Understanding AI use in mental health practice.
Generative AI is becoming part of how some people seek mental health information, emotional support, and conversation.
This interactive resource helps ADMH students and early-career addictions and mental health workers explore emerging evidence, consider important practice and ethical issues, and approach conversations about AI use with curiosity and professional judgment.
Educational resource: This guide does not assess, diagnose, determine whether AI use is safe or unsafe, or replace professional judgment.
AI use is becoming part of the conversation.
Emerging ADMH professionals need opportunities to examine the evidence, limitations, ethical considerations, and implications for practice.
Explore the guide
Build your understanding
Explore six areas designed to support critical thinking about generative AI and its emerging place in addictions and mental health practice.
How Is AI Entering Mental Health Conversations?
Explore emerging research on how AI is entering mental health conversations, including evidence involving young adults, existing adult users, and practitioners.
Explore the evidence →
What Does the Evidence Say?
Examine potential usefulness, limitations, concerns, and areas where current evidence remains uncertain.
Review the research →
Understanding Different AI Tools
Distinguish general-purpose generative AI from purpose-built mental health chatbots and other digital mental health interventions.
Compare the tools →
Exploring AI Use With a Client
Explore person-centred, non-judgmental ways of opening conversations about a client's use of AI.
Explore conversation prompts →
Practice & Ethical Considerations
Consider privacy, misinformation, bias, AI literacy, over-reliance, professional boundaries, and the importance of human connection.
Consider the issues →
Resources for Emerging ADMH Professionals
Explore research, professional guidance, and practical resources for continued learning about AI and mental health practice.
View resources →
How Is AI Entering Mental Health Conversations?
Emerging research offers several windows into how AI is entering mental health conversations. Some evidence comes directly from young adults, while other studies examine existing adult AI users or practitioners' reports about their patients. These different sources help build context, but they should not be treated as if they measure the same population or answer the same question.
Young-adult perspective
25
young adults, ages 18–30
Petersson et al. (2025) interviewed young adults about their perceptions of how AI-based technology could be used in mental health care.
Existing adult users
270
adult users across 29 countries
Luo et al. (2025) studied adults who regularly used ChatGPT for emotional and mental health support. Their ages ranged from 18 to 67 years, so this was not a young-adult-only sample.
Practitioner reports
1,200+
U.S. licensed psychologists
APA's 2026 survey asked psychologists about AI use discussed by their patients. These findings describe psychologists' reports and are not estimates of AI use in the general population (American Psychological Association [APA], 2026).
A practice lens
Start with curiosity, not assumptions.
Knowing that a client uses AI does not by itself tell us why they are using it, what they find helpful or unhelpful, or how the interaction is affecting them. Research describes varied purposes and experiences, supporting exploration of the individual's experience rather than assuming that all AI use looks the same (Luo et al., 2025).
Uses identified in emerging research
What might someone be looking for?
Select one area at a time. These categories summarize patterns identified across the research. They are not a checklist for determining whether someone's AI use is appropriate or inappropriate.
Read the evidence carefully
These studies do not all answer the same question.
Each source provides a different view of an emerging landscape. Keeping those differences visible prevents us from treating unlike evidence as though it measured the same thing.
Young-adult perceptions
Petersson et al. (2025)
Explored how 25 young adults aged 18–30 perceived possible uses of AI-based technology in mental health care.
Existing adult users
Luo et al. (2025)
Examined 270 adults across 29 countries who already regularly used ChatGPT for emotional and mental health support.
Practitioner reports
APA (2026)
Surveyed more than 1,200 U.S. licensed psychologists about AI use that had come up with their existing patients.
These findings should not be combined into a single estimate of how many young adults use AI for mental health.
Apply it to practice
A client brings AI into the conversation.
CLIENT
“I've been talking to an AI chatbot about my anxiety because I don't really feel comfortable talking to anyone else about it.”
Pause & reflect
Before offering advice or correcting the client, what might you want to understand about their experience with AI?
What has the AI interaction been like for them?
What do they find helpful or unhelpful about using it?
What information or responses have they received?
What makes talking with another person feel less comfortable right now?
These prompts are reflective practice questions created for this educational guide. They are intended to support exploration and conversation and are not a validated assessment, risk-assessment instrument, or clinical decision-making tool.
Keep the tool in context
Conversational does not mean clinically equivalent.
