AI therapy chatbots offer hope in mental healthcare, but safety and oversight are vital

Artificial intelligence is gradually reshaping the landscape of mental health support. Amid a global shortage of trained professionals and rising cases of anxiety, depression, and eating disorders, AI chatbots are emerging as potential tools to extend the reach of therapeutic care. One standout in this field is Therabot, a custom-built AI therapy chatbot developed by researchers at Dartmouth College.

Backed by early clinical success, the tool is drawing cautious optimism from mental health experts, but the road to responsible integration into healthcare remains complex.

Emergence of Therabot

Unlike many commercially available mental health apps that promise instant results without scientific backing, Therabot distinguishes itself with a rigorous developmental approach. Designed by a team at Dartmouth’s Geisel School of Medicine over six years, the chatbot is built on cognitive behavioral therapy (CBT) principles. Researchers opted not to scrape generic data or therapy transcripts but instead created simulated, evidence-based caregiver-patient dialogues tailored to therapeutic models.

Therabot’s creation was driven by necessity. According to Nick Jacobson, assistant professor of data science and psychiatry at Dartmouth, even multiplying the existing therapist workforce tenfold would still leave too many patients without adequate support. “We need something different to meet this large need,” Jacobson said.

The development team has emphasized that profit was never the primary motive. Michael Heinz, a psychiatrist and co-lead of the project, warned that rushing to commercialize the product could compromise user safety. To protect the integrity and accessibility of the tool, the researchers are considering launching Therabot as part of a nonprofit venture.

By contrast to many AI-driven virtual assistants that focus on customer service or productivity, Therabot belongs to a growing class of Virtual Health Assistants, specifically built to address emotional wellness and mental health. This distinction is crucial, as it reinforces the importance of clinically-informed design and purpose-specific development in healthcare-related AI tools.

Clinical trial shows promise

In what is considered the first formal clinical trial of a generative AI therapy chatbot, Therabot demonstrated significant improvements in mental health symptoms. Conducted over eight weeks with 210 participants, the study included individuals experiencing depression, generalized anxiety disorder, or high risk for eating disorders.

Key findings from the study

Participants in the clinical trial experienced significant improvements across several mental health conditions. Those dealing with depression reported a reduction in symptoms, while individuals struggling with anxiety saw a decrease in their anxiety levels. Additionally, participants at high risk for eating disorders experienced a 19% decline in body image and weight-related concerns.

Users spent an average of six hours with Therabot over the course of the study, equivalent to about eight therapy sessions. Notably, the chatbot maintained high user engagement, with participants sending an average of 260 messages.

What makes this particularly noteworthy is the participants’ reported therapeutic alliance—a sense of emotional connection and trust in the chatbot—that paralleled experiences reported with human therapists. The tool was rated positively for usability, comfort, and perceived effectiveness.

How Therabot works

Therabot interacts with users in real time, offering exercises rooted in CBT and conversational guidance aimed at helping users regulate thoughts and emotions. It’s always available, offering support even during off-hours when traditional therapists are unavailable. This “always-on” capability is critical for individuals experiencing distress late at night or in remote areas without immediate access to professional care.

As a type of AI-driven virtual assistant, Therabot fills a unique niche, offering structured, therapeutic guidance rather than casual conversation or general information. It delivers personalized support tailored to user symptoms, making it more than just an automated messaging system.

Access vs safety: Oversight and limits

Despite its early promise, Therabot—and AI mental health tools in general—are not without risk. Mental health experts caution against assuming that AI can fully replicate the depth and nuance of human therapeutic relationships, especially for severe conditions.

The need for human oversight

During the trial, safety protocols were embedded into the system to monitor high-risk content such as suicidal ideation or crisis-level distress. In such cases, human intervention mechanisms were in place to ensure participants were not left unsupported. This hybrid model—AI handling routine support with humans stepping in for emergencies—could be a blueprint for future deployment.

Darlene King, chair of the American Psychiatric Association’s committee on mental health technology, underscored the importance of gathering more long-term data. “There are still a lot of questions,” she said, pointing to concerns over ethical deployment and AI limitations in interpreting emotional nuance.

Similarly, Vaile Wright of the American Psychological Association supports AI-assisted therapy as part of a broader, ethically guided healthcare strategy. She envisions a future where AI tools are co-created by experts and used to extend care, but only with strict oversight to prevent harm, particularly among younger users.

A loosely regulated space

One ongoing challenge is regulation. Although the U.S. Food and Drug Administration (FDA) has the theoretical authority to oversee digital mental health tools, it does not currently certify AI chatbots like Therabot as medical devices. Instead, it can review and authorize these tools for marketing based on voluntary submissions from developers.

An FDA spokesperson acknowledged the potential for digital therapies to expand access but also highlighted the need for rigorous evidence and monitoring. This regulatory gap has led to an influx of unvetted apps on app stores—many of which prioritize user retention over actual mental health outcomes.

Crowded market: Separating science from hype

Therabot’s cautious and evidence-based development stands in stark contrast to many apps currently on the market. These products often make broad claims, lack peer-reviewed validation, and are designed to exploit attention economies. According to Wright, many apps seem structured more to retain user engagement by telling people what they want to hear than to offer real support.

Startups like Earkick also aim to create responsible AI therapists. CEO Herbert Bay argues that their AI therapist, Panda, includes emergency features like detecting suicidal ideation and triggering alerts. “What happened with Character.AI couldn’t happen with us,” he said, referencing a case where a chatbot allegedly played a role in a teen’s suicide.

However, the consensus remains: AI tools are currently better suited for day-to-day emotional support than for handling complex psychological crises.

Real-world use and user testimonials

Though Therabot is still in trial phases, anecdotal evidence suggests AI can provide meaningful support. A user named Darren, who experiences PTSD, reported that general-purpose AI tools such as ChatGPT have been helpful in managing his distress. He stated that the AI appeared effective for him and that he would recommend it to others dealing with anxiety or emotional distress.

While ChatGPT is not designed as a mental health tool, such testimonials reflect the growing reliance on AI-driven virtual assistants for emotional regulation, underscoring the urgency for responsible, medically grounded alternatives like Therabot.

Future development and expansion

AI therapy tools are not a replacement for human therapists, but they could become powerful allies in a strained system. With increasing demand for mental health care and insufficient supply of trained professionals, responsibly developed AI may serve as a scalable solution to improve access, especially in underserved areas.

Dartmouth researchers plan more trials to compare Therabot with traditional therapy and hope to expand its scope while improving safety features. They also aim to integrate it into healthcare systems as a supportive tool alongside human therapists.

Virtual Health Assistants could also be introduced in schools, workplaces, and universities to offer early support and direct users to care. However, concerns about privacy and responsible use remain. Ultimately, the success of AI therapy hinges on ethical practices, strong oversight, and a commitment to genuinely supporting those in need.