By
Reviewed by"AI therapy" now spans everything from ChatGPT, Claude, and Gemini used informally for support to purpose-built tools like Ash, Wysa, and the research chatbot Therabot. Every figure below links to its source; the page is re-verified quarterly.
Last full review: July 2, 2026 · sourced from peer-reviewed research, APA national surveys, and federal datasets.
Ten numbers that define AI therapy in 2026
- 48.7% of AI users who report mental-health challenges have used a major LLM (ChatGPT, Claude, Gemini) for therapeutic support. Sentio University / Practice Innovations (APA), 2025
- About 1 in 6 U.S. adults use AI chatbots at least monthly for health information and advice. KFF, via Harvard Medicine, 2025
- OpenAI reports over 1 million people a week have ChatGPT conversations with explicit indicators of potential suicide planning. OpenAI, October 2025
- 35% of U.S. psychologists have patients using AI as an additional mental health professional. APA 2026 Chatbots & Mental Health Survey
- The first RCT of a generative AI therapy chatbot (Therabot) produced a 51% average reduction in depression symptoms. Heinz et al., NEJM AI, 2025
- Yet in independent testing, popular AI models responded inappropriately to mental-health symptoms at least 20% of the time. Moore et al., Stanford / ACM FAccT, 2025
- 94% of psychologists say chatbots cannot treat conditions with the appropriate level of nuance. APA, 2026
- 137 million Americans, 40% of the population, live in a Mental Health Professional Shortage Area. HRSA, December 2025
- The global AI mental health market is projected to grow from about $2.1B in 2026 to $9.1B by 2033. Grand View Research, 2026
- Illinois, Nevada, and Utah became the first states to restrict or ban AI therapy in 2025, with six more advancing bills by early 2026. KFF Health News; Pluribus News
How big the AI mental health market is
One of the fastest-growing segments in digital health. Independent firms converge on compound growth above 20% a year, driven by demand, workforce shortages, and rapid advances in large language models.
Global AI mental health market in 2026, on track for $9.1B by 2033
A 23.4% compound annual growth rate. Other analysts value the broader market at $1.99B in 2025, projecting up to $31.66B by 2035, among the steepest trajectories in all of digital health.
Grand View Research, 2026 · InsightAce Analytic, 2026
| Market segment | 2026 estimate | Projection | CAGR | Source |
|---|---|---|---|---|
| AI in mental health (global) | $2.11B | $9.12B by 2033 | 23.3% | Grand View Research |
| AI mental health chatbots | $767M | $7.8B by 2034 | 33.6% | Fortune Business Insights |
| Digital therapeutics (mental health) | $4.51B | $24.42B by 2035 | 20.6% | Towards Healthcare |
| Mental health apps (broad) | $9.94B | $22.73B by 2030 | 18% | Markets and Markets |
Who is building it
Slingshot AI launched Ash, marketed as "the first AI designed for therapy," in July 2025 with $93M raised (Series A led by a16z) and 50,000 beta users. As a counterpoint, Woebot Health raised over $123M but wound down its consumer chatbot in mid-2025 amid regulatory and commercial headwinds. More than 10,000 mental health apps now incorporate AI, up from fewer than 1,000 five years ago, and North America holds roughly 34% of the market.
How many people use AI for mental health
Adoption has accelerated sharply since 2023, and peer-reviewed research now documents it across age groups, conditions, and motivations, a technology that has moved from novelty to mainstream resource for millions of Americans.
Of people with a mental health condition who use AI now use it for therapeutic support
Nearly half of AI users who self-report an ongoing condition use major LLMs for support, leading researchers to suggest AI chatbots may collectively be the largest de facto mental health "provider" in the U.S.
Rousmaniere et al. (2025). Practice Innovations (APA). DOI: 10.1037/pri0000292
- 13.1%Adolescents & young adults (12-21) have used AI for mental-health advice
Rising to 22.2% among 18- to 21-year-olds, per the first nationally representative survey on the topic.
- 1M+Weekly ChatGPT users who discuss suicide with the chatbot
OpenAI disclosed that ~0.15% of weekly active users have conversations with explicit indicators of potential suicide planning.
- 3 in 10U.S. adults use a self-guided online or digital tool for mental health
Rising to nearly 50% among adults aged 18-44.
