AI in General Medicine
In general medicine, the AI most residents meet first is the large language model: for drafting notes, summarising literature and checking a differential. Ambient scribes that write the clinic note from the conversation are in wide use in the US and spreading. Trials so far show mixed gains in diagnosis accuracy, and these tools still make confident errors.
What AI does in General Medicine
Residents quiz themselves with an LLM acting as examiner. Check every answer against a textbook.
Examples: General chat LLMs
An LLM suggests diagnoses from a case summary. Useful as a second look on unusual cases.
Examples: OpenEvidence, Glass Health, general chat LLMs
Software turns the consultation audio into a structured note for you to edit.
Examples: Abridge, Heidi
A medical search tool answers a clinical question with cited papers and guidelines.
Examples: OpenEvidence, ESC Chat (cardiology)
An LLM turns the chart into a first-draft summary that the doctor checks.
Examples: Hospital-integrated LLM tools
Models read vitals and labs in the EHR and flag patients at risk. Results vary a lot between hospitals.
Examples: Prenosis Sepsis ImmunoScore, EHR vendor models
routine = in everyday use in many centres · early adoption = in use at some sites, evidence still building · research = studies only
Key ideas for this specialty
New to the basics? Start with the theory lessons.
Tools
AI scribe and clinical assistant for notes and letters.
Regulatory status: Site lists HIPAA, GDPR and regional privacy compliance; no medical-device clearance stated
Enterprise ambient documentation for health systems.
Regulatory status: Not listed
Indian imaging AI company (chest X-ray, head CT); relevant to every physician reading chest films.
Regulatory status: Site states FDA cleared and CE marked
AI on a digital stethoscope for murmurs, AF and low ejection fraction.
Regulatory status: Site states 9 FDA clearances
Clinical AI for mental health intake, useful as an example of a regulated chatbot.
Regulatory status: Reported as Class IIa UKCA medical device (search result)
A listing here is not an endorsement. Ask any vendor for its CDSCO licence and class before clinical use.
Where to learn
Hands-on diagnosis, prognosis and treatment models; needs Python and some deep learning background.
India's national ethics framework; useful before you put any AI tool into a thesis or ward workflow.
Interviews with people building and testing medical AI; hosts are Raj Manrai and Andrew Beam.
Weekly digest of new AI and medicine papers from a practising cardiologist.
Five short courses that teach clinicians how to judge AI tools; CME available for physicians.
Indian course built for clinicians, from IIT Madras faculty.
Do it yourself
Write 10 de-identified case summaries from your own ward with known final diagnoses. Ask an LLM for a top-3 differential and score it. Note where it is confidently wrong.
Open the dataset or tool →Open the free demo of MIMIC-IV-ED on PhysioNet and predict admission from triage vitals with a simple logistic regression, then compare with a random forest.
Open the dataset or tool →Use public or properly de-identified data only, and get ethics approval before any project on your own patients.
Papers worth reading
- Large Language Models in Healthcare and Medical Applications: A Review Bioengineering, 2025
- Large Language Model Influence on Diagnostic Reasoning JAMA Network Open, 2024
- Ambient AI Scribes in Clinical Practice: A Randomized Trial NEJM AI, 2025
- Automation Bias in Large Language Model–Assisted Diagnostic Reasoning among Physicians Trained in AI Literacy — A Randomized Clinical Trial NEJM AI, 2026
Societies and guidance
2023 ethical guidelines for AI in biomedical research and healthcare.
Recent updates
GPT-4o raised doctors' scores in a three-country trial, but the control group had no references
A randomised trial of 249 physicians in Indonesia, Kenya and the Netherlands (npj Digital Medicine) found GPT-4o access raised vignette scores by 7 to 18 percentage points. The control doctors were not allowed guidelines or internet searches.
Source: iatroxChest X-ray AI did not improve resident accuracy in a prospective crossover study
In Academic Radiology, five radiology residents read 1,861 chest radiographs with four commercial AI tools. Accuracy did not improve overall and fell for pleural effusion and nodules because of false positives, though three of five residents read 6 to 17 seconds faster.
Source: Omnicuris MedshotsFDA picks first company for its TEMPO digital health pilot
The FDA chose Dexcom as the first participant in TEMPO, a pilot with CMS to gather real-world evidence on digital health devices, including AI insights, for chronic disease care. It plans up to ten more participants in each of four clinical areas.
Source: US FDANature Medicine: LLM decision support in Kenyan clinics improved care quality but not short-term outcomes
A cluster-randomised trial of an LLM tool ('AI Consult') inside the record system, in 16 primary care clinics with over 9,600 patients, improved documentation and treatment planning. Treatment failure was 2.2% with the tool and 2.0% without, and no harm signal was seen.
Source: University of BirminghamJAMA Network Open: LLM feedback raised sepsis bundle compliance in two emergency departments
A cluster-randomised trial at two UC San Diego emergency departments used an LLM to check SEP-1 sepsis care and give feedback. Compliance was 82.9% versus 70.1%, with no significant difference in mortality or ICU admission.
Source: JAMA Network OpenGoogle's AMIE tested with real patients before primary care visits
In a 100-patient feasibility study at Beth Israel Deaconess, patients text-chatted with AMIE before appointments under live physician supervision. No safety stops were needed, the final diagnosis was in AMIE's top 7 in 90% of cases, and it was the single top diagnosis in 56%.
Source: Google ResearchLinks checked 3 October 2026. “Not listed” means we could not confirm the detail, not that it is absent. Educational material, not clinical advice.