Sobowale et al. (2025) evaluated five direct-to-consumer GenAI chatbots using youth personas and the CAPE-II evaluation framework.
High-quality ratings by evaluation area
86.7%
Accessibility
104 of 120 ratings
31%
Therapeutic approach
14 of 45 ratings
39%
Monitoring & risk evaluation
31 of 80 ratings
The chatbots scored strongly on accessibility but considerably lower on therapeutic approach and monitoring and risk evaluation. The researchers also identified concerns involving crisis handling, fabricated content, privacy transparency, and transparency about model training and knowledge (Sobowale et al., 2025).
Important: This study evaluated chatbot performance. It did not measure how common AI use is among young adults or why an individual young adult chooses to use AI.
Key takeaway
The fact that someone uses AI may be less informative than understanding why they are using it, what they are seeking from the interaction, what they are receiving from it, and how it is influencing their understanding, choices, and relationships.
This is a practice-oriented synthesis of the emerging evidence presented above rather than a clinical rule.
Evidence behind Section 01
References
References are presented in APA 7th edition style. Select “View source” to open the original publication.
American Psychological Association. (2026). Patients are bringing AI to therapy. https://www.apa.org/pubs/reports/chatbots-mental-health-2026
View source ↗Luo, X., Wang, Z., Tilley, J. L., Balarajan, S., Bassey, U.-A., & Cheang, C. I. (2025). Seeking emotional and mental health support from generative AI: Mixed-methods study of ChatGPT user experiences. JMIR Mental Health, 12, e77951. https://doi.org/10.2196/77951
View source ↗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. https://doi.org/10.2196/76973
View source ↗Sobowale, K., Humphrey, D. K., & Zhao, S. Y. (2025). Evaluating generative AI psychotherapy chatbots used by youth: Cross-sectional study. JMIR Mental Health, 12, e79838. https://doi.org/10.2196/79838
View source ↗✓ Section 01 complete
Next: examine what the broader evidence says about potential usefulness, limitations, and uncertainty.
What Does the Evidence Actually Tell Us?
Research findings can sound more certain when reduced to a headline. Evidence literacy requires examining what was actually studied, what was measured, and how far the findings can reasonably be applied.
Start with the claim
“Research shows that AI chatbots improve mental health.”
Before accepting that statement, what would you want to know? Select each question to look behind the headline.
Look behind the headline
What did Feng et al. (2025) actually study?
Feng et al. (2025) conducted a systematic review and meta-analysis of AI chatbot interventions for adolescents and young adults. Looking at the study details changes how broadly its findings can be interpreted.
Population
Ages 15–39
Participants across the included trials fell within this age range.
Evidence included
31 RCTs
29,637 participants were represented across the randomized controlled trials.
Meta-analysis
26 trials
Twenty-six trials provided sufficient data for quantitative synthesis.
Overall mental distress
SMD −0.35
A small-to-moderate pooled reduction in overall mental distress relative to control conditions.
Findings varied
Context mattered
Effects varied across participant groups, comparison conditions, deployment formats, and chatbot designs.
Important boundary
Generative evidence remains uncertain
Retrieval-based systems showed the most consistent effects, while overall effectiveness of generative systems remained inconclusive.
Interpretation boundary
The review provides evidence about the chatbot interventions included in its eligible trials. It does not establish that every AI chatbot, every generative AI system, or every use of AI for mental health will produce the same outcomes.
Claim vs. evidence challenge
Which conclusion stays closest to the study?
Research claim
“The review proves that generative AI is an effective mental health treatment.”
The evidence supports
A specific, qualified conclusion
Across the included Feng trials, chatbot interventions showed small-to-moderate effects in reducing overall mental distress, with effectiveness varying across study and intervention characteristics.
The evidence does not establish
A universal conclusion
• Every chatbot is effective.
• All AI systems have equivalent effects.
• Results automatically apply to general-purpose generative AI.
Key takeaway
A research finding is only as broad as the population, intervention, comparison, outcomes, and study conditions that produced it.
For emerging ADMH professionals, evidence literacy means asking what a study actually supports before applying a broad claim to a client, a technology, or mental health practice.
Evidence behind Section 02
Reference
Section 02 intentionally focuses on one systematic review and meta-analysis so the emphasis stays on how to interpret evidence.
