SymptomAI: Google Research Develops Conversational AI Agent for Primary Symptom Triage

Google Research presented SymptomAI, a conversational AI agent for end-to-end symptom interviews and differential diagnostic assessment. In a randomized national-scale study with 13,917 participants, five variants of SymptomAI based on Gemini Flash 2.0 were tested. Two weeks after interaction, participants reported diagnoses from healthcare providers. A clinical expert panel reviewed transcripts and ranked SymptomAI’s differential diagnoses; clinicians preferred SymptomAI’s top-5 diagnoses over those from other clinicians more than 50% of the time. All agent-driven prompting strategies outperformed the user-driven baseline. SymptomAI’s top-1 diagnosis for infectious diseases aligned with physiological trends from Fitbit data, including changes in heart rate, respiration, skin temperature, and sleep. The study highlights the potential of AI for scalable symptom assessment, though limitations include reliance on static transcripts and the snapshot nature of symptom reporting.