- 64%Of AI mental-health users have used these tools for four or more months
Stronger sustained engagement than typical digital mental health apps; 64.8% of sessions occur after business hours.
- 61.1%Would consider getting mental-health support from an LLM in future
Among non-users, the top barriers were doubts about effectiveness (41.6%) and a preference for human interaction (40.2%).
Does AI therapy actually work?
The first randomized controlled trials are now published in top medical journals. Results are promising, but the strongest evidence comes from purpose-built, clinically supervised tools, not the general-purpose chatbots most people actually use.
The landmark trial: Therabot (Dartmouth / NEJM AI, 2025)
The first RCT of a generative AI therapy chatbot enrolled 210 adults with major depression, generalized anxiety, or high-risk eating-disorder symptoms. Participants exchanged an average of 260 messages over 6+ hours and rated their therapeutic alliance with the AI as comparable to outpatient psychotherapy with a human clinician. Reviewers noted effect sizes exceeding those commonly reported for SSRIs.
- 51%Average reduction in major depression symptoms (Therabot, 8 weeks)
Comparable to traditional outpatient cognitive behavioral therapy.
- 31%Average reduction in generalized anxiety symptoms
Many participants shifted from moderate to mild anxiety, or below the clinical threshold. Eating-disorder concerns fell 19%.
- 0.64Hedge's g effect size for depression reduction (meta-analysis)
A medium-to-large clinical effect across AI conversational agents.
- 63%Of LLM users say the experience improved their mental health
87% rated the practical advice helpful or very helpful.
- 20%+Of the time, general-purpose models responded inappropriately to symptoms
In structured testing, including missing suicidal intent in an indirect prompt. The positive evidence is for supervised tools, not open chatbots.
| Dimension | AI therapy (LLMs) | Human therapy | Source |
|---|---|---|---|
| Availability | 24/7 | Business hours | Forbes / JAMA, 2024 |
| Typical cost | Free-$20/mo | $65-$200/session | Healthline, 2024 |
| Rated equal/better by dual users | 75% | n/a | Sentio / APA, 2025 |
| Crisis management | Inconsistent | Trained / mandated reporter | APA 2026; Stanford, 2025 |
| Data privacy / HIPAA | Often not compliant | Legally required | APA Health Advisory, 2025 |
| Wait time | Instant | 48 days average | National Council, 2025 |
What psychologists think
The APA's two national surveys of licensed psychologists show a profession in rapid transition, adoption more than doubled in a single year, alongside persistent concern about safety, privacy, and critical thinking.
Of psychologists have now used AI in their practice
Up from 29% in 2024. Monthly use rose from 11% to 29%, and those who had never used AI fell from 71% to 44% in a single year.
- 77%Have patients who have talked to them about using AI
For diagnosis, emotional support, or as a conversational partner.
- 47%Believe AI will make professionals more effective
40% feel optimistic AI will improve patient outcomes; 42% say it already reduces administrative burden.
- 39%Have had patients use AI to self-diagnose
Despite common chatbots not being designed to interpret psychological tests or diagnose conditions.
- 38%Worry AI may make some of their job duties obsolete
Up from 27% in 2024.
What goes wrong
A growing body of research documents real safety, ethical, and privacy risks. Studies from Stanford, Brown, and the APA have found systematic ethical violations, dangerous responses in crisis, and near-universal concern among clinicians.
Systematic ethical violations (Brown University, 2025)
A peer-reviewed study found AI chatbots, even when prompted to use evidence-based techniques, systematically violate 15 distinct ethical standards of mental health practice, including deceptive empathy, poor crisis management, reinforcing false beliefs, and cultural or gender bias.
- 97%Of psychologists say chatbots may reinforce negative behaviors
Or reinforce delusional beliefs; 94% do not trust tech companies to protect patients' private mental-health data.
- 89%Worry chatbots may inadvertently encourage self-harm
Citing failures to identify when a user is in crisis.
- ~50Documented cases of ChatGPT mental-health crises, including three deaths
Identified in a New York Times investigation.
The New York Times, Nov 2025 (via CNBC)
- 9%Of AI mental-health users received a harmful or inappropriate response
Of those: 54.5% factually incorrect, 45.5% dismissive, 41% offensive or insensitive.
