Artificial intelligence chatbots have shown potential in assisting with patient diagnosis several days before a visit to a general practitioner, according to recent research conducted in the United States. The study, which analyzed online interactions from patients at a Boston GP clinic, suggests that AI tools could help family doctors prepare for consultations and potentially influence treatment approaches.

The research involved 98 patients who engaged in conversations with an AI chatbot up to five days prior to their scheduled appointments. During these interactions, the AI provided each patient with a list of possible diagnoses and, in some cases, recommended next steps to discuss with their doctor. Physicians monitored these conversations in real time and received transcripts for review.

Published in The Lancet, the study found that the chatbot’s primary diagnosis suggestion aligned with the final clinical diagnosis in more than 50% of cases. When considering the chatbot’s top three suggestions, the correct diagnosis appeared in 75% of cases. These results indicate that patient-facing AI tools can participate in meaningful clinical dialogues within existing healthcare workflows.

However, the researchers also identified some limitations and risks. In one instance, an AI-generated suggestion of cancer as a possible diagnosis caused significant anxiety for the patient, leading a doctor to classify that response as “somewhat harmful.” Another conversation revealed an example of the chatbot generating inaccurate, or “hallucinated,” information. Due to these concerns, the researchers emphasized that while AI chatbots can support clinical conversations, the current evidence does not justify their autonomous use for diagnosis, triage, or management decisions.

The study highlights the potential for AI to augment healthcare delivery by facilitating earlier identification of symptoms and streamlining patient-doctor interactions. Nonetheless, experts caution that further research and development are necessary to ensure the safety, accuracy, and reliability of AI systems before broader clinical implementation.