional design and did not directly test whether increasing symptom severity caused changes in service preference. In particular, participants who scored high on measures of probable depression or generalised anxiety disorder may not have differed significantly from those who did not in terms of their attitudes towards mental health chatbots. The researchers only note that participants with higher levels of probable depression or generalised anxiety disorder had “marginally lower negative attitudes toward mental health chatbots than their counterparts”.

This finding highlights an important distinction regarding the research question. The reviewed evidence suggests that young people generally report more positive attitudes towards traditional human-led services than towards chatbots. However, this does not necessarily mean that preference for chatbots decreases as problems become more severe, because none of the reviewed studies directly examined the relationship between symptom severity and preference for one form of support over the other. Therefore, it is possible that, for some young people, convenience and accessibility may remain important even when their concerns are relatively severe. Similarly, some young people may greatly value the anonymity and reduced stigma associated with chatbots, which may allow them to engage with mental health support even when experiencing more serious difficulties. Severity may therefore interact with other factors, rather than independently determining young people’s preference for chatbot or human support.

It is also important to note that Koulouri et al. (2022) specifically recommend considering mental health chatbots as “an integral part of the broader mental health ecosystem”. The researchers note that the mental health chatbot landscape is “dichotomised” and that the field requires “more rigorous exploration of the efficacy of mental health chatbots in a comprehensive mental health service system”. This point highlights another potential distinction when addressing the research question. Rather than viewing chatbots and human services as alternatives, chatbots may form one part of a broader pathway to mental health support. For less severe or initial concerns, mental health chatbots may represent an attractive option because of their convenience, accessibility and potentially reduced stigma. For more complex concerns, they may instead serve as an entry point or complement to professional human support.

3. Trust, Stigma and Preference for AI Services

Severity is unlikely to be the sole factor that shapes young people’s attitudes towards mental health chatbots and their relationship to traditional services. Some evidence suggests that young people’s choice between modalities may be informed by their perceptions of trust and stigma associated with each. AI services may provide an alternative for young people who feel embarrassed or uncomfortable about seeking help, whereas human services are often viewed as more beneficial overall.

Hoffman et al. (2024) conducted a study that directly examined young adults’ attitudes towards AI- and human-delivered psychotherapy, including the role of help-seeking self-stigma. The researchers note that mental health chatbots are “increasingly available and used for mental health support” but that little is known about “attitudes toward and preferences for” such services. The researchers measured young adults’ attitudes towards human- and AI-delivered psychotherapy and found that “greater self-stigma associated with human-delivered psychotherapy was related to more negative attitudes toward human psychotherapy and more positive attitudes toward AI psychotherapy”. Interestingly, the study also found that “greater stigma associated with AI psychotherapy was related to more negative attitudes toward AI psychotherapy and more positive attitudes toward human psychotherapy”.

This finding further complicates the relationship between severity and young people’s attitudes towards mental health chatbots. It shows that young people are unlikely to adopt such services simply due to convenience, particularly if they also hold negative perceptions of AI-based psychotherapy. On the contrary, Hoffman et al. (2024) suggest that young adults who are “particularly hesitant to pursue human psychotherapy due to self-stigma” may be more receptive to alternatives such as mental health chatbots.

This conclusion can be further applied to the findings of Rackoff et al. (2025). The researchers found that young people generally viewed traditional services as more beneficial but were significantly more likely to report cost, time and stigma as barriers to accessing such services than to chatbot-based support. Thus, for young people with serious mental health concerns, the combination of perceived benefits and barriers may be particularly important when deciding whether and where to seek support. However, if such barriers can be addressed, those young people may find that the benefits of traditional services make them more appealing than chatbots. On the other hand, young people who experience stigma and inconvenience as significant barriers to mental health support may be more receptive to chatbots because of their greater convenience and potentially lower perceived stigma, even when experiencing more serious concerns.

It is important to note that this interpretation assumes that young people would be willing to use mental health chatbots for serious concerns. However, none of the reviewed studies establish a causal link between stigma and choice of service, and none specifically examined the role that severity plays in young people’s attitudes towards mental health chatbots. In particular, Hoffman et al. (2024) note that their research design was cross-sectional and that therefore “the associations found in this study should be interpreted as correlational rather than causal”. Similarly, while the researchers identified several factors that contribute to young people’s attitudes towards chatbots and traditional services, those attitudes do not necessarily translate into actual behaviour. Rackoff et al. (2025) highlight this point by showing that young people generally reported more positive attitudes towards traditional mental health services, while actual use of mental health chatbots remained low. Almost half of their participants reported using a chatbot for some purpose, but few had used a chatbot specifically for mental health reasons.

