Back to Blog

Blog

Could AI Bring Therapists More Clients? What the Evidence Suggests

What the NHS self-referral chatbot study, the RAND cost data and US treatment trends suggest about whether AI adds to or takes from demand for therapy.

July 7, 2026
7 min read
By Declan Ahern
AI mental health chatbotmarket growthfuture of therapytherapist demand

Whether AI takes clients from therapists or brings them more is not settled, and the evidence so far is indirect. This post goes through it: an NHS self-referral chatbot, the RAND cost data, US treatment trends and surveys of AI use.

In February 2024, researchers published a study of NHS Talking Therapies services in England that had added an AI chatbot to the process by which people refer themselves for therapy.1 Referrals to the services that used it rose 15% over the study period, against 6% at control services that did not. The rise was larger in groups that usually reach therapy less often: referrals from nonbinary people rose 179%, and from people in ethnic minority groups 29%.

The researchers analyzed written feedback from 42,332 people and identified two likely reasons for the wider reach. The chatbot was not a person, and talking to it helped people see that they needed treatment.

The study bears on what happens when software makes the first conversation about mental health cheap and easy. It shows that when the first step gets easier, more people come toward care, so AI could bring therapists more clients. That depends on the software being built to lead people toward a clinician when they need one.

Cost decides who starts therapy

One large test of how price affects demand for mental health care is decades old. In 1986 Emmett Keeler and colleagues used data from the RAND Health Insurance Experiment to study how cost sharing, the part of the bill an insured person pays, changed the use of mental health care. They found that cost sharing sharply reduced the number of episodes of treatment, and had only a small effect on how long or how intensively people were treated within an episode. Price also played a larger part in decisions about mental health care than in decisions about other medical care, though a smaller one than most observational studies had suggested.2

So price mostly affects whether people begin an episode of treatment, and much less what happens after. In the US since then, growth in mental health care has come mostly from treating more cases.

Between 2000 and 2021 the number of mental health and substance use cases treated in a year rose from about 35.7 million to about 90.1 million, an increase of about 153%. After adjusting for inflation and population, 87.3% of the growth in spending came from the larger number of cases, and the rest from higher spending per case.3 The counts are of cases treated in each year, so they do not show how many came from people who had never had treatment. The authors could not observe severity, so some of the growth may be milder conditions being treated. The post on the shift from access to outcomes looks at what that growth has meant for the price of a session.

Online therapy is a recent version of the same test. At the J.P. Morgan Healthcare Conference in January 2026, Talkspace's chief executive said that 40% of the company's clients are new to therapy.4 That is the company's own figure, and it does not say what those clients would have done without Talkspace.

The gap AI is entering

Many people with a mental illness get no treatment in a given year. In the 2024 National Survey on Drug Use and Health, 52.1% of US adults with a mental illness in the past year received any mental health treatment in that year, which leaves close to half who received none.5

AI already reaches some of them. In a July 2026 review, Rebekah Bodner, John Torous and colleagues estimated that about 27% of people who use AI have used it for some form of mental health support, and that perhaps 40% are open to it.6 They are careful about the figure. The surveys behind it gave estimates from 3% to 70%, because each defined "mental health support" in its own way, and the authors say a precise number is not yet possible.

The review does not show what those people would have done without AI. Some might have done nothing, and some might have called a therapist, in which case a chatbot took the place of that call. The review estimates how many people use AI this way and does not measure that balance.

When a chatbot stands in for care

The case against expansion is that AI is good enough for many people to stop there. Someone gets through a bad month with a chatbot and never books an appointment. If the difficulty was mild and passed, nothing was lost. The loss comes when a chatbot stands in for care the person needed and nobody is in a position to notice.

This is why the NHS study does not settle the question for AI in general. All six of its authors list Limbic, the company that makes the chatbot, as an affiliation. The study was observational, so the services that chose the tool may have differed from the others in ways it could not measure. It counted referrals, and a referral is a long way from a course of treatment. Above all, that chatbot was built as a front door to a human service, and every conversation it had was meant to end in a referral. A general-purpose chatbot has no such door, and a product paid for by engagement has a reason to keep people talking to it.

Supply matters too. If more people want therapy and the number of therapists stays the same, the result is longer waits, and training more clinicians takes years. What Argentina teaches us about the future of therapy looks at the country that has gone furthest in that direction.

