Safer Use of Chatbots
for Mental Health
On This Page
About
This guide is for Canadian adults who use, or are considering using AI chatbots for emotional support, life advice, self-reflection, companionship, or mental-health-related questions.
Drawing on research available as of May 2026, it covers where chatbot use may become problematic, where it may be potentially beneficial, when to pause or stop, and how to use these tools more safely.
Research on chatbots and mental health is still new, and there is still a lot we do not know.1,2, 27 Many findings come from early studies and need to be repeated by other researchers before we can be confident in them. The risks and benefits will become clearer as we learn more.1,2 Where newer or formal guidance exists, readers should follow that guidance.
This guide was written by a clinical psychologist and a data scientist, with review and contributions from psychology, psychiatry, and social work professionals, and a member of the public.
Note on Chatbots
In this guide, we use “chatbots” as a broad term for AI tools that can generate conversational responses. We are mainly referring to three types of tools:
General-purpose chatbots like ChatGPT and Claude which use GenAI technology that can help with many different tasks.
Companion chatbots which may use GenAI technology, and are intended to simulate a relationship.
Wellness chatbots which may use GenAI technology and are intended to support stress, emotional well-being, self-reflection, or related concerns, like Woebot.
GenAI means generative artificial intelligence: AI systems that can create new content, such as text, images, audio, or other outputs, based on patterns learned from data.
Appendix: Current Limitations in AI Safety and Performance
Current artificial intelligence systems still have important limitations that affect their safety and performance.
Using earlier details
The system may miss, forget, or give too little weight to important details shared earlier (a disability, a death of a loved one).10, 13
Handling long conversations
Long or complicated exchanges can make the output less reliable and may be linked to riskier use.10,,16, 27
The system may become less reliable or be linked to riskier use during long or complex back-and-forth exchanges.10,,16, 27
Responding consistently
The response may change from one message to the next.11, 10
Making things up
The system may produce false or made-up information while sounding confident.1,2
Over-validating
The response may support the user’s point of view too strongly.6,17,18,19
Missing the full life picture
The system does not have reliable access to the person’s full history, relationships, culture, risks, values, or support system.2, 13
Detecting risk
The system may fail to notice or flag signs, especially subtle signs, that human support, professional care, or urgent help is needed.10,13,20l
References & Disclaimer
References
Open references
1. Torous, J., Linardon, J., Goldberg, S. B., Sun, S., Bell, I., Nicholas, J., Hassan, L., Hua, Y., Milton, A., & Firth, J. (2025). The evolving field of digital mental health: current evidence and implementation issues for smartphone apps, generative artificial intelligence, and virtual reality. World Psychiatry, 24(2), 156–174. https://doi.org/10.1002/wps.21299
2. American Psychological Association. Health advisory on AI chatbots and wellness apps. https://www.apa.org/topics/artificial-intelligence-machine-learning/health-advisory-chatbots-wellness-apps
3. Mental Health Research Canada — National Survey 25 (n=4,666). August 2025.
4. Zhang, Y., Zhao, D., Hancock, J. T., Kraut, R., & Yang, D. (2025). The rise of AI companions: how human-chatbot relationships influence well-being. arXiv preprint arXiv:2506.12605.
5. Fang, C. M., Liu, A. R., Danry, V., Lee, E., Chan, S. W., Pataranutaporn, P., ... & Agarwal, S. (2025). How AIai and human behaviors shape psychosocial effects of extended chatbot use: A longitudinal randomized controlled study. arXiv preprint arXiv:2503.17473.
6. Cheng, M., Lee, C., Khadpe, P., Yu, S., Han, D., & Jurafsky, D. (2026). Sycophantic AI decreases prosocial intentions and promotes dependence. Science, 391(6792). https://doi.org/10.1126/science.aec8352
7. Dohnány, S., Kurth-Nelson, Z., Spens, E., Luettgau, L., Reid, A., Gabriel, I., ... & Nour, M. M. (2026). Technological folie à deux: feedback loops between AI chatbots and mental health. Nature Mental Health, 1-10.
8. Saracini, C., Cornejo-Plaza, M. I., & Cippitani, R. (2025). Techno-emotional projection in human–GenAI relationships: A psychological and ethical conceptual perspective. Frontiers in Psychology, 16, 1662206.
9. Boyd, R. L., & Markowitz, D. M. (2026). Artificial intelligence and the psychology of human connection. Perspectives on Psychological Science, 21(2), 192-220.
10. Cheng, Y., Kang, Z., Jiang, K. (2026). The slow drift of support: Boundary failures in multi-turn mental health LLM dialogues.10.48550/arXiv.2601.14269.
11. Chen, L., Zaharia, M., & Zou, J. (2024). How is ChatGPT’s behavior changing over time?. Harvard Data Science Review, 6(2). Chen et al. — ChatGPT performance changes over time. arXiv:2307.09009, 2023.
12. Chen, L., Zaharia, M., & Zou, J. (2023). How is ChatGPT's behavior changing over time?. 10.48550/arXiv.2307.09009.
13. Pichowicz, W., Kotas, M., & Piotrowski, P. (2025). Performance of mental health chatbot agents in detecting and managing suicidal ideation. Scientific Reports, 15(1), 31652.
14. Moore, J., Grabb, D., Agnew, W., Klyman, K., Chancellor, S., Ong, D. C., & Haber, N. (2025). Expressing stigma and inappropriate responses prevents LLMs from safely replacing mental health providers. In Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency (FAccT '25). Association for Computing Machinery, New York, NY, USA, 599–627. https://doi.org/10.1145/3715275.3732039
15. Badawi et al. / Priyadarshana et al. — Comparison of 9 s on 4,500 mental health responses rated by psychiatric evaluators; meaningful safety differences found across models. arXiv:2601.18630, 2026.