Thus, the reviewed evidence suggests that mental health chatbots may provide a convenient and potentially less stigmatising option for some young people. However, young people may also value the personal and professional support provided by human therapists, particularly when they perceive their concerns as requiring more comprehensive support. Overall, the evidence suggests that traditional services are generally viewed more positively than mental health chatbots, while chatbots may offer an accessible alternative or entry point for individuals who experience barriers to human support. Importantly, the evidence does not establish that chatbots are used only for minor concerns or that young people necessarily prefer human therapy as severity increases. None of the reviewed studies specifically examined the effect of severity on young people’s attitudes towards mental health chatbots, highlighting the need for further research directly examining how symptom severity interacts with stigma, accessibility, trust and service preference.

3.1 Young people use AI chatbot therapy due to human therapy accessibility issues

Young adults may turn to AI chatbots for mental health support partly because of the stigma associated with traditional therapy, including both public stigma and self-stigma. Therapy can require money, trust and time, and some people may hesitate to invest in these resources when they are uncertain about whether treatment will be helpful. Hoffman et al. (2024) describe self-stigma as the internalisation of negative attitudes or beliefs about seeking psychological help. Their research focuses on young adults and suggests that self-stigma can influence attitudes towards both AI-based and human-delivered psychotherapy.

This is important because even when human therapy is available, young people may still avoid seeking help if they feel embarrassed, uncomfortable or concerned about being judged. AI chatbots may make the first step towards seeking support feel easier because they can offer a more private and potentially less socially threatening environment in which users can discuss personal concerns. Unlike interactions with another person, users may perceive an AI chatbot as less likely to judge them, which may reduce some of the discomfort associated with disclosing sensitive information.

This shows why accessibility is not only about whether a therapist is physically available. Self-stigma can create a psychological barrier that prevents people from seeking support even when services are accessible. For young adults experiencing this type of stigma, an AI chatbot may therefore provide a more private way to begin discussing their concerns and potentially reduce some of the barriers that prevent them from seeking help. However, this does not mean that AI chatbots provide the same quality or depth of care as a human therapist. Rather, they may offer an additional point of access for people who might otherwise avoid seeking any form of mental health support.

3.2 Stigma behind AI chatbots

AI, however, can also be associated with mistrust and negative perceptions, as some people may question whether it can be used responsibly, whether it is secure or whether their privacy will be protected. These concerns are also reflected in research examining the acceptance of AI chatbots. Dergaa et al. (2023) explain that misconceptions and uncertainty surrounding how AI works can contribute to negative attitudes towards its use, particularly in situations involving private or sensitive information. One concern is that AI may produce inaccurate information, potentially leading users to misinterpret their symptoms or rely on inappropriate advice. The authors also discuss concerns relating to data privacy, plagiarism and other forms of misuse of AI-generated information. These concerns may reduce trust in AI, even when the technology could potentially provide useful support.

Privacy is particularly important when considering the use of AI for mental health support. Someone discussing their mental health may be sharing information that they would normally disclose only to a trusted person, meaning that trust in how an AI system handles personal information becomes especially important. Dergaa et al. (2023) explain that privacy concerns can contribute to negative attitudes towards AI because these systems may rely on large amounts of data and users may not always understand how their information is collected, processed or protected. If someone is uncertain about how their personal information is being used, they may be less willing to trust an AI chatbot with sensitive concerns. Overall, Dergaa et al. (2023) suggest that improving people’s understanding of AI may help reduce misconceptions and support more informed attitudes towards its use.

3.3 AI chatbots as an accessible first step

Despite stigma, AI chatbot therapy has shown potential in addressing some barriers that human therapy may not always overcome as easily, particularly by reaching people more quickly and encouraging them to seek support. AI chatbots can give people, especially young people around the age of 17-25, more immediate and potentially lower-cost access to mental health support. However, these forms of support also have drawbacks, including the possibility of providing inaccurate or overly simplified information and lacking genuine human empathy.

It is also important to consider that the experience of interacting with AI itself may influence how users think about mental health and psychological support. Song et al. (2025) specifically examined the impact of human-AI cooperation on mental illness stigma. Their research provides another way of understanding AI chatbot therapy because AI may do more than simply provide information. The interaction itself may also influence users’ attitudes towards mental illness and help-seeking. This is particularly important when considering young people who may already feel uncomfortable discussing their mental health. A chatbot creates a different type of interaction than sitting face-to-face with another person. The user may feel that they have more control over the conversation because they can decide what they want to type and how much information they want to provide. This could make starting a conversation about a difficult or sensitive subject feel easier.

However, this does not mean that AI can provide everything that a human therapist can. The lack of genuine human empathy is an important difference. An AI chatbot can produce a response that sounds understanding, but that response is still generated by technology rather than based on lived emotional understanding. In some cases, AI may also provide responses that appear overly agreeable or simply reflect what the user wants to hear. This is one reason AI could potentially be used to help people begin talking about their mental health without being treated as a replacement for human therapists.

3.4 Stigma behind therapy: how AI gets people past

AI psychotherapy may help some young people take an initial step towards seeking mental health support, which could eventually make engagement with human therapy feel more manageable. As previously discussed, Hoffman et al. (2024) found that young adults could hold positive attitudes towards AI-based psychotherapy while having more negative attitudes towards human psychotherapy. Rather than showing that these attitudes naturally change over time, however, their findings suggest that AI-based support may offer an alternative starting point for individuals who feel uncomfortable with traditional therapy.

