Google's SymptomAI: A Conversational AI Agent Built on Gemini Flash 2.0 Assesses Symptoms

On July 22, 2026, Google Research and Google DeepMind introduced SymptomAI, a conversational AI agent built on the Gemini Flash 2.0 model. In a text interview, it asks the patient about their symptoms and produces a list of suspected diagnoses. In a national study of 13,917 participants, clinicians rated its differential diagnosis best in 53.3% of cases.

How SymptomAI works

SymptomAI holds a natural-language conversation with the user: it asks follow-up questions about their complaints and builds a differential diagnosis with recommendations from the answers. In the study, each participant was randomly assigned to one of five agent variants, which differed in how flexible the interview was, from a rigid script to free-form conversation.

• Published: July 22, 2026, by the Google Research and Google DeepMind teams
• Base model: Gemini Flash 2.0
• Participants: 13,917 people, in a national randomized study
• Five interview strategies: Base, Fixed Canonical, Flexible Canonical, Dynamic Live, Dynamic Final
• Two weeks later, participants reported their actual doctor's diagnosis in a survey

How accurate was the agent?

In a blind evaluation of the transcripts, clinicians rated SymptomAI's differential diagnosis best in 53.3% of cases. On the top-5 accuracy metric, experts judged the agent's list more accurate than those from other physicians, and all strategies with active questioning significantly outperformed the Base variant. The authors' key takeaway: much of diagnosis comes down to the conversation itself.

"A significant share of clinical diagnoses can be reached from a language interview alone," the

Google Research publication states.

What do wearables have to do with it?

For some participants, the researchers collected up to 30 days of Fitbit tracker data, recorded before the conversation with the agent. The data show clear physiological shifts, in cardiovascular function, breathing, skin temperature and sleep quality, that preceded the onset of symptoms. This hints that models may one day be able to link a patient's complaints to objective biosignals.

Where the method's limits lie

The authors list the limitations explicitly. A differential diagnosis is inherently ambiguous, and the clinicians in the study evaluated static transcripts, while the agent itself conducted an active interview, so the conditions are not equal. SymptomAI cannot see body language, outward signs or the medical record, and its diagnoses were produced for the study only and are not a confirmed clinical assessment.

What this means

Large conversational models have come close to structured history-taking: on a sample of nearly 14 thousand people, an agent built on Gemini Flash 2.0 beat human physicians on differential-diagnosis quality in half of the cases. It is not yet a replacement for a doctor's appointment, but it is a claim that initial symptom triage can be scaled through chat.

Frequently asked questions

What is SymptomAI?

It is an experimental conversational AI agent from Google Research and Google DeepMind, built on Gemini Flash 2.0: it asks the patient about their symptoms and generates a list of likely diagnoses with recommendations. For now, it is a research project, not a medical service.

Can SymptomAI be used to diagnose a condition?

No. The authors stress that the agent's diagnoses were produced for research purposes only and are not a confirmed clinical assessment. It cannot replace a visit to a doctor.

What model does SymptomAI run on?

The agent is built on Gemini Flash 2.0. The study tested five interview variants, from a rigid script (Fixed Canonical) to free-form dynamic conversation (Dynamic Live, Dynamic Final).