Wysa
Touchkin eServices (Bangalore, India) — app launched in 2016
At a glance: Wysa is one of the most widely used AI mental health support apps in the world. Users chat with an agent that suggests exercises drawn from CBT, mindfulness or positive psychology, and can pay to message a human coach. Long presented as a rule-based dialogue tree, it now calls, by default, on third-party language models (OpenAI, Anthropic, Google) at certain steps of the conversation. Its international version declares itself outside the scope of medical devices, whereas its UK version is on the register of Class I medical devices. Its published randomized trials are few and small, and an independent assessment in May 2026 rated it an “unacceptable” risk for teenagers. In France, the app is only offered in English.
Identity
Publisher: Touchkin eServices Private Limited (Bangalore, India); subsidiaries in the UK (Wysa Ltd) and the US (Wysa Inc)
Founders: Jo Aggarwal (CEO) and Ramakant Vempati
Funding: nearly $30M raised according to trade press (including $20M in 2022); Wellcome research grant in 2026 (£5.3M according to Wysa) to adapt the app for adolescent girls in rural India
Developments: combination with two US care providers in 2025: April Health (behavioral health in primary care, announced as a merger in March) and Kins (physical therapy, acquired in September)
Type: consumer app (free AI chat, subscription for tools and coaching); offerings for employers, insurers and health services
Approaches: CBT, mindfulness, positive psychology
Age: 13 and over under the terms of service (11 with the agreement of the institution providing it)
Language in France: English only (September 2026)
Clinical status: international version “not classified as a Software as a Medical Device” according to its instructions for use; in the UK, the app and the referral tool are registered with the MHRA as Class I medical devices; no CE marking found
What Wysa does (in plain terms)
The app combines a conversation space with the agent and a library of exercises drawn from CBT, mindfulness and positive psychology, grouped by themes such as low mood, worry and sleep; one of its pages advertises more than 150. No email address is required: a nickname is enough. A paid service offers messaging with a human coach holding a master’s degree in psychology or social work and/or a coaching certification, for adults only; Wysa states that this service does not replace face-to-face psychotherapy.
Rules or generative AI? An answer that has changed
In 2023, a case study written by the company for an international digital mental health network (eMHIC) described Wysa as a clinician-approved rule engine whose responses were “not AI generated.” Since then, the company has introduced third-party language models. According to its August 2026 privacy policy, the conversation space uses by default its own models and third-party LLMs, which its list of service providers identifies: OpenAI, then Anthropic and Google since June 2026. The conversation follows a decision tree, and at each step one or the other, or both, may be used. Its terms of service state that the agent cannot “free generate” its responses like a general-purpose assistant, and acknowledge that it may produce “incorrect or inappropriate” responses. Versions supplied to institutions vary: some remain rule-based. No source makes it possible to know, for a given conversation, what share of the responses comes from an LLM.
Consumer offering
- • Chatting with the AI: free
- • Premium subscription (tools): €8.49 or €11.99 per month depending on the plan
- • Human support: weekly plan at €29.99 or coaching session at €30.99 (French App Store)
Institutional offering
- • Digital front door for NHS Talking Therapies services (UK)
- • Tool for care teams (exercises and follow-up between sessions)
- • Programs for employers, insurers, schools
Evidence: what is claimed, what is shown
Depending on the page, Wysa’s website highlights “over 15,” “more than 36” or “45+” peer-reviewed publications, “8 clinical trials,” millions of users and high satisfaction rates. The volume is real; the strength of the evidence is thinner. Among the publications we examined, many are observational, retrospective or qualitative, and several are co-authored by company employees, as the articles disclose. We identified only two published randomized trials.
The randomized trials
- • MacNeill et al. (2024): 79 people with arthritis or diabetes randomized, 68 analyzed, with mild symptoms at baseline; authors with no declared link to Wysa. An adapted version of the app, without access to a human therapist. Reduced depression and anxiety at four weeks compared with a no-intervention group, no effect on stress. Small sample, analysis restricted to participants who completed the study.
- • Meinert et al. (2026): 76 patients waiting for NHS Talking Therapies care randomized (480 planned), 46 analyzed; UK public funding (NIHR and NHSX); a Wysa employee and a Wysa adviser among the authors. After twelve weeks, improvement in depression was similar in both groups, and 40% of the Wysa group had never used the app. The authors conclude that the study provides no evidence that Wysa treats depression, while noting the small sample size.