Feng, X., Tian, L., Ho, G. W. K., Yorke, J., & Hui, V. (2025). The effectiveness of AI chatbots in alleviating mental distress and promoting health behaviors among adolescents and young adults: Systematic review and meta-analysis. Journal of Medical Internet Research, 27, e79850. https://doi.org/10.2196/79850
View source ↗✓ Section 02 complete
Next: understand why different AI tools and interventions should not be treated as though they are the same.
Same Label. Different Tools.
“AI chatbot” can describe systems with different purposes, architectures, and levels of evidence. AI literacy starts by identifying what kind of tool is actually being discussed before assuming that findings about one system apply to another.
Tool A
General-purpose GenAI
Built to handle many kinds of tasks and conversations. A person may choose to discuss mental health with it even though mental-health intervention is not its specific purpose.
Tool B
Purpose-built mental-health GenAI
Developed specifically around a mental-health or wellbeing purpose. Research interventions may incorporate defined therapeutic content, target populations, or structured delivery.
Tool C
Structured / rule-based chatbot
Uses predetermined rules, scripts, or conversational pathways rather than generating every response through a large language model.
The important distinction
All three may be described as chatbots, but they are not interchangeable. A tool’s intended purpose and its technical architecture answer different questions.
Start with two questions
Identify the tool before interpreting the evidence
Select the cards below. The first question is about intended purpose; the second is about how the system produces or selects responses.
Question 1
What was it designed for?
Question 2
How does it work?
Don’t mix the labels
Purpose and architecture are not the same thing
Two lenses
Ask both questions
What was the tool designed for? How does its conversational architecture work?
More structured
Rule-based approaches
A purpose-built mental-health tool can use constrained rules or scripts.
More open-ended
Generative / LLM approaches
Generative architecture can appear in purpose-built interventions or in broad general-purpose systems.
“Purpose-built” describes intended purpose. “Generative” describes something about how responses are produced.
One label does not automatically tell you the other.
Research spotlight
The technology is changing faster than the evidence base
Hua et al. (2025) systematically reviewed mental-health chatbot research from 2020–2024 and classified systems as rule-based, machine-learning based, or LLM-based. The review shows both rapid technological change and uneven levels of evaluation.
Evidence mapped
160 studies
Published from 2020–2024.
LLM growth
45%
Of studies published in 2024 used LLM-based architectures.
Clinical efficacy
16%
Of LLM-based studies were at the clinical-efficacy testing stage.
Interpretation boundary
Rapid growth in research attention does not mean that every architecture has reached the same level of clinical evaluation.
Purpose-built does not mean proven
What does the emerging GenAI intervention literature show?
Olisaeloka et al. (2026) mapped purpose-built generative-AI mental-health chatbot interventions. The review found a developing and heterogeneous field rather than one standardized type of intervention.
21 studies
Included after a systematic search of seven databases.
11 countries
Evidence came from multiple international settings.
Early & varied
Many interventions were early-stage and CBT-based, with varied target conditions and delivery features.
Being designed for mental health tells us about intended purpose. It does not automatically establish effectiveness, clinical validation, or suitability for a particular person.
Evidence transfer challenge
Can the evidence travel?
Study
Researchers test a purpose-built, CBT-informed mental-health chatbot with a defined participant group and report improvement in a measured outcome.
Someone concludes
“This means a person using a general-purpose generative AI chatbot for the same concern should experience the same benefit.”
Can we transfer the finding directly?
Key takeaway
“AI chatbot” is not one intervention. Before interpreting research or applying a finding, identify what the system was designed to do and how it works, then ask whether the technology in the evidence is actually comparable to the technology being discussed.
For emerging ADMH professionals, AI literacy does not require becoming a computer scientist. It means knowing enough about different tools to ask better questions about the evidence behind them.
Evidence behind Section 03
References
Hua, Y., Siddals, S., Ma, Z., Galatzer-Levy, I., Xia, W., Hau, C., Na, H., Flathers, M., Linardon, J., Ayubcha, C., & Torous, J. (2025). Charting the evolution of artificial intelligence mental health chatbots from rule-based systems to large language models: A systematic review. World Psychiatry, 24(3), 383–394. https://doi.org/10.1002/wps.21352
View source ↗Olisaeloka, L., Richardson, C. G., Wang, A. Y., Munthali, R. J., & Vigo, D. V. (2026). Generative AI mental health chatbots: A scoping review of intervention design and user experience. npj Digital Medicine. https://doi.org/10.1038/s41746-026-02972-0
View source ↗✓ Section 03 complete
Next: explore how AI use can enter a person-centred conversation without turning the guide into a diagnostic or assessment tool.
