SODP Learning Collaborative on Recognizing, Documenting, and Responding the Patient AI Use: Week One

The Society of Digital Psychiatry (SODP) has concluded the first week of our learning collaborative on recognizing, documenting, and responding to patient AI use. We will be sharing updates week-by-week with SODP membership so they can follow along with the ongoing discussion!

Purpose:

Generative AI use is becoming a clinically meaningful factor of many patients' experience, in both beneficial and maladaptive ways. Patients are increasingly using AI as a substitute or supplement for mental health care, and around ⅓ of clinicians report encountering patients using AI for mental health support. However, there is no established protocol for recognizing when patient AI use is causing harm, nor is there protocol for appropriate response or intervention. Developing those protocols requires an understanding how the structural tendencies of LLMs (such as sycophantic affirmation of users) have the potential to interact with psychiatric vulnerabilities and produce pathological effects. 

Given the nascent state of research, the SODP has brought together mental health professionals from a wide range of backgrounds and cultural contexts to explore these challenges. Alongside a weekly discussion, participants complete a free, online training on patient AI use, available at ai-digital-literacy.net. 

Discussion Summary:

This week’s conversation started off with the question: How do the structural tendencies of LLMs interact with psychiatric vulnerabilities?

Participants raised concerns related to the developmental effect of Artificial Intelligence, particularly for younger people and those using AI chatbots as companions. Specific concerns were raised around identity development, as the social substitution offered by Artificial Intelligence could deepen social isolation while simultaneously failing to offer the developmentally beneficial friction of human interaction; AI could offer a “good enough” form of interaction that shields vulnerable individuals from having to engage with difficult but developmentally productive socialization. Further, LLM use could end up acting as a harmful substitute for more effective coping techniques; for example, a patient with GAD may repeatedly reach for an LLM for maladaptive reassurance, deepening anxious thought loops rather than practicing techniques to learn how to sit with discomfort.

There was also substantive discussion around the issue of disclosure of AI use to clinicians. Concerns were raised around stigma as a barrier to disclosure, especially amongst peer groups or from the clinicians themselves; worry about external judgment by peers, parents, or clinicians might prevent important discussion about a patient’s AI use. As one participant noted, many patients with SMI already hide many parts of their lives from those around them, and AI use is not exclusive in that respect. This presents a challenge to clinicians, as AI use may be clinically relevant but never surfaces over the course of treatment. Participants highlighted the need for asking about AI use as a part of standard practice, especially during intake procedures.

Finally, the group discussed why patients may turn to AI rather than traditional mental health care. The 24/7 convenience of a chatbot, combined with its low demand on users as compared to what a clinician may ask of a patient, makes it an attractive option for patients supplementing or supplanting care. The group also pointed out how the prevalence of emotional support use cases for AI reveals structural gaps in mental health care systems. Further, although anthropomorphism is a likely tendency in interacting with AI, the very fact that AI is not human may make it feel like a safer space for sensitive disclosure than with peers, family, or clinicians. Other participants pointed out how AI use may present significant benefits across a neurodiverse population beyond just emotional support, particularly in managing executive dysfunction.

Related Readings:

Keshavan, M., Torous, J., & Yassin, W. (2026). Do generative AI chatbots increase psychosis risk? World Psychiatry, 25(1), 150–151. https://doi.org/10.1002/wps.70017
 
Hudon, A., & Stip, E. (2025). Delusional Experiences Emerging From AI Chatbot Interactions or "AI Psychosis". JMIR mental health, 12, e85799. https://doi.org/10.2196/85799 

Dohnány, S., Kurth-Nelson, Z., Spens, E., Luettgau, L., Reid, A., Gabriel, I., Summerfield, C., Shanahan, M., & Nour, M. M. (2026). Technological folie à deux: feedback loops between AI chatbots and mental health. Nature. Mental health, 4(3), 336–345. https://doi.org/10.1038/s44220-026-00595-8 

Siddals, S., Torous, J., & Coxon, A. (2024). "It happened to be the perfect thing": experiences of generative AI chatbots for mental health. Npj mental health research, 3(1), 48. https://doi.org/10.1038/s44184-024-00097-4 

Collaborative Facilitators:

Please reach out to our facilitators if you are interested in learning more!

Mason Granof, Clinical Research Coordinator
The Division of Digital Psychiatry at BIDMC and Harvard Medical School
mgranof@bidmc.harvard.edu

Kathryn Ledley, SODP Co-Leader & Clinical Research Assistant
The Division of Digital Psychiatry at BIDMC and Harvard Medical Schoolkledley@bidmc.harvard.edu