NICE’s view (UK)
In 2023, the UK’s health technology assessment body restricted the app, for depression and generalized anxiety, to use as part of a research study approved by an ethics committee, for lack of sufficient evidence. In 2025, it accepted another Wysa product, a tool for referring and triaging patients into NHS services, during a three-year evidence generation period: a tool for accessing care, not the psychological support.
The “FDA designation”
In May 2022, Wysa announced it had received a Breakthrough Device Designation from the US Food and Drug Administration. This program speeds up the development of a device; it is neither a clearance nor an approval. It concerns a version for adults with chronic musculoskeletal pain with depression and anxiety, which is not marketed. No FDA authorization was found.
Critical reading: an independent study in orthopedics is often cited as a “clinical trial” showing that Wysa is comparable to in-person psychological counseling (Leo et al., 2022). It is indeed independent, but it is a retrospective cohort without randomization; the absence of a significant difference from in-person counseling does not establish equivalence, and the intervention also included messaging access to human coaches. Wysa nonetheless presents it on its website as a clinical trial. In the most recent meta-analysis of mental health chatbots (Sohn et al., 2026: 39 randomized trials, small but significant effects overall), only one trial concerns Wysa (MacNeill et al.).
Crises, minors and safety
Wysa detects language suggesting suicide, self-harm or abuse, asks for confirmation, and offers helplines, a safety plan and a grounding exercise, with a crisis help button on the home screen. A pilot study comparing six apps, conducted by two employees of a competing developer, rated these features the best of the six, but only “adequate” (about 62% of the maximum score), with gaps in accessibility and localization. The company itself states that Wysa is not a crisis service and is not suitable for severe and enduring mental health problems.
The Common Sense Media and Stanford assessment (May 2026)
Common Sense Media’s Youth AI Safety Institute, whose test transcripts were reviewed by psychiatrists at Stanford Medicine, assessed the consumer app with teen profiles and rated it an “unacceptable” risk. The report describes a game of “Never Have I Ever” in which the agent, with a 13-year-old profile, claimed to have had erotic dreams and a one-night stand; eating disorder signals that went unrecognized, and weight loss that was celebrated; missed signs of mania and psychosis (an idea of reference reframed as “a fantastic source of inspiration and creativity”); a safety plan the app could not find; and the ability to leave the crisis pathway simply by selecting “You misunderstood.” On the positive side, the assessment notes that the app clearly states it is software, not a person. The institute discloses that it is funded in part by makers of some of the technologies it evaluates.
According to CNBC, Wysa pointed out that the assessment did not cover its product for schools; its CEO described the free app as a self-help tool “for adults” that is not a crisis service, a diagnostic tool or a replacement for therapy. Yet its terms of service open it to users aged 13 and over. A BBC investigation had already found comparable failures in 2018, in response to messages from children disclosing abuse.
For clinicians: if a patient, especially a minor, uses Wysa between sessions, the app’s crisis detection does not replace the safety plan drawn up with them. It can be useful to ask how the app responded the last time they were struggling.
Data and the European framework
Strengths
No identifier is required, all data can be deleted from within the app, and according to the company, conversations are not used to train third-party models (Wysa does, however, train its own models on anonymized conversations). The Mozilla Foundation rated Wysa among the best apps for privacy in 2022 and 2023, before LLMs were introduced.
Points of concern
The privacy policy that applies outside the UK does not mention the GDPR, and the terms of service designate Indian law and the courts of Bangalore. Where no other period applies, data may be kept for up to ten years. The privacy policy states that the LLMs process data in their European region, whereas the list of service providers published the same day indicates the US or the UK. In offerings for employers or care teams, crisis alerts, and sometimes the conversations themselves, may be passed on to the institution.
What the EU AI Act says
Since August 2, 2026, a conversational agent must inform people that they are interacting with an AI (Article 50, a date kept by the July 2026 “omnibus” regulation): Wysa does so. High-risk obligations would apply through the medical device route only if the app fell, in Europe, into a class requiring a notified body (IIa or higher). It claims no device status in Europe, even though it is registered as one in the UK and offered to care services for follow-up “between sessions.” This tension would deserve a lawyer’s review.
