By
Reviewed byIn short
The future of AI in therapy is likely to involve human therapists supported by AI, stronger evidence for narrow uses, clearer oversight, and care that blends app-based support with human sessions. Expect voice and multimodal tools, more personalization, and AI handling lighter between-session work, while diagnosis, trauma, and crisis stay human. Treat AI as a supplement to professional care, not a substitute, and never rely on an app in a crisis. These are forecasts about how care may develop. Clinical trials establish findings about particular tools, and surveys describe current use; neither guarantees how quickly a model of care will spread.
Augmentation, not replacement, is the real trajectory
The clearest signal about the future of AI in therapy is that the technology is being absorbed into existing care rather than overturning it. The near-term shift is augmentation: therapists using AI for notes, scheduling, homework, and progress tracking so they spend more of each session on the human part of the work.
For a patient, preparation can be useful before any new technology enters care. A social anxiety test can help describe fears about interpersonal situations, and grounding techniques offer a practical exercise to discuss with your clinician. Bring your questions and observations to appointments so the care plan reflects what is happening outside the session.
For patients, this looks like app-based support between appointments that loops back to a clinician. The lighter end of mental-health support (psychoeducation, habit-building, mild everyday stress, and skills practice) is where tools fit best and where adoption is moving fastest. In the American Psychological Association's 2026 survey of more than 1,200 licensed psychologists, 77% said their patients report using AI. The strongest near-term role for AI may sit even earlier in the journey: helping someone name what they are dealing with, understand their options, and arrive at a clinician sooner and better prepared. That percentage describes psychologists reporting patient use, rather than the share of all patients or Americans who use AI.
If you want the direct answer to whether AI will take therapists' jobs, read our companion piece on whether AI will replace therapists.
The future I want is not AI as the therapist. It is AI as the thing that finds you the right one, faster, and supports the work between sessions. Front door, not final destination.
Evidence is getting better, and narrower
Today the research base is thin and uneven. Some studies suggest chatbot-delivered CBT can ease mild symptoms of anxiety and low mood for some people, but most tools are not regulated medical devices and many marketing claims run ahead of the data. Reviews of large language models in psychotherapy published in 2024 and 2025 map the same territory from the technical side: rapid capability growth and promising support use cases, next to an evaluation and safety gap that has to close before anything clinical is on the table.
The likely direction is more rigorous, condition-specific evidence rather than broad claims that AI treats mental illness. Expect AI therapy studies that test narrow, well-defined uses such as guided CBT exercises for mild anxiety, structured journaling support, or relapse-prevention check-ins, where outcomes can actually be measured. Dartmouth's Therabot trial, published in NEJM AI in 2025 as the first randomized controlled trial of a generative AI support chatbot, is the template: one purpose-built tool, one defined population, measured outcomes.
The honest read is that the field will earn credibility one specific use case at a time. Tools that publish real outcomes and define what they do and do not treat will pull ahead of those that lean on vague wellness language. For patients, that means the question shifts from does AI therapy work in general to does this tool work for this problem.
The Therabot trial randomized 210 adults to the study chatbot or a waitlist control, with a primary assessment after 4 weeks. Because it did not compare the chatbot directly with therapist-delivered care, the result cannot establish equivalence to a licensed professional. Source: Heinz and colleagues, NEJM AI, 2025.
Regulation is maturing
For most of the category's history, AI mental-health tools have operated in a gray zone, often positioned as wellness products to sidestep medical-device oversight. That is changing. Regulators and professional bodies are paying closer attention, and Illinois restricted independent AI provision of therapy under its 2025 Wellness and Oversight for Psychological Resources Act. The law permits specified administrative and supplementary uses under a licensed professional's responsibility, with additional requirements for supplementary support. It should not be summarized as a ban on every use of AI in a clinical practice.
Expect clearer best practices for AI chatbots in therapy over the next few years: what a tool may claim, how it must handle crisis and risk, how it discloses that it is not a licensed clinician, and how it treats sensitive emotional data. Privacy is central here, because these tools collect some of the most personal information people share, and current practices vary widely.
This maturing is good for patients even if it slows some launches. Guardrails around crisis handling, honest marketing, and data protection are exactly what a high-trust mental-health category needs. The tools that treat regulation as a floor to build on, rather than an obstacle, are the ones likely to last.
Hybrid care becomes the default
The most likely shape of the future is hybrid: human therapy with AI woven around it. A typical model looks like scheduled sessions with a licensed clinician, supported by app-based tools that reinforce skills, prompt practice, and track mood between visits, with that data feeding back to the therapist.
This division of labor plays to each side's strengths. Software is well suited to consistency, availability, and structured exercises delivered the same way every time. Humans remain essential for diagnosis-level work, trauma, relationships, serious mental illness, and anything involving risk.
Hybrid care also has the potential to widen access, and the access gap is large: HRSA's State of the Behavioral Health Workforce report states that 137 million US residents lived in a designated mental health professional shortage area as of December 2, 2025. This dated estimate describes access pressures; it does not show that software can meet the same needs as additional clinicians. Where there are long waitlists or no nearby clinician, structured AI support can be a reasonable bridge to human care. The win is more people getting some support, with clearer paths to a human when they need one.
Multimodal, voice, and more personalization
On the product side, what ships next is about how people interact. Voice and phone AI support interfaces are spreading, letting people talk out loud instead of only typing, which can feel closer to a real conversation for some users. Multimodal tools that combine text, voice, and tracking are likely to become more common.
Personalization is the other big change. Tools are moving toward adapting to your history, your goals, and patterns in how you feel over time, instead of starting cold every session. Done well, that means more relevant prompts and exercises. Done poorly, it raises real questions about data use and how much an app should infer about your mental state.
Here the line between likely and hype matters most. Better voice and personalization are plausible and already underway. Claims that AI will deliver a genuine therapeutic relationship, read you the way a clinician does, or safely manage crises are not supported and should be treated with skepticism. Newer interface features do not change the structural limits on judgment, accountability, and real human connection.
What it means for patients and therapists
If you are seeking help, the practical guidance does not change much as the technology advances. Use AI tools as a supplement, not a substitute, especially for anything beyond everyday stress. Treat them as skills practice and between-session support, check the privacy policy before sharing sensitive details, and see a licensed human for trauma, ongoing low mood, relationship work, or any risk to your safety. No AI tool diagnoses, treats, or cures mental illness, and none are crisis services. Our guide on how to use AI as a therapist covers safe setup.
If you are a therapist, expect the biggest changes in your workflow. AI will reshape how you handle notes, homework, and between-session contact long before it touches the relational, judgment-heavy core of the work, which is the part software is worst at. Learning to use these tools well, and to advise clients on which ones are credible, is becoming part of the role. The clinicians who thrive will be the ones who fold AI in without overselling what it can safely do.
Questions to ask before AI enters your care
Ask your clinician what the software does, what information it receives, and who checks its output. If it records sessions or creates summaries, discuss consent, correction of errors, retention, and whether the information leaves the practice. Clarify whether you can decline the feature and continue receiving care. A clear explanation is more useful than a broad assurance that the system is secure.
Between appointments, establish who sees app messages and when. A mood log that nobody reviews should not be mistaken for monitoring. Agree on a separate route for urgent concerns, and judge usefulness by changes that matter in daily life, such as practicing an agreed skill or arriving at appointments better prepared. More messages alone do not demonstrate better care.
