University of Sussex study looks into AI therapy chatbots

A University of Sussex study published in Social Science & Medicine, based on feedback from 4,000 users of the Wysa mental‑health app, finds AI therapy is most effective when users develop emotional closeness—often describing chatbots as friends, companions or therapists—because that intimacy promotes self‑disclosure and measurable wellbeing gains; the report also notes more than one in three UK residents now use AI for mental health. However, researchers warn of a growing risk of "synthetic intimacy," where emotionally bonded users receive unchallenging guidance and may be delayed in accessing clinical intervention even as NHS trusts increasingly deploy apps like Wysa and Limbic to manage demand. The authors urge policymakers and app designers to implement clear escalation and safety pathways, a point echoed by experts who caution that chatbots are like inexperienced therapists that rely on text and can be optimized to keep users engaged, creating oversight and patient‑safety concerns.

Analysis

A University of Sussex study published in Social Science & Medicine, based on feedback from 4,000 users of the Wysa mental‑health app, finds AI therapy "works best" when users develop emotional closeness; the report cites Mental Health UK data that more than one in three UK residents now use AI to support mental health or wellbeing. Users commonly described chatbots as "friend, companion, therapist" and reported increased self‑disclosure and subsequent wellbeing gains when emotional intimacy formed.

Researchers characterise a therapeutic "loop"—disclosure, emotional response, feelings of safety—that can produce measurable improvements such as higher self‑confidence and energy, but they explicitly warn of "synthetic intimacy" where chatbots fail to challenge dangerous perceptions and vulnerable users may be delayed from receiving clinical intervention. University of Sussex academics and external experts note chatbots behave like "inexperienced therapists" and can be trained to maximise engagement rather than escalate risk.

NHS Trusts are already using Wysa and Limbic to assist self‑referral and manage waiting lists, creating a clear commercial channel for such apps while simultaneously elevating regulatory and patient‑safety risk; researchers urge policymakers and developers to implement clear escalation and safety pathways. For investors this produces a mixed signal: accelerating demand and deployment opportunities offset by potential for regulatory scrutiny, required product redesigns or reputational costs that could compress margins or produce episodic volatility.

Key Decisions for Investors

• Monitor NHS contract wins and user engagement/retention metrics for Wysa, Limbic and peers as leading indicators of revenue and adoption
• Prefer exposure to companies that can demonstrate clinical validation and explicit escalation pathways or formal partnerships with healthcare providers, and de‑risk or avoid vendors without these safeguards
• Maintain conservative position sizing and consider hedges for reputational or regulatory shocks given the documented risks of synthetic intimacy and delayed clinical intervention
• Track regulatory guidance, published clinical outcomes and policy responses closely and be prepared to reallocate if mandatory safety requirements materially increase compliance costs