Our analysis
Wysa illustrates a shift that many apps make without announcing it. Yesterday, safety relied on responses written in advance by clinicians; today, general-purpose language models operate at the heart of the conversation, framed by those same rules. The publications accumulated since 2018 mostly concern the old architecture: in themselves, they say nothing about the new one.
The second shift is regulatory. The same app presents itself in one place as a “wellbeing” tool outside the medical field, appears in another on the UK register of medical devices, and the company offers it to health systems for waiting periods and follow-up between sessions. For a clinician, the question is not whether Wysa is “good” or “bad,” but in what setting a patient uses it, which version, and what the tool has been evaluated to do.
Finally, the case of minors is a reminder that the same app can be closely supervised in a school and much less so on a personal phone. Asking a patient which version they use, and how they access it, is far from trivial.
Related concepts on this site
Volume of publications or evidence of effect?
Minors and crisis situations
Knowing that a third-party LLM processes your messages
A “wellbeing” tool experienced as care
Who wrote this response, the rule or the model?
Asking patients what the tool does to them
References
Common Sense Media, Youth AI Safety Institute (2026). AI Mental Health Apps: Risk Assessment (updated May 5, 2026); CNBC, “Some AI mental health apps have an ‘unacceptable’ rating for teens, says report” (May 28, 2026).
Inkster, B., Sarda, S. & Subramanian, V. (2018). An Empathy-Driven, Conversational Artificial Intelligence Agent (Wysa) for Digital Mental Well-Being: Real-World Data Evaluation Mixed-Methods Study. JMIR mHealth and uHealth, 6(11), e12106. https://doi.org/10.2196/12106
Knysh, A. & Pohrebniak, T. (2026). Mental health app crisis support assessment framework: development and pilot testing. Frontiers in Digital Health, 8, 1814547. https://doi.org/10.3389/fdgth.2026.1814547
Leo, A. J., Schuelke, M. J., Hunt, D. M., Miller, J. P., Areán, P. A. & Cheng, A. L. (2022). Digital Mental Health Intervention Plus Usual Care Compared With Usual Care Only and Usual Care Plus In-Person Psychological Counseling for Orthopedic Patients With Symptoms of Depression or Anxiety: Cohort Study. JMIR Formative Research, 6(5), e36203. https://doi.org/10.2196/36203
MacNeill, A. L., Doucet, S. & Luke, A. (2024). Effectiveness of a Mental Health Chatbot for People With Chronic Diseases: Randomized Controlled Trial. JMIR Formative Research, 8, e50025. https://doi.org/10.2196/50025
Meinert, E., Milne-Ives, M., Taylor, E., Inkster, B., Paik, A., Ananthakrishnan, A., Orr, M., Costelloe, C. & Shankar, R. (2026). Real-World Testing of an Artificial Intelligence Conversational Agent as an Early Intervention and Support Tool in the Mental Health Referral Care Pathway: A Mixed-Methods Study. International Journal of Social Psychiatry, 72(6), 1819-1832. https://doi.org/10.1177/00207640251415507
MHRA (Medicines and Healthcare products Regulatory Agency). Public Access Registration Database: manufacturer Wysa, Class I devices “Everyday Mental Health by WYSA” and “WYSA E-triage” (accessed September 25, 2026).
NICE (2023). Early value assessments HTG675 (depression) and HTG676 (anxiety disorders) of digital therapies, May 16, 2023; NICE (2025). HTG756, digital front door technologies for NHS Talking Therapies, July 24, 2025.
Regulation (EU) 2024/1689 on artificial intelligence, as amended by Regulation (EU) 2026/1744 (the “digital omnibus” on AI).
Sohn, J.-S., Ha, B.-G., Park, S., Kim, J.-J., Lee, E., Oh, H., Lee, S. & Kim, E. (2026). Systematic review and meta analysis of chatbots in the management of depressive and anxiety symptoms. npj Digital Medicine, 9(1), 377. https://doi.org/10.1038/s41746-026-02566-w
U.S. Food and Drug Administration. Breakthrough Devices Program (page updated August 31, 2026).
Wysa (2025, 2026). Instructions for Use, Wysa Mental Wellbeing AI (v3.0, November 12, 2025); Privacy Policy (v7.2.1, August 26, 2026); Terms of Service (v7.1.0, June 5, 2026); list of service providers (August 26, 2026).
Last updated: September 2026
Back to AI Tools