ArticleNature medicine2026
ChatGPT Health performance in a structured test of triage recommendations.
Article in Nature medicine, 2026. The graph could read no effect estimate from its abstract, so it casts no vote on the map. Cited by 19 papers.
What it found
Each row is one number read from the abstract, on the scale the paper reported it, with its interval. Left of the dashed line favours the treatment, right favours the comparator. Under each row is the sentence it came from. New to these charts? A ten-minute tutorial.
The abstract states no effect estimate the extractor could read, or names no intervention and outcome on the map, so this paper lights no cell and moves no belief. It is still indexed, cited and linked below.
The trial behind it
Trials whose registry record cites this paper, or whose number appears in the abstract. A trial that started after this paper was published is citing it as background, not reporting it.
Neither the registry nor the abstract names a trial number. If this is a trial report, that itself is worth knowing.
Who cites it
19 citing papers in PubMed.
- Using AI to offer personalized health advice.Nature · 2026Article
- Use of Generative AI for Substance-Related Information Among Young Adults (Aged 18-29 Years) in the United States: Cross-Sectional Survey.JMIR formative research · 2026Article
- An Evidence-Based Framework for Patient-Facing Artificial Intelligence Integration in the Emergency Department.Journal of the American College of Emergency Physicians open · 2026Article
- Closing the Solidarity Gap Requires Closing the Accountability Gap for Patient-Facing AI.Journal of medical Internet research · 2026Article
- Toward a test of medical AI superintelligence.Nature medicine · 2026Article
- Artificial intelligence in psychiatric care and education: a qualitative study of factors influencing adoption in Singapore.International journal of medical education · 2026Article
- Patient-Facing AI System for Symptom Guidance Using Simulated Encounters and Physician Review: Analytical Validation Study.JMIR formative research · 2026Article
- Deployment protocol and base model jointly determine residual risk in rule-based clinical LLM supervision.Research square · 2026Article
- Health system learning enables generalist neuroimaging models.Nature medicine · 2026Article
- Article
- Digital Mental Health Research Priorities, Revisited for the AI and Large Language Model Era.JMIR mental health · 2026Article
- Solidarity or Segregation? ChatGPT Health and US Health Care Disparities.Journal of medical Internet research · 2026Article
- The evolving physician-AI relationship: a five-tier framework for integrating intelligent systems into clinical practice and medical education.ESMO real world data and digital oncology · 2026Review
- Article
- Article
- Synthetic gatekeeping in health chatbots: configured answer systems as the unit of AI evaluation and governance for medical information.Frontiers in digital health · 2026Article
- Managing vulnerability: cross-sector reflections on the ethics and equity implications of AI use among young adults with chronic conditions.Frontiers in digital health · 2026Article
- Emerging shadow health systems: regulating health-focused generative AI chatbots from a European perspective.Frontiers in public health · 2026Article
- Artificial intelligence in dementia care: challenges, controversies, and policy implications.Frontiers in dementia · 2026Article
Corrections and comments
PubMed lists nothing against this paper. Absence here is not a guarantee, only a check that was made.
Authors and funding
19 authors.
Funding
No grant is acknowledged in the PubMed record.
Abstract
ChatGPT Health was launched in January 2026 as OpenAI's consumer health tool and has reached millions of users. Here we conducted a structured stress test of triage recommendations using 60 clinician-authored vignettes across 21 clinical domains under 16 factorial conditions, yielding 960 total responses. Performance followed an inverted U-shaped pattern, with the most dangerous failures concentrated at clinical extremes-nonurgent presentations (35%) and emergency conditions (48%). Among gold-standard emergencies, the system undertriaged 52% of cases, directing patients with diabetic ketoacidosis or impending respiratory failure to 24-48 h evaluation rather than the emergency department, while correctly triaging classical emergencies such as stroke and anaphylaxis. When family or friends minimized symptoms, indicating anchoring bias, triage recommendations shifted significantly in edge cases (odds ratio = 11.7, 95% confidence interval = 3.7-36.6), with the majority of shifts toward less urgent care. Crisis-intervention messages activated unpredictably across suicidal ideation presentations, occurring more frequently when patients described no specific method than when they did. Patient race, sex and barriers to care did not show significant effects, although confidence intervals did not exclude clinically meaningful differences. These findings reveal missed high-risk emergencies and inconsistent activation of crisis safeguards, raising safety concerns that warrant prospective validation before consumer-scale deployment of artificial intelligence triage systems.
Indexed as
Identifiers
What OpenQuestion holds
Registered trials
Read under generation 80e0d062 · epoch 390. Bibliography from PubMed, PubMed Central and OpenAlex; grants from NIH RePORTER; trial links from ClinicalTrials.gov; estimates, votes and beliefs from the OpenQuestion graph.