By
Reviewed byIn short
AI is applied across therapy and counseling in three broad layers. Client-facing tools include chatbots, self-help apps, and between-session support. Clinician-facing tools handle documentation and notes, training and supervision, treatment-planning support, and screening or triage. System-level tools improve access, scheduling, and analytics. Each application has real benefits and real limits, and none of them diagnose, treat, or cure mental illness or replace a licensed clinician.
How AI is applied across therapy and counseling
AI is applied in therapy and counseling at three levels: tools that face the client, tools that support the clinician, and tools that run the wider system. The clearest way to understand the field is to ask who the tool actually serves. Some applications face the client directly. Some sit behind the scenes and support the clinician. And some operate at the level of a whole practice or health system, shaping who gets care and how smoothly it runs.
Each application needs a specific purpose and a clear limit, because useful automation can also introduce errors, privacy problems, or missed warning signs. None of these tools is a licensed therapist, and none of them diagnose, treat, or cure mental-health conditions. If you are in crisis or thinking about suicide, call or text 988 in the US to reach the Suicide and Crisis Lifeline, available 24 hours a day.
For someone preparing to seek care, the Stress Level Test can help describe current strain, while the SMART Goals Worksheet can organize a practical goal to bring to an appointment. These are self-reflection resources. An intake chatbot or a screening score should lead to a conversation with a professional when symptoms affect daily life.
The most valuable uses of AI in this field are the quiet ones: notes, scheduling, screening, helping a clinician do more of what only they can do. The flashy use case, replacing the therapist, is the least valuable and the most risky.
Client-facing applications
Chatbots and conversational agents are the most visible application. The American Psychological Association's Chatbots and Mental Health Survey reports psychologists' observations of patients using AI; it is a practitioner survey, so its findings should not be presented as the share of all US adults using chatbots. These tools let someone talk through a worry, practice a coping skill, or simply feel heard at any hour, and a randomized trial of Dartmouth's Therabot reported symptom improvements with a research chatbot under clinician oversight. The benefit is round-the-clock availability and a low barrier to starting. The limit is that a chatbot does not truly understand context, can miss risk, and is not a crisis service, so it should point users toward human help when stakes are high. To go deeper, see our overview of AI psychotherapy and our safe-use checklist of best practices for AI chatbots in therapy.
Self-help apps package structured techniques from approaches like cognitive behavioral therapy and dialectical behavior therapy into exercises, mood tracking, and guided lessons. They offer affordable, repeatable skill-building that someone can do at their own pace. Their weakness is that adherence is hard to sustain and most apps are not regulated medical devices, so quality varies widely.
Between-session support is a growing use case where an AI tool reinforces what happens in therapy by sending reminders, prompting practice, or offering a place to vent until the next appointment. The benefit is continuity, since change often happens between sessions rather than during them. The limit is that without clinician oversight the support can drift away from the actual treatment plan.
Clinician-facing applications: documentation and planning
Documentation and note-taking is one of the fastest-growing applications. In the APA's 2025 Practitioner Pulse Survey, 56% of respondents reported having used AI to assist their work at least once. That measures adoption, rather than evidence that any particular use improves care. AI tools can draft progress notes from a session, summarize key themes, and reduce the paperwork that drives clinician burnout. The benefit is time given back to the clinician and, by extension, to clients. The limit is accuracy and privacy: a draft can contain errors or hallucinated detail, so a clinician must review every note, and sensitive recordings demand strong data protection.
Treatment-planning support uses AI to suggest evidence-based interventions, surface relevant measures, or organize assessment data into a coherent picture. The benefit is a second set of eyes that can catch options a busy clinician might overlook. The limit is that these suggestions reflect their training data, can carry bias, and are no substitute for clinical judgment about a specific person.
Screening and triage tools use AI to flag symptom severity, sort intake forms, or route a new client toward the right level of care. The benefit is faster, more consistent intake and earlier identification of people who need urgent attention. The limit is that a false negative can be serious, so screening should inform a clinician's decision rather than replace it.
Before a session is recorded or transcribed, ask what is being captured, where it will be stored, who can access it, and whether it will be used to train a model. The clinician should explain the available alternatives and review the draft against the actual session. Incorrect names, invented symptoms, and omitted safety concerns need correction before a note becomes part of the record.
Clinician-facing applications: training and supervision
Training and supervision is a quieter but meaningful application. AI can power simulated clients that let trainees practice difficult conversations in a safe setting, or analyze recorded sessions to give feedback on things like talk ratio, reflective listening, or use of a specific technique. The benefit is more practice and more objective feedback than a supervisor alone can provide.
The limit is that simulated practice is not the same as sitting with a real person in distress, and automated feedback measures what is easy to count rather than the harder-to-define qualities of a strong therapeutic relationship. These tools work best as a supplement to human supervision, since leaning on them alone can narrow training toward whatever the model happens to score.
System-level applications
At the level of a practice or health system, AI is applied to the logistics of care rather than the care itself. Scheduling tools predict no-shows, fill cancellations, and match clients to clinicians with the right specialty or availability. The payoff is fuller calendars and shorter waits. The risk is that optimizing for throughput can quietly deprioritize complex cases that take longer.
Access is another system-level application. AI-assisted directories, intake bots, and triage lines can help more people find appropriate care, including in places with few providers. The best evidence so far comes from England's NHS Talking Therapies: in a study of 129,400 patients across 28 services, the services that added an AI-enabled self-referral tool saw referrals rise by 15 percent, compared with 6 percent in matched control services, and the increases were larger among groups that are usually underserved. The benefit is reach into underserved areas. The limit is the digital divide, since the people who most need care are sometimes the least able to use these tools.
Analytics close the loop by aggregating outcome measures across a caseload to show what is working and where someone may be stalling. The benefit is data-informed care and earlier course correction. The limit is that outcome data is noisy and easy to misread, so it should prompt a conversation rather than dictate a decision. For where all of this is heading, see our look at the future of AI in therapy.
Weighing benefits against limits
Across all three layers, AI is strongest at the repetitive, scalable, and measurable parts of therapy and counseling: availability, paperwork, scheduling, screening, and tracking. It is weakest at the parts that depend on genuine understanding, relationship, and judgment, which remain the heart of effective care. What AI therapy studies actually show is covered in our research roundup, and for a condensed side-by-side, see our summary of AI therapy pros and cons.
The practical takeaway is to treat each application as a tool with a job, not as a stand-in for a clinician. A documentation assistant should save time without replacing clinical review. A chatbot should support a person without pretending to treat them. Within those boundaries, AI can widen access and ease burden; treating it as a replacement introduces real risk. If you plan to try these tools yourself, read our practical guide on how to use AI as a therapist first. And if you want to work with a person rather than a tool, browse licensed therapists and counselors in our directory.
Choosing an application by the problem it solves
For a client, start with a bounded task such as summarizing questions for an appointment or practicing an agreed exercise. For a clinician, begin with a workflow whose output can be checked, such as formatting an existing note. For a practice, check whether automated scheduling or triage changes who gets access to care, including people who need language assistance or a non-digital route.
Ask who owns the final decision and what happens when the software fails. A useful pilot tracks corrections, missed referrals, complaints, and the burden of checking outputs. Faster completion alone cannot establish that the application is appropriate for sensitive clinical work. The World Health Organization's guidance identifies human autonomy, transparency, accountability, and equity as central considerations.