Keep the conversation open before trying to correct it.
When a client mentions using AI for mental health information, emotional support, or conversation, the first task is not to approve or condemn the technology. A person-centred response begins by understanding the client's experience and what the AI interaction means to them.
Practice focus
APA's 2026 practitioner guidance recommends proactively asking about AI use, exploring what patients find helpful or unhelpful, asking permission before sharing information or concerns, and leaving room for ongoing discussion.
A simple conversation flow
Ask → Explore → Reflect → Respond
This is a learning framework for practising person-centred communication. It is not a validated assessment, screening, or risk-assessment tool.
Ask
Create space for the person to tell you about their AI use without assuming what the experience has been.
Explore
Learn what they were looking for, what they received, and what they found helpful, unhelpful, or uncertain.
Reflect
Show that you have heard the meaning of the experience before moving into information, concerns, or advice.
Respond
When relevant, ask permission to share information or concerns and keep professional judgment in the conversation.
Practice scenario
“I've been talking to an AI chatbot when I'm anxious because sometimes it's easier than talking to people.”
The statement tells you that AI is part of the client's experience. It does not yet tell you whether the interaction has been helpful, unhelpful, accurate, influential, or what need it is meeting.
What would you say first?
Applying familiar ADMH skills
OARS can help keep the conversation person-centred.
APA does not present its AI guidance as motivational interviewing. Here, OARS is being applied as a familiar ADMH communication framework for practising curiosity, listening, and collaboration around a new topic.
O · Open questions
“What have those conversations been like for you?”
A · Affirmations
Acknowledge the person's willingness to talk openly about something they may have been unsure about sharing.
R · Reflections
“It sounds like having something available right away has been important to you.”
S · Summaries
“So it has helped you organize some of what you're feeling, and there have also been responses you weren't sure about.”
Before giving information
Ask permission.
After understanding more about the client's experience, there may be information or concerns that are relevant. APA's practitioner guidance recommends asking whether you can share that information rather than moving straight into a lecture or correction.
Which response best preserves collaboration?
What Section 04 is doing
Supporting the conversation.
The focus is how to invite disclosure, understand the person's experience, listen reflectively, and introduce information collaboratively.
What comes next
Practice and ethical considerations.
Privacy, inaccurate information, bias, over-reliance, crisis limitations, and other professional considerations belong in Section 05, where they can be examined directly rather than being crowded into this conversation exercise.
Key takeaway
The goal is not to approve or condemn a client's AI use before understanding it.
A person-centred conversation creates room to understand the client's experience, reflect what matters to them, and—when relevant—ask permission before sharing information or concerns. Professional judgment remains essential.
Reference
American Psychological Association. (2026, June 16). Your patients are using AI. Here's how to talk with them about it. https://www.apa.org/topics/artificial-intelligence-machine-learning/practitioners-talk-ai-patients.html
View source ↗✓ Section 04 complete
Next: examine privacy, misinformation, bias, over-reliance, crisis limitations, and other practice and ethical considerations.
A practice consideration is not an automatic conclusion.
When AI enters a mental-health conversation, several issues may deserve attention at the same time. Recognizing a concern does not, by itself, determine whether a client's AI use is safe or unsafe. It can help an emerging ADMH professional ask better questions, verify important information, and use professional judgment.
Practice considerations map
What might deserve attention?
Select a consideration to explore the evidence-informed question behind it. These are lenses for reflection, not a checklist that produces a clinical or safety verdict.
Research spotlight
Safety is more than a filter.
A 2026 systematic scoping review examined safety mechanisms in purpose-built generative-AI mental-health chatbot interventions.
21
included studies
11
countries
Layered
safeguards were needed
The review identified technical safeguards, co-design and privacy measures, role clarification, and crisis-response approaches. Human oversight was limited, crisis protocols were mostly underdeveloped, and systematic adverse-event monitoring was sparse. The authors argue for a broader sociotechnical approach rather than relying on a single technical safeguard.
Olisaeloka et al., 2026
Apply it · One situation, several considerations
Client
“I told the chatbot everything about what has been happening. It says I probably have borderline personality disorder. Honestly, it understands me better than most people do, so lately I just talk to it instead.”
What practice considerations can you notice without diagnosing the client or deciding whether their AI use is safe or unsafe?
Select all that apply.
