An AI buffer before the first appointment
A research team in South Korea says it has developed a conversational AI system to support one of the hardest moments in mental-health care: the first psychiatric interview. The project, led by researchers from the Korea Advanced Institute of Science and Technology with clinical collaborators at Gangnam Severance Hospital, is designed to let patients talk with an AI system before sitting down with a doctor.
The idea is straightforward. Initial psychiatric consultations often ask patients to explain distress, symptom history, and personal context under pressure and in limited time. That can be uncomfortable for patients and difficult for clinicians, who must quickly identify the details most relevant to diagnosis and treatment. The new system is meant to reduce that friction by helping patients organize what they want to say in advance while also collecting clinically useful information.
How the system works
According to the researchers, the tool is built on a large language model that can adapt the flow of an interview in real time. Rather than following a fixed questionnaire, it compares a patient’s responses against specialized psychiatric knowledge and uses that comparison to generate the next questions. The goal is to make the exchange feel more like a guided clinical conversation than a rigid form.
The team also says the system incorporates techniques drawn from counseling practice. Those include expressing empathy, restating a patient’s words in a more organized way, and clarifying ambiguous statements. That matters because psychiatric intake is not just about collecting facts. It also depends on whether a patient feels comfortable enough to continue disclosing emotionally difficult information.
In that sense, the system is being positioned as a support layer rather than an autonomous clinician. It is not described as diagnosing patients or replacing psychiatrists. Instead, it is intended to prepare both sides for a more focused in-person appointment.
What the early testing suggests
The researchers evaluated the system using 1,440 virtual patients. In those tests, they reported that the AI was able in most cases to obtain key clinical information needed for treatment within 30 minutes. That result, if it translates to real-world use, would speak directly to one of the main bottlenecks in psychiatric care: the need to surface a large amount of relevant history quickly without overwhelming the patient.
The study was presented on April 13 at ACM CHI 2026 and published in the proceedings of the 2026 CHI Conference on Human Factors in Computing Systems. That places the work within a human-computer interaction context as much as a purely medical one, which is important. The challenge here is not only model accuracy, but also how a patient experiences the exchange and whether clinicians receive information in a form they can trust and use.
Why this matters for mental-health systems
Mental-health services in many countries are under strain, with growing demand, staffing shortages, and persistent barriers to access. Tools that can improve intake quality without asking doctors to spend more time per case are likely to draw attention. If patients can arrive better prepared and clinicians can begin with a clearer, structured summary, the first appointment may become more productive.
Still, several practical questions remain outside the scope of the supplied report. Real patients are not virtual patients, and psychiatric interviewing raises obvious issues around safety, privacy, escalation, and bias. It also matters how clinicians verify or reinterpret AI-organized summaries, especially when symptoms are complex or patients communicate indirectly.
Even so, the work points to a more concrete use of generative AI in health care than many headline-grabbing chatbot claims. Rather than promising fully automated care, it focuses on a narrow but important administrative-clinical handoff: helping people say what is wrong, and helping doctors hear it faster.
This article is based on reporting by Medical Xpress. Read the original article.
Originally published on medicalxpress.com
