Glass Health

Glass Health sits in the narrow overlap between clinical decision support and ambient AI scribing, and it handles both better than most tools that try to do either alone. The core trick is its knowledge graph backbone: when you feed it a patient presentation, it doesn't just pattern-match against GPT's training data. It grounds differential diagnoses in structured medical ontologies, which means fewer hallucinated zebra diagnoses and more defensible reasoning chains. I've used it primarily for differential generation during case reviews. You drop in age, presenting complaint, pertinent positives and negatives, and it returns a ranked list with likelihood estimates and supporting evidence. The autonomy here is real: it iterates through diagnostic trees without you steering each branch, pulling in epidemiology, symptom clusters, and red flags. Compare that to ChatGPT with a clinical prompt, where you're constantly course-correcting and second-guessing references. Glass cites UpToDate, guidelines, and peer-reviewed sources inline, which saves the ten minutes you'd otherwise spend validating every claim. The ambient scribe feature listens to patient encounters and drafts SOAP notes in real time. It's faster than Dragon, more structured than Otter, and it actually understands medical shorthand. But it's not flawless: complex multi-problem visits sometimes get flattened into a single assessment, and you'll still need to edit for nuance. I'd reach for it when I need a first draft that doesn't require a full rewrite, not when I need publication-ready documentation. Failure modes: it struggles with rare presentations outside the knowledge graph's coverage, and the free tier rate-limits you hard enough that it's only useful for occasional consults. The mobile app is serviceable but the web interface is where the real workflow lives. Competitor-wise, UpToDate's diagnostic tools are more exhaustive but require more manual navigation; Suki and Nuance DAX handle scribing but lack the diagnostic reasoning layer. Glass is the better pick if you want both in one place and you're willing to pay for the integrated workflow. One concrete use case: I ran a case of atypical chest pain through it mid-consultation. It flagged costochondritis and GERD as expected, but also surfaced herpes zoster as a differential I'd missed, with a cited incidence rate for the patient's age group. That kind of safety net is worth the subscription if you're in primary care or urgent care settings where breadth matters more than depth.