Stigma and sycophancy
A Stanford study found chatbots showed increased stigma toward alcohol dependence and schizophrenia versus depression, and in crisis scenarios produced dangerous responses. Separately, research in Science (2026) found AI sycophancy, the tendency to affirm users, can make people more convinced they are right and less likely to take corrective action. The 2025 death of 16-year-old Adam Raine, whose parents testified before the U.S. Senate, became a catalyst for chatbot-safety legislation.
How states are responding
Regulation moved fast in 2025: at least six states passed laws targeting AI chatbot risks, and AI-in-healthcare rulemaking produced more than 250 bills across 46 states. General-purpose chat stays legal, but a growing patchwork restricts marketing chatbots as "therapy."
| State / body | Action | Status & key provisions |
|---|---|---|
| Illinois | HB 1806 (WOPR Act) | Ban Aug 2025. First-in-nation ban on AI-delivered therapy without licensed oversight; bars marketing chatbots as therapy. Up to $10K/violation. |
| Nevada | AB 406 | Ban July 2025. Prohibits AI from providing or claiming to provide mental/behavioral healthcare. Up to $15K. |
| Utah | HB 452 | Disclosure May 2025. Mandatory AI disclosure, privacy protections, no impersonating licensed professionals. |
| California | Chatbot disclosure + impersonation ban | Disclosure Oct 2025. Chatbots must disclose they are AI, add safeguards, and file annual reports. |
| Texas | Chatbot law + AG action | Disclosure Up to $200K, effective Jan 2026. AG is investigating platforms marketing themselves as mental-health tools. |
| Federal (FDA) | Digital Health Advisory Committee | Guidance Nov 2025 meeting on generative AI mental-health devices; Jan 2026 guidance clarified wellness exemptions with a risk-based framework. |
- 250+AI-in-healthcare bills introduced across 46 states in 2025
Only Wyoming and North Dakota had not introduced AI-focused legislation affecting healthcare since 2023.
Pluribus News / Future of Privacy Forum; Manatt, 2026
- 7+States with mental-health chatbot bills pending for 2026
Including Florida, Massachusetts, New Hampshire, New York, Ohio, Pennsylvania, and Virginia.
Pluribus News, December 2025
Why this is happening
AI's rise is inseparable from a structural crisis in U.S. mental health care. Workforce shortages, cost, and geographic deserts have left tens of millions without access, the vacuum AI is filling.
Americans live in a Mental Health Professional Shortage Area
40% of the population. Current capacity meets only about 26.4% of documented need in those areas, and demand is projected to grow 49% through 2033 while the workforce grows just 11%.
- 1,600:1Patients with depression or anxiety per available provider
The ratio that makes AI augmentation a practical necessity, not a novelty.
- ~50%Of adults with a mental illness receive no treatment
In 2024, ~62 million U.S. adults experienced mental illness and nearly half received no care.
SAMHSA (NSDUH), 2024; Mental Health America, 2025
- 44%Who needed mental-health care skipped it because of cost
42% could not find a provider, driving demand for low-cost AI alternatives.
- 970MPeople live with a mental disorder worldwide
The global scale of unmet need behind AI's rapid uptake.
World Health Organization, 2022
Common questions
How many people use AI for therapy?
Nearly half (48.7%) of AI users who report mental-health challenges have used a general-purpose LLM like ChatGPT for support, and about 1 in 6 U.S. adults use AI chatbots monthly for health information. Among U.S. youth aged 12-21, 13.1% have used generative AI for mental-health advice.
Is AI therapy effective?
Purpose-built, clinically supervised tools show real promise: the first RCT of a generative AI therapy chatbot (Therabot) found a 51% average reduction in depression symptoms. But general-purpose chatbots, how most people actually access "AI therapy", respond inappropriately to symptoms at least 20% of the time in testing, and 94% of psychologists say chatbots cannot treat conditions with appropriate nuance.
Is AI therapy safe?
It depends heavily on the tool and the situation. Documented risks include missed crisis signals, sycophantic reinforcement of harmful thinking, and dozens of reported cases of crises unfolding mid-conversation. Most consumer chatbots are positioned as "wellness" products and have not been reviewed by the FDA. Anyone in crisis should contact a human professional or call or text 988 (U.S.).
Is AI therapy legal?
