← All research storiesHealth
A medical chatbot interviewed 98 patients before their visits, and the right diagnosis was on its list 90% of the time
In a real clinic, an AI chatbot took patients’ medical histories before urgent-care visits while doctors watched, with no safety stops and a 90% diagnosis hit rate.
What the scientists did
Before you see a doctor, someone usually asks what’s wrong, when it started and what you’ve tried. Researchers wanted to know whether an AI chatbot could safely do that first interview in a real clinic, not just in tests with actors.
At an urgent primary-care clinic in Boston, adult patients chatted by text with AMIE, a research AI from Google, in the days before their appointment. AMIE asked about their symptoms and history, then wrote a summary and a list of possible diagnoses for their doctor. Human supervisors watched every chat live, ready to stop it if anything unsafe happened.
Words to know
Differential diagnosis
A list of the possible conditions that could explain a patient’s symptoms, which doctors narrow down with questions and tests.
- For a sore throat: a cold, strep throat or allergies
- For a headache: dehydration, migraine or lack of sleep
Feasibility study
An early, small study that checks whether something can work safely in the real world before a bigger trial.
- Testing a new app with about 100 patients
- Checking that doctors and patients are willing to use a tool
Large language model (LLM)
An AI trained on huge amounts of text so it can read and write in everyday language.
- The technology behind chatbots
- AMIE, the medical chatbot in this study
What they found
Across 98 patients, supervisors never had to stop a conversation under the safety rules set in advance. AMIE’s list of possible diagnoses included the patient’s final diagnosis 90% of the time. Doctors said the AI’s summaries helped them prepare in 75% of cases and changed how they approached care in more than half.
In the earlier preprint version of the study, blinded experts rated the AI’s diagnosis lists and care plans as similar in overall quality to the doctors’, but the doctors’ plans were more practical and more cost-conscious.
Why it matters to you
Someday, before a doctor’s appointment, you might describe your symptoms to an AI at home so your doctor walks in already knowing your story and can spend more time talking with you. This study does not mean a chatbot can replace a doctor, and you should never use a chatbot instead of medical care. If you feel sick, talk to a doctor.
Where this could lead
Pre-visit check-ins
Patients could answer an AI’s questions before an appointment, giving doctors a summary to review ahead of time.
3–10 years, after larger trialsMore face time with doctors
If history-taking is partly handled ahead of time, visits could focus more on examining and talking with the patient.
3–10 yearsThese are possibilities the research points toward, not promises. Most early findings take years of testing before they reach everyday life, and some never do.
Why scientists care
Many medical AIs score well on exam-style tests, but real patients are messy, worried and vague. Testing an AI in a working clinic, with safety supervisors, is an important step toward finding out whether these tools actually help patients and doctors.
Keep in mind
This was one clinic, about 100 patients and no comparison group of patients who didn’t use the AI. Every chat was watched by a human, which won’t be true in everyday use. The study was run by Google, which makes the AI, together with the hospital. Larger, independent trials are needed before anything like this is used widely.
How far along is the science?
- Peer-reviewedReviewed by independent experts for a journal before publication. Peer review is a quality check, not a grade of evidence strength: confidence still depends on study design, limitations, replication and the wider evidence.
- PreprintA manuscript shared publicly, often before or alongside journal review. Check its separate journal and review status (each story says what we know). Results may change.
- Conference talkPresented to other scientists at a meeting. Usually short and early; a full paper may come later.
- Agency reportPublished by a science agency such as NASA, NOAA or NIH. Reviewed internally, but not by an outside journal.
- Press releaseAn announcement from a university or company. Useful, but always check the research it describes.
How soon could it reach you?
- Could reach you soonCould show up in products, services or advice soon. Our rough editorial guess is about 1 to 3 years, not a forecast from the research.
- A few years outPromising, but needs more testing or engineering first. Our rough editorial guess is several years, with a lot of uncertainty.
- Long-term scienceBasic research that builds knowledge. Real-world uses, if any, are far off.
- Small stepA small update to earlier work. Useful to researchers, but not a change you would notice yet.
Every story links to its original source: the paper itself, or the conference program when no paper is out yet. Explanations are written with AI, usually from the paper’s abstract (each story says what it was written from), then checked against the source. We only use journals, recognized preprint servers, science agencies, scientific conferences and established science news outlets.