Hoffman et al. (2024) specifically connect this relationship to self-stigma, which can occur when negative social attitudes towards seeking psychological help are internalised and begin to affect how individuals view themselves. Their research found that young adults with higher self-stigma associated with human psychotherapy tended to report more negative attitudes towards human therapy and more positive attitudes towards AI psychotherapy. This is important because it suggests that AI may provide another option for someone who feels uncomfortable with the idea of traditional therapy.

Instead of immediately having to sit down with another person and discuss personal concerns, an AI chatbot may provide a more private and less intimidating place to begin. For some users, this could make discussing mental health concerns feel more manageable and potentially support later engagement with professional care. However, this does not mean that AI should replace human therapists. AI may instead serve as an easier first interaction with mental health support, while human therapy can still provide the professional knowledge, personal connection and deeper level of care that AI may not be capable of providing.

Conclusion

Overall, the evidence reviewed suggests that AI-based mental health support and human therapy are not necessarily competing options, but may interact in ways that either strengthen or undermine young people’s trust and willingness to seek help. AI chatbots may offer an accessible and potentially lower-stigma first step for young people who avoid seeking help because of cost, time or embarrassment, while human therapy is generally viewed more positively and may be perceived as more appropriate when greater professional support is needed.

This relationship may depend on how AI referrals to human care are communicated, the stigma associated with each form of support and the perceived severity of the problem. For example, a supportive and well-communicated referral may build trust, whereas an abrupt or impersonal referral may feel dismissive. These effects are also likely to vary according to individual preferences, levels of stigma, service accessibility and the nature of the mental health concern.

However, the current evidence does not allow us to conclude that AI support causally changes trust or actual help-seeking behaviour, as much of the research relies on self-reported perceptions rather than directly measured behavioural outcomes. Taken together, the evidence suggests that AI may be most valuable as a complement to human care rather than as a replacement for it, particularly when it acknowledges users’ emotions, communicates its limitations clearly and supports an appropriate transition towards professional help when needed.

Dergaa, I., Fekih-Romdhane, F., Glenn, J. M., Saifeddin Fessi, M., Chamari, K., Dhahbi, W., Zghibi, M., Bragazzi, N. L., Ben Aissa, M., Guelmami, N., El Omri, A., Swed, S., Weiss, K., Knechtle, B., & Ben Saad, H. (2023). Moving beyond the stigma: Understanding and overcoming the resistance to the acceptance and adoption of artificial intelligence chatbots. New Asian Journal of Medicine, 1(2), 29–36.

Habeb, M., Ciobanu, A. M., Al-Ani, M., & Mottershead, R. (2025). Stigma in mental health: The status and future direction. Cureus, 17(6), e85398.

Hoffman, B. D., Oppert, M. L., & Owen, M. (2024). Understanding young adults’ attitudes towards using AI chatbots for psychotherapy: The role of self-stigma. Computers in Human Behavior: Artificial Humans, 2(2), 100086.

Koulouri, T., Macredie, R. D., & Olakitan, D. (2022). Chatbots to support young adults’ mental health: An exploratory study of acceptability. ACM Transactions on Interactive Intelligent Systems, 12(2).

Kretzschmar K, Tyroll H, Pavarini G, Manzini A, Singh I. (2019) Can Your Phone Be Your Therapist? Young People’s Ethical Perspectives on the Use of Fully Automated Conversational Agents (Chatbots) in Mental Health Support. Biomedical Informatics Insights, 11.

Kuhail, M. A., Alturki, N., Thomas, J., Alkhalifa, A. K., & Alshardan, A. (2025). Human-Human vs Human-AI Therapy: An Empirical Study. International Journal of Human–Computer Interaction, 41(11), 6841–6852.

Ni, Y., & Yang, T. (2025). “Even GPT can reject me”: Conceptualizing Abrupt Refusal Secondary Harm (ARSH) and reimagining psychological AI safety with Compassionate Completion Standard (CCS) [Preprint]. arXiv, 2512.18776.

Petersson, L., Ahlborg, M. G., & Häggström Westberg, K. (2025). “I believe that AI will recognize the problem before it happens”: Qualitative study exploring young adults’ perceptions of AI in mental health care. JMIR Mental Health, 12, e76973.

Rackoff, G. N., Zhang, Z. Z., & Newman, M. G. (2025). Chatbot-delivered mental health support: Attitudes and utilization in a sample of U.S. college students. Digital Health, 11.

Song, T., Jamieson, J., Zhu, T., Yamashita, N., & Lee, Y.-C. (2025). From interaction to attitude: Exploring the impact of human-AI cooperation on mental illness stigma. Proceedings of the ACM on Human-Computer Interaction, 9(2), 1–31.