16. Du, Y., Tian, M., Ronanki, S., Rongali, S., Babu Bodapati, S., Galstyan, A., Wells, A., Schwartz, R., Eliu A Huerta, E. A., & Peng, H. (2025). Context length alone hurts LLM performance despite perfect retrieval. Findings of the Association for Computational Linguistics: EMNLP, 23281–23298. https://doi.org/10.18653/v1/2025.findings-emnlp.1264
17. Noshin, K., Ahmed, S. I., & Sultana, S. (2026). AI Sycophancy: How Users Flag and Respond. arXiv preprint arXiv:2601.10467.Noshin et al. — AI sycophancy: how users flag and respond. arXiv:2601.10467, 2026.
18. OpenAI. (2025, April 29). Sycophancy in GPT-4o: What happened and what we’re doing about it. OpenAI. https://openai.com/index/sycophancy-in-gpt-4o/
19. Malmqvist, Lars. (2024). Sycophancy in large language models: Causes and mitigations. 10.48550/arXiv.2411.15287.
20. Ramaswamy, A., Tyagi, A., Hugo, H. et al. ChatGPT Health performance in a structured test of triage recommendations. Nat Med 32, 1671–1675. https://doi.org/10.1038/s41591-026-04297-7
21. Hudon, A., & Stip, E. (2025). Delusional experiences emerging from AI chatbot interactions or “AI Psychosis”. JMIR Mental Health, 12(1), e85799.
22. Harvard Gazette. What to make of AI psychosis. https://news.harvard.edu/gazette/story/2026/04/what-to-make-of-ai-psychosis/
23. Golden, A., & Aboujaoude, E. (2026). A transdiagnostic model for how general purpose AI chatbots can perpetuate OCD and anxiety disorders. npj Digital Medicine.
24. Rector, N. A., Katz, D.E., Quilty, L.C., Laposa, J.M., Collimore, K., & Kay, T. (2019). Reassurance seeking in the anxiety disorders and OCD: Construct validation, clinical correlates and CBT treatment response. Journal of Anxiety Disorders, 67. doi: 10.1016/j.janxdis.2019.102109. Epub 2019 Jun 22. PMID: 31430610.Rector et al. — Reassurance seeking is a common maintaining factor across anxiety disorders; its reduction in CBT is associated with improved clinical outcomes. Journal of Anxiety Disorders, 2019.
25. Sun, X., Wang, Y., & McDaniel, B. T., (2026). AI companions and adolescent social relationships: Benefits, risks, and bidirectional influences. Child Development Perspectives. https://pmc.ncbi.nlm.nih.gov/articles/PMC12928748/
26. Palaniyappan, L., Paquin, V., & Barou-Laforie, E. (2026). High-Risk Human–AI Engagement: Clinical Assessment and Management Considerations. The Canadian Journal of Psychiatry, 07067437261445770.
27. Keshavan, M., Torous, J., & Yassin, W. (2026). Do generative AI chatbots increase psychosis risk?. World Psychiatry, 25(1), 150.
28. Tseng, E., & A Liang, C. (2026, April). " Chat, Should I Leave Him?" Risks, Rewards, and Roles for AI in Relationship Advice. In Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (pp. 1-19).
29. Bean, A.M., Payne, R.E., Parsons, G. et al. Reliability of LLMs as medical assistants for the general public: a randomized preregistered study. Nat Med 32, 609–615. https://doi.org/10.1038/s41591-025-04074-y
30. Bodner, R., Lim, K., Siddals, S., Goldberg, S., & Torous, J. (2026). Barriers to understanding how many people use AI for mental health support: an estimate and narrative review. npj Digital Public Health, 1(1), 21.
For the cited version of this guide, contact katherinegibb@pm.me.
Important note
Open disclaimers
- This guide provides general educational information for Canadians aged 18 and older. Children and adolescents may face different risks and require separate guidance and appropriate adult support. It is not a substitute for individualized assessment, diagnosis, treatment, psychotherapy, professional advice, or crisis support, and should not be considered clinical, legal, or technical advice. Information may not reflect the latest clinical, technical, or security developments. Prefer newer, reliable resources when available, and refer to the latest version of this guide at therapistintheloop.org or katherinegibb.ca. No method described can guarantee complete privacy or safety. This guide is educational and general only and intended for adults (18+). Children and adolescents may face different risks and need separate guidance and appropriate adult support. It should not be considered as personalized therapy, clinical or technical advice, or crisis support. It does not replace a licensed professional. For emergencies, call 9-1-1.
- The information in this guide may not reflect the most recent technical updates or clinical findings, and does not include all considerations pertaining to the use of AI for mental health. This guide uses the best available information, including peer-reviewed studies, early research that may not yet have been peer reviewed, theoretical papers, and relevant guidance. Some sources are preliminary because AI tools are changing faster than research can fully evaluate them.
- Authors and contributors disclose relevant affiliations in the interest of transparency. Mention of any product or service does not imply endorsement. Their involvement reflects their contribution to the development or review of this guide and does not necessarily imply affiliation, sponsorship, or endorsement by Therapist in the Loop, their organizations, or any product or service mentioned.
Human Support
If you need human support, reach out.
9-8-8 Suicide Crisis Helpline
Call or text 9-8-8 · 24/7 · Canada-wide
988.caKids Help Phone
Call 1-800-668-6868 or text CONNECT to 686868 · Ages 5-29 · 24/7
kidshelpphone.caFind a Helpline
Directory of Canadian and international crisis lines
findahelpline.com