Artificial Intelligence in Therapy? The Use of AI in Mental Health Services

The use of artificial intelligence in mental health services has been one of the hot topics in recent years. Some LLM-based models, such as Therabot and Woebot, have been specifically developed for this purpose and are capable of "mimicking" psychological treatment protocols. However, even these specialized models have not been able to obtain the necessary approvals for public use. At the same time, many people use general-purpose AI tools for these purposes.

APA’s Assessment of AI Tools for Mental Health Services

The APA’s comprehensive report on AI tools in March 2025 begins with an extensive explanation of the different types of artificial intelligence, or AI. In addition to these explanations, it refers to relevant papers examining the differences among these tools.

In general, the APA limits the use of AI in all processes of mental health services to several areas:

The use of diagnostic tools for the purpose of early detection in cases showing signs of the initial course of a disorder. (The professional enters the information and receives the results.)
Monitoring clients’ concerns and issues between sessions (recording and responding to immediate concerns—for example, recording information about the intensity and frequency of anxious thoughts while providing an immediate practical strategy).
Recording and analyzing biomarker data (information collected by wearable devices such as smartwatches).
Case studies for students or recent graduates (the supervisor provides information about a real case, and AI plays the role of the client), which is purely educational.

Finally, the APA warns about ethical concerns. The stated applications can only be carried out using tools specifically approved by the APA, and the use of general-purpose tools is discouraged. Also, as with all psychiatric and psychological records, at least one, up to three professionals (depending on the protocol) must review and approve the analysis and presentation of information.

The APA then republishes notices from the WHO (World Health Organization) and the EU (European Union) and emphasizes the need for FDA approval.

Ultimately, it is emphasized that, to date, no tool has received the necessary approvals for public or widespread use.

Evidence on use of AI in Mental Health Services

Olawade et al. (2024), in a review, evaluated current trends and future prospects for the use of AI in improving mental health. The paper, noting the significant role of AI in the advancement of various sciences as well as its widespread applications for general users, considers its use in mental health services to be an inevitable part of the future. A total of 211 studies published between 2019 and 2023 were initially included in the analysis: 87 articles were excluded because they were not in English, and 32 were excluded because they did not include study measures, leaving 92 articles as the basis of the review. Overall, these articles examined the effects of 20 AI tools.

Ultimately, Olawade et al. reported relatively high accuracy for these tools in early diagnosis, predicting the course of disorders, and serving as supportive tools in treatment. As a supportive tool, AI is capable of providing precise personalized methods that can substantially increase recovery rates.

Analytical tools developed based on psychological data can serve as therapists’ assistants and provide a great deal of information that would certainly be useful to therapists. However, this analysis does not recommend using AI tools—even specialized tools in this field—as the primary therapist. These tools effectively address many of the challenges associated with accessing psychological treatment, such as stigma, limited access, cost, and so on, but they still cannot be used without professional supervision.

Thakkar et al. (2024), by providing detailed information about the history, types, and benefits of AI, thoroughly examine the capabilities of these tools. They subsequently propose two primary roles that are consistent with the current capabilities of these tools: increasing awareness and providing therapeutic support. Based on their current capabilities, various tools are highly useful and applicable for increasing public awareness of mental health.

In addition, by removing many of the limitations of traditional treatment, these tools are among the best options for therapeutic support. The paper also explains well why these tools have strong capabilities for early diagnosis. It further reports significant advances in specialized mental-health tools in understanding and analyzing emotions and affect. However, using these tools for direct intervention requires further development.

Thakkar et al. subsequently report ethical concerns, concerns related to information leak or the misuse of information by these tools, problems with information validity, the inability to collect additional information or violations of ethical frameworks in this process, and potential cultural biases. Ultimately, Thakkar et al. describe the development and advancement of specialized mental-health tools as highly useful and promising and, with the goal of improving quality, recommend that the necessary research and clinical trials be conducted more rapidly.

Dehbozorgi et al. (2025), in a meta-analysis examining nearly 3,000 papers, evaluated the applications of AI in the field of mental health. In addition to the applications examined in the studies discussed above, this paper reports significant advances in specialized large language models (LLMs) for psychological intervention. The paper also introduces a scale for evaluating AI in intervention, which can guide future studies. One section of the paper addresses the challenge of acceptance.

The issue is that even when a tool performs correctly, users may show lower acceptance because of its machine-based nature. Ultimately, Dehbozorgi et al., through their evaluation, examine many of the previous challenges associated with the use of AI and introduce new challenges for future investigation.

• 
  
  Conclusion: Can AI Be Useful in Mental Health Services?

Based on the APA report, as well as the research and review studies conducted in this area, the use of general-purpose AI tools, particularly chatbots, for mental health services raises numerous problems, including a lack of validity in results and substantial bias in responses. However, specialized AI tools developed in recent years are clearly highly useful tools for professionals in predicting the course of illness, early diagnosis, and providing therapeutic support and between-session monitoring. Nevertheless, their use by members of the general public without professional supervision is not recommended. The study by Dehbozorgi et al. points toward the development of specialized tools that may be suitable for use by general users in the future.