OpenEvidence

Contribution to AI

Point-of-care reference had been a subscription business for three decades: curated topic reviews written by named expert authors, sold to hospitals by the seat. The alternative that appeared in 2023 was a question box — a clinician types a case in ordinary language and gets back a synthesis with inline citations to the papers it rests on, in seconds, free to any verified American clinician, with pharmaceutical advertising covering the cost. Uptake was unusually fast; by mid-2025 the company reported use by more than 40 per cent of US physicians. Two consequences outlast the product itself. The first is contractual. Retrieval systems can only quote what they are permitted to read, and clinical evidence sits almost entirely behind publisher paywalls, so multi-year licences with NEJM Group in February 2025 covering everything published since 1990, followed by the JAMA Network's specialty journals, the NCCN guidelines and Cochrane, showed that a negotiated route to authoritative text existed at a moment when general-purpose model builders were defending scraping suits. The second is uncomfortable and just as instructive. Placing drug advertising beside a machine-generated treatment answer moves promotion to the instant of decision, and the objections raised in the medical journals themselves became an early public argument about commercial influence over AI-mediated clinical judgement — a question no subscription encyclopaedia had forced anyone to ask.

Drafted with AI and edited by hand (claude-opus-5, reviewed 2026-08).