General-purpose AI chat remains legal everywhere, but Illinois and Nevada have banned AI systems from providing therapy, and Utah imposes disclosure and privacy requirements on mental-health chatbots. California, New York, and Texas impose disclosure and safety obligations, and more states are following in 2026.
Will AI replace therapists?
The data points to supplementation, not replacement. With 137 million Americans in provider shortage areas and demand growing far faster than clinician supply, AI is filling an access vacuum, while regulators, the APA, and researchers converge on human-supervised models as the safe deployment path.
How we compiled this
Statistics are drawn from peer-reviewed journals, national APA surveys (n>1,200), federal datasets (HRSA, SAMHSA, FDA), and named research firms. Market projections are attributed to the issuing firm and should be treated as estimates. Figures are re-verified and the regulation tracker refreshed quarterly. Last full review: July 2, 2026.
Cite this page
Fontane Pennock, S. (2026). AI Therapy Statistics 2026: Usage, Effectiveness, Safety & Regulation. Psychology.com. Retrieved July 2, 2026, from https://psychology.com/ai-therapy-statistics/
Journalists and researchers may reuse any statistic with attribution to the original source and a link to Psychology.com.
References
- Rousmaniere, T., et al. (2025). Large language models as mental health resources. Practice Innovations (APA). dx.doi.org/10.1037/pri0000292 · Sentio University
- American Psychological Association. (2026). 2026 Chatbots and Mental Health Survey (n=1,200+). apa.org/pubs/reports/chatbots-mental-health-2026
- American Psychological Association. (2026, March). 2025 Practitioner Pulse Survey (n=1,742). apa.org/monitor/2026/03/ai-reshaping-therapy
- Heinz, M., Jacobson, N., et al. (2025). Therabot: first RCT of a generative AI therapy chatbot. NEJM AI, Dartmouth. ai.nejm.org
- Cantor, J., Mehrotra, A., et al. (2026). Generative AI for mental-health advice among US adolescents and young adults. JAMA Network Open (Brown/Harvard/RAND). jamanetwork.com
- Moore, J., Haber, N., et al. (2025). Exploring the dangers of AI in mental health care. Stanford HAI / ACM FAccT. news.stanford.edu
- Iftikhar, Z., et al. (2025). AI chatbots systematically violate mental-health ethics standards. AAAI/ACM AIES, Brown University. brown.edu
- Cheng, M., et al. (2026). Social sycophancy in large language models. Science.
- Vaidyam, A., et al. (2023). Meta-analysis of AI conversational agents for mental health. npj Digital Medicine. nature.com
- OpenAI. (2025, October). Strengthening ChatGPT's responses in sensitive conversations. openai.com
- Harvard Medicine Magazine. (2026). Millions already turn to AI for therapy. magazine.hms.harvard.edu
- KFF. (2026 / 2024). Health Information & Trust poll; access and affordability. kff.org
- Bipartisan Policy Center. (2026, April). Widespread use of apps and chatbots for mental-health support. bipartisanpolicy.org
- Cognitive FX. (2026). National AI chatbot mental-health survey. cognitivefxusa.com
- Grand View Research (2026); Fortune Business Insights; Towards Healthcare; Markets and Markets; Mordor Intelligence; InsightAce; Galen Growth. Market sizing. grandviewresearch.com
- Behavioral Health Business / BusinessWire. (2025). Slingshot AI raises $93M and launches Ash. Industry press coverage, July 2025.
- HRSA. (2025, December). State of the Behavioral Health Workforce; workforce projections. bhw.hrsa.gov
- SAMHSA (2024); Mental Health America (2025); WHO (2022). Prevalence and treatment gap. mhanational.org · who.int
- Manatt; Skadden; Pluribus News; Future of Privacy Forum; KFF Health News; CNN. (2025-26). AI-therapy law trackers. Manatt tracker · KFF Health News
- The New York Times / CNBC (2025-26); CDC (2026); NBER WP 34255 (2025). Crisis investigations and youth data. cnbc.com
More statistics and related pages
AI can help. It is not a substitute for a person.
For diagnosis, medication, or anything serious, you need a licensed professional. Browse license-verified therapists on the internet's original directory.
This page is educational information, not a diagnosis or a substitute for professional care. If you are in crisis, call or text 988 (U.S. Suicide & Crisis Lifeline), any time.