A conversational support platform based on artificial intelligence (AI) developed in Israel can significantly reduce symptoms of anxiety and depression, form a meaningful therapeutic alliance with users, and provide round-the-clock emotional support.
This was discovered by researchers at Reichman University in Herzliya and published in the prestigious journal JAMA Network Open under the title, “Efficacy of a conversational AI agent for psychiatric symptoms and digital therapeutic alliance: A randomized clinical trial.”
Led by Prof. Anat Shoshani of Reichman’s Baruch Ivcher School of Psychology, and conducted in collaboration with the Kai.ai platform, the study explored whether AI systems could offer an innovative response to the global mental health crisis.
Kai.ai is an AI-driven mental health and wellness platform designed to provide accessible, affordable, and personalized emotional support. By using therapeutic techniques such as cognitive behavioral therapy (CBT) and acceptance and commitment therapy (ACT), it acts as a conversational companion on messaging apps like WhatsApp.
Users engage in brief conversations with the AI to track their mood, journal, and manage stress, guiding them through scientifically backed exercises, including breathing routines and reflective writing. The platform is designed with input from licensed therapists and psychologists to ensure safety and provide a bridge to professional care when needed.
Almost 1,000 Israeli students who suffered from emotional distress, anxiety, and depression were assigned to one of three groups: traditional in-person group therapy with psychologists; a waiting list control group; or a group given access to Kai, which was available to users at any time.
The study was conducted during a period of prolonged security tensions in Israel, which heightened the need for accessible and immediate mental health support.
Conversational AI platform reduces symptoms of depression, anxiety
The findings showed that users of the conversational AI platform experienced a significant reduction in symptoms of anxiety and depression compared to the wait-list group. In reducing anxiety, the system even outperformed human-led group therapy.
Notably, 58% of participants who initially presented with clinical levels of anxiety moved into the healthy range following the intervention. Among participants experiencing depression, nearly half reported substantial improvement.
One of the study’s most striking findings was the formation of a “therapeutic alliance” – a sense of trust, understanding, and emotional connection between users and the AI platform. Participants rated the system as “empathetic, professional, and supportive at levels comparable to human therapists.” Many also said they found it easier to open up to a digital system, free from concerns about judgment or embarrassment.
Shoshani told The Jerusalem Post in an interview that “anxiety is highly situational. Clinical therapy is invaluable, but it is episodic. AI sits in the user’s pocket – on the bus, in the library, and during sleepless nights – providing continuous support precisely when it’s needed. The goal is not to create machines that sound more human, but to build systems that help make our society more human.”
While many mental-health apps experience high dropout rates, the study found that 61% of users continued to use the platform throughout the 12-week study. On average, three days each week. According to the researchers, the sense of continuity, personalized availability, and the system’s ability to “remember” users and their personal context fostered a deeper connection that sustained engagement over time.
Notwithstanding these promising findings, she and her team stress that AI is not a substitute for human psychological treatment, particularly in complex conditions such as post-traumatic stress disorder (PTSD). The Kai.ai platform has a human support team available 24/7, and when it detects signs of acute distress or risk, it triggers an alert to bring in immediate human intervention.
The finding that AI had no significant effect on PTSD symptoms is “extremely important because it helps define the boundaries of what conversational AI can and cannot do. It appears particularly promising for mild-to-moderate anxiety, depressive symptoms, emotional regulation, stress management, and day-to-day well-being support.
“Trauma is different, as the symptoms are often more complex and may require focused interventions. The platform was not originally designed as a trauma-specific treatment model,” she said.
Shoshani said that she was “surprised by the magnitude of the anxiety findings, especially because face-to-face group therapy is itself an active and clinically meaningful intervention. Anxiety often emerges in real time and often requires support at the exact moment distress escalates.”
“Thoughts late at night, panic before social situations, spiraling thoughts before exams, or anticipatory anxiety during ordinary daily situations are difficult to fully address within a once-weekly therapy format,” she continued.
Clinical psychologists meet patients inside a therapy room once a week, “but this doesn’t suit everybody, especially not in the current reality in Israel, where distress levels in the population have been so high,” she asserted.
“Emotional crises rarely wait patiently for next week’s appointment. If clinicians can help shape and guide AI support between appointments, it may significantly boost the therapeutic impact,” she added.
She warned, however, that “one should not overstate this finding. This does not mean AI is ‘better than therapy’ in a global sense. It suggests that certain forms of distress, particularly anxiety, may benefit from highly accessible, ongoing support models that complement traditional care.”
