Artificial intelligence is becoming increasingly common in healthcare, but an important question remains: can medical chatbots communicate safely with real patients?
Researchers at a major Boston medical center have investigated this question in a prospective study published in *The Lancet* in October 2026.
The findings suggest that conversational AI could help patients prepare for medical appointments and provide doctors with useful information. However, concerns about reliability, privacy and clinical safety remain.
<h3>How Was The Study Conducted?</h3>
The researchers evaluated a medical AI system called Articulate Medical Intelligence Explorer (AMIE), developed by Google. The study took place between April and November 2025 and involved 114 patients, of whom 98 completed both the AI conversation and their medical appointment. Participants used a secure text-based interface from home after arranging an urgent primary care consultation.
The chatbot asked about symptoms, collected relevant medical history and suggested possible diagnoses for discussion with a doctor. It also prepared a summary that healthcare professionals could review before meeting the patient. Unlike many earlier experiments involving simulated medical scenarios, this research examined interactions with actual patients.
<h3>Was The Chatbot Safe?</h3>
Patient safety was the central focus. Every conversation was supervised in real time by a qualified internal medicine physician. Supervisors could intervene if the chatbot provided potentially dangerous information, overlooked urgent symptoms or caused significant emotional distress.
Across the 98 completed conversations, none required an emergency safety-stop intervention. However, supervising doctors identified one instance in which the AI generated inaccurate information and provided additional clinical clarification in five cases.
These findings are encouraging, although they cannot establish that the technology would remain equally safe without medical supervision.
<h3>What Did Patients Think?</h3>
Most participants evaluated their conversations positively. Patients appreciated the chatbot's ability to listen, explain medical information and create a comfortable atmosphere. Their attitudes towards medical AI also became more positive after using the system and remained improved following their appointment. Nevertheless, two concerns persisted.
Some participants questioned whether their personal medical information would remain confidential. Others were uncertain about the chatbot's honesty and reliability.
These findings suggest that technical performance alone may be insufficient to establish trust.
<h3>Did Doctors Benefit?</h3>
Doctors were able to review AI-generated transcripts or summaries before 44 consultations. In 75% of these cases, physicians reported that the information helped them prepare for the appointment. In 57%, they believed it might have influenced their clinical approach. However, one doctor considered the interaction somewhat harmful because a patient may have become anxious after the chatbot mentioned lymphoma as a possible diagnosis.
This example highlights the importance of communicating medical uncertainty carefully. Even technically plausible suggestions can cause unnecessary concern when presented without appropriate context.
<h3>Can AI Replace A Doctor?</h3>
The researchers emphasised that the study assessed feasibility and safety under supervision. It did not establish whether using AI improves diagnosis, treatment decisions or long-term patient health. The relatively small number of participants also limits how confidently the findings can be applied to larger populations. Future studies will need to examine clinical outcomes, potential errors and performance without continuous physician monitoring.
<h3>What Could Change In Healthcare?</h3>
Medical chatbots could eventually help patients organise their symptoms, prepare questions and share important information before consultations. For doctors, structured summaries might make appointments more efficient and allow more time for discussing treatment. However, responsible implementation will require strong privacy protections, transparent communication and continued clinical oversight. The study offers an early indication that AI may become a useful addition to primary care, provided that patient safety remains the priority.