Professional judgment
Recognize the issue without turning it into a verdict.
A privacy concern, questionable claim, or changing relationship with AI may deserve exploration. None of these observations alone diagnoses a condition or creates a universal safe/unsafe conclusion.
Practice boundary
AI does not replace professional responsibilities.
Organizational policy, supervision, scope of practice, documentation requirements, professional standards, and established crisis procedures continue to guide practice.
Key takeaway
AI-related practice considerations are interconnected.
Recognizing privacy, information quality, bias and context, reliance, or crisis limitations can help an emerging ADMH professional ask better questions while maintaining professional judgment, without turning AI use itself into a diagnosis or a simple safe/unsafe category.
References
American Psychological Association. (2025). Health advisory: Use of generative AI chatbots and wellness applications for mental health. https://www.apa.org/topics/artificial-intelligence-machine-learning/health-advisory-chatbots-wellness-apps
View source ↗Olisaeloka, L., Richardson, C. G., Wang, A. Y., Munthali, R. J., & Vigo, D. V. (2026). Safety mechanisms and risk mitigation in generative AI mental health chatbots: A systematic scoping review. Healthcare, 14(10), 1395. https://doi.org/10.3390/healthcare14101395
View source ↗✓ Section 05 complete
Next: explore professional guidance, research, and resources for continued learning about AI and mental health practice.
Know where to look when the answer is still developing.
AI technologies, evidence, and professional guidance continue to change. Emerging ADMH professionals do not need one source for every question. They need to know what kind of question they are asking and where reliable guidance may be found.
Resource navigator
Where would you look first?
Start with the question you are trying to answer. Select a pathway to see the type of source that may be most useful.
Before you rely on a resource
Four questions can help you place it in context.
01
Is it current?
AI technologies and guidance can change quickly. Check the publication or update date.
02
Who produced it?
Distinguish research, public-health guidance, professional standards, and commercial claims.
03
Who does it apply to?
Check the profession, population, jurisdiction, setting, and intended audience.
04
What kind of evidence is it?
A study, systematic review, advisory, policy, and organizational procedure answer different questions.
Professional learning
Use guidance in context.
When an AI-related question affects practice, consider the evidence alongside supervision, organizational procedures, privacy obligations, scope of practice, and any standards that apply to your role.
A changing field
Knowing the limit of a source is part of AI literacy.
A resource can be useful without answering every question. Look for what it covers, what it does not cover, and whether newer evidence or guidance changes the picture.
Final takeaway
AI literacy is an ongoing practice.
The evidence, technologies, and professional guidance surrounding AI will continue to develop. Emerging ADMH professionals do not need to know everything about AI. They do need to ask critical questions, recognize the limits of their knowledge, consult reliable sources, and use supervision, organizational guidance, and professional judgment when questions arise.
Resources referenced in this section
World Health Organization. (2026, March 20). Towards responsible AI for mental health and well-being: Experts chart a way forward. https://www.who.int/news/item/20-03-2026-towards-responsible-ai-for-mental-health-and-well-being--experts-chart-a-way-forward
View source ↗World Health Organization. (2025, March 25). Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models. https://www.who.int/publications/i/item/9789240084759
View source ↗Information and Privacy Commissioner of Ontario. (2026, February 19). Responsible use of AI & protecting client privacy in community counselling. https://www.ipc.on.ca/en/media-centre/presentation/responsible-use-ai-protecting-client-privacy-community-counselling
View source ↗American Psychological Association. (2026, June 16). Your patients are using AI. Here's how to talk with them about it. https://www.apa.org/topics/artificial-intelligence-machine-learning/practitioners-talk-ai-patients.html
View source ↗College of Physicians and Surgeons of Ontario. (2025, August). Using artificial intelligence in clinical practice. https://www.cpso.on.ca/physicians/policies-guidance/advice-to-the-profession/using-artificial-intelligence-in-clinical-practice
Profession-specific example for physicians.
View source ↗✓ Section 06 complete
You've completed the six-section practice guide.
Return to any section when you want to revisit the evidence, tools, conversation approaches, practice considerations, or professional resources.
Why this matters
AI is entering mental health conversations. Practice needs to understand what that means.
The goal of this guide is not to promote AI as mental health care or to tell professionals whether a client's AI use is safe or unsafe. Instead, it provides a space to explore emerging evidence, consider important questions, and strengthen informed, person-centred conversations.