Where Citt sits, and how the prediction could fail

At Citt we built the AI support that clients use between sessions to sit inside a therapist's care. The therapist sets it up for each client, and its instructions tell it to bring recurring themes back to the therapist. That is our answer to the design question above, and what we want AI to make possible sets out the reasoning behind it.

The expansion argument is a possibility, and it can be checked against data. Two findings would count against it: new referrals to therapy services falling while AI use for mental health rises, or surveys showing that people with moderate or severe symptoms use chatbots in place of care they wanted and could get. NHS England publishes monthly counts of new referrals to NHS Talking Therapies services, so the first can be watched in public, though a fall in referrals could also have other causes.7

Footnotes

  1. Habicht, J., Viswanathan, S., Carrington, B., Hauser, T. U., Harper, R., & Rollwage, M. (2024). Closing the accessibility gap to mental health treatment with a personalized self-referral chatbot. Nature Medicine, 30(2), 595-602. A multisite observational study of 129,400 people referred to NHS Talking Therapies services. All six authors list Limbic, the chatbot's maker, as an affiliation. It measures referrals; it does not measure treatment received or outcomes. ↩

  2. Keeler, E. B., Wells, K. B., Manning, W. G., Rumpel, J. D., & Hanley, J. M. (1986). The demand for episodes of mental health services. RAND Corporation, R-3432-NIMH. Based on data from the RAND Health Insurance Experiment; it concerns insurance cost sharing in the US and predates online care and AI. ↩

  3. Mark, T. L., Fernando, L., Shieh, P., & Dunn, A. (2026). US national spending on mental health and substance use disorder treatment driven by case growth, 2000-21. Health Affairs, 45(4), 365-371. Three of the four authors are at the US Bureau of Economic Analysis. Exhibit 2 gives 35,670,113 cases in 2000 and 90,083,559 in 2021. The paper's text calls this a 253 percent increase; those counts give an increase of about 153% (the 2021 figure is about 253% of the 2000 figure). The counts are cases treated per year, not first-time clients. Severity was not observed, so part of the growth may be milder cases. ↩

  4. Talkspace presentation by Jon Cohen, chief executive, at the 44th Annual J.P. Morgan Healthcare Conference, January 2026. Transcript. The 40% figure is the company's own statement; the transcript does not say how it was measured. ↩

  5. Substance Abuse and Mental Health Services Administration. (September 2025). Mental health treatment among people who had a mental health condition in the past year. NSDUH Population Statistics Report, Vol. 1, No. 7. Covers 2024; "any mental illness" is estimated from a statistical model, and "treatment" includes medication and telehealth. ↩

  6. 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. Published July 3, 2026. Draws on 19 records from 2024 onward. The authors call the 27% an approximation and say a systematic review was not possible. ↩

  7. NHS England Digital. NHS Talking Therapies dashboards. Monthly, quarterly and annual data on new referrals, entry to treatment and outcomes. NHS England notes that the data cannot show whether a person was referred again later, so the count is of new referrals, not of people coming to therapy for the first time. ↩

Frequently Asked Questions

Will AI chatbots take clients away from therapists?

The studies in this post do not measure that for general-purpose chatbots. Research on insurance cost sharing and on an AI self-referral tool in NHS services suggests that an easier first step brings more people toward therapy, which makes growth possible but does not show it will happen.

What did the NHS self-referral chatbot study find?

Services in NHS Talking Therapies that added an AI chatbot to their self-referral process saw referrals rise 15%, against 6% at control services, with larger rises among nonbinary and ethnic minority people. The study was observational, and all of its authors are affiliated with the company that makes the chatbot.

How many people use AI for mental health support?

A July 2026 review estimated that about 27% of people who use AI have used it for mental health support. The surveys it drew on ranged from 3% to 70%, because they defined mental health support differently.

How many US adults with a mental illness get treatment?

In SAMHSA's 2024 National Survey on Drug Use and Health, 52.1% of US adults with a mental illness in the past year received any mental health treatment in that year. Treatment in the survey includes medication and telehealth as well as therapy.

Stop rewriting notes from scratch

Citt.ai drafts session notes under your review, keeps between-session context in one place, and stays therapist-led. Start free. No card required.