AI nurse triage is the part of a clinic's AI voice agent that asks a caller about symptoms and decides how urgently the caller must be seen. Triage means sorting patients by urgency. It is the hardest job that an AI medical receptionist can do, because a wrong answer can harm a patient. Products like this are sold mainly to clinics and hospitals in the United States.
Nurse triage by phone is an old service. Many American clinics and health plans run a nurse line. A patient calls with a symptom, and a nurse asks questions from a written protocol. A protocol is a fixed list of questions for each symptom, written by doctors. The answers lead to one of a few outcomes, called dispositions. Examples are "call emergency services now", "go to an emergency room", "be seen within 4 hours", "book an appointment within 3 days" and "care for yourself at home". The AI agent follows the same protocols. It asks the questions by voice, records each answer and reaches a disposition.
Here is an example. A woman in Oregon calls her clinic at 11:10 p.m. She says it has burned when she passes urine for two days. The agent asks the four questions in the protocol for this symptom. Does she have a fever? Pain in the back or side? Vomiting? Could she be pregnant? She says she has a fever of 38.9 °C (102 °F) and pain in her lower back. A urine infection with fever and back pain may have reached the kidneys. So the protocol's disposition is "be seen within 4 hours".
The agent does not end the call there. It passes the call to the on-call nurse, with the answers already written down. The nurse confirms the decision in two minutes and directs the woman to a 24-hour urgent-care clinic.
Why keep a nurse involved? Most such agents follow a rule: the AI drafts the decision, and a person approves it. This rule is called human in the loop. The agent may handle easy outcomes alone, such as advice for a mild cold. But any urgent outcome goes to a nurse. And if the caller mentions a red-flag symptom, such as chest pain or difficulty breathing, the agent stops asking questions and tells the caller to call emergency services at once.
The agent hears the answers through speech recognition. So it must also cope with accents and unclear answers, and ask again when it is not sure. The same idea of sorting by urgency is used for scans, ECGs and emergency rooms, as explained in triage. In India, a similar first step appears in telemedicine apps, as an AI symptom checker.