Shoshani explained that the participants were university students, many of whom served long periods in military reserve duty over recent years.
“This is a generation that has carried much of the burden of ongoing national stress on its shoulders. We know there is a strong relationship between exposure to war-related stressors and symptom severity. Many participants had not yet had an opportunity to care for their own mental health, and some were facing very long waiting lists for therapy.
“Others were trying to catch up academically and socially after prolonged disruptions, while some continued reserve service during the study period itself. The moment someone needs support may occur in the middle of the night, during a panic attack, after quarreling with someone, or while sitting alone in a dorm room,” she said.
Kai carries forward relevant context from earlier conversations, so support can feel continuous rather than fragmented. It also includes safety monitoring, therapeutic structures, and psychologically informed conversational frameworks designed to create a more supportive emotional experience.
In mental health, the best response is often not only the right response, but the right response at the right moment. What AI platforms like Kai may offer is the possibility of combining clinical knowledge with psychological timing: support that is not only evidence-based but also present when the person is actually struggling, Shoshani added.
Since participants rated the AI system as comparable to human therapists in warmth and professionalism, the fact that humans can form emotional trust with AI systems “tells us something profound about human psychology. Humans are highly sensitive to cues of responsiveness, warmth, validation, consistency, memory, and emotional attunement,” said Shoshani.
“When a system repeatedly responds in a way that feels empathic, nonjudgmental, and personally relevant, people may begin to experience a form of emotional trust, even when they know intellectually that the system is not human,” she continued. “AI offers a new kind of psychological interaction that shares some mechanisms with therapy while remaining fundamentally different from human care.”
People often disclose secrets, shame, fears, and highly personal experiences more easily to AI systems than to other humans. “Fear of judgment, embarrassment, rejection, or burdening another person may become less intense in AI interactions,” she added.
Not everyone in the population needs intensive psychotherapy, the Reichman psychologist stressed. “Even within our study, participants represented varying levels of distress and psychological need. The goal is not to keep people away from therapy; it’s to make sure people receive the right level of support at the right time,” said Shoshani.
There are risks in users becoming emotionally dependent on conversational AI systems, particularly among vulnerable populations, because the design of AI systems for mental health must be approached with enormous ethical sensitivity and responsibility, she warned.
Human well-being should remain the central priority, not maximizing engagement, dependency, or compulsive use. AI is not designed to replace friends, partners, family relationships, or human intimacy. In many cases, it actively encourages users toward healthier real-world connections, help-seeking, and daily functioning.
One of the most promising possibilities is integrating AI platforms directly into ongoing psychotherapy. Therapists could help shape personalized between-session support plans, guide the platform toward specific therapeutic goals, reinforce coping strategies, monitor progress, or support patients after formal treatment ends, she suggested.
“This was not a study of people with severe psychiatric illness, acute suicidality, psychosis, severe personality pathology, or complex chronic trauma. The participants were university students experiencing psychological distress, generally in the mild-to-moderate range,” Shoshani said.
“Kai is not only built around reducing symptoms; it also includes well-being, gratitude, strengths, meaning, relationships, self-compassion, and personal growth. People are not only collections of symptoms. They are whole people trying to live, study, love, function, recover, and grow,” she continued.
Mental-health AI needs safeguards at many levels because it operates in one of the most sensitive areas of human life.
“People may turn to these systems when they are anxious, ashamed, lonely, traumatized, or in crisis. In those moments, a response that sounds fluent and caring is not enough. It also has to be clinically responsible. That is why external evaluation is so important,” she said.
Cultural sensitivity matters enormously because emotional expression, help-seeking, family roles, and distress can look very different across cultures. Working with adolescents is not the same as working with adults. AI should help people reflect, regulate emotions, and consider options, but it should not make decisions for them or replace human judgment, she said.
“The platform also has to recognize situations where ordinary emotional support is not enough – suicidality, self-harm, psychotic symptoms, abuse, trauma reactions, severe deterioration, or emotional escalation that requires referral to human care. In those situations, the system needs to know when to stop being conversational and become protective,” Shoshani explained.
People may express distress differently, and cultural adaptation absolutely matters.
“But at a very human level, most people need similar things during difficult periods: someone or something that listens without judgment, helps them reflect, offers perspective, encourages hope, and creates a sense that they do not have to carry everything alone,” she said.
If technology can responsibly create more moments in which people feel less alone, think more clearly, breathe more deeply, and find their way back to life and to other people, then its value is not only technological or clinical.
“It is profoundly human,” she concluded.
