Fiscal second-quarter results put Doximity’s health system customers above 190, up from 165 a quarter earlier, with 10 of the 20 biggest US hospitals on the roster.
That pushes the medical social network past the ambient documentation that made it a clinical AI name, into decision support, where a wrong answer carries clinical risk.
Prompting the chart instead of reading it
NewYork-Presbyterian now runs Doximity’s tools inside electronic health records at 10 hospitals and hundreds of sites. An early adopter is UC San Diego Health, whose new patient context feature pulls laboratory results, notes and records to answer questions about that specific patient.
The measurable effect shows up in how doctors type. Prompts written by doctors on the feature have shrunk to an average of 12 words from 24, since the tool fetches key history on its own rather than being handed it.
Selling pharma on physician search
The commercial logic is separate from the clinical one. Doximity built an AI Search advertising product around the queries clinicians run, and 16 of the 20 largest pharmaceutical companies have bought it. Bookings have reached eight figures, though that is not recognized revenue, and the company expects gradual growth through 2027.
Chief executive Jeff Tangney sized the target at roughly $14B in annual US pharmaceutical paid search spending, money until now claimed by general web search engines.
Guarding against its own models
Doximity has already published evidence of the risk it is taking on. Testing across nine AI models turned up wrong medication dosing advice in at least 2.3% of responses, and roughly 44% of those mistakes were serious or potentially lethal. That produced DoseCheck, a layer built to catch dangerous dosing before a physician sees it. The company also folded Wolters Kluwer’s UpToDate reference content into the product.
