OpenEvidence built its reputation answering clinical questions. Its most interesting recent moves are about something else: teaching. Free CME credits earned from the questions physicians actually asked, professional society education distributed inside the answer flow, and now EvidenceGrade, a feature that shows clinicians how much weight the evidence behind each answer can bear. The most widely used medical AI platform in the world is becoming, quite deliberately, an education platform too.
One access note before the analysis. OpenEvidence withdrew from the UK and EU in April 2026 and remains unavailable to UK clinicians as of July 2026, a gap we examined in our withdrawal analysis. The education strategy described here is therefore something UK doctors can learn from rather than log into, which makes the underlying question, how AI changes medical learning, more important here, not less.
What OpenEvidence has actually announced
The education layer has been assembling in public. In April 2025 the platform introduced free continuing medical education for verified US clinicians: physicians review the clinical questions they asked, complete short learning assessments, and receive credit transcripts, turning point-of-care curiosity into documented learning. Society partnerships followed, and they go beyond guideline licensing: the American College of Emergency Physicians agreement, for example, brings ACEP clinical policies, point-of-care tools and educational materials directly into the platform, alongside relationships with bodies including NCCN, ACC, ADA and AAFP. A March 2026 partnership with Cochrane put gold-standard systematic reviews in the workflow. And in July 2026 came EvidenceGrade, which grades and visualises, in real time, the certainty of the published evidence behind each answer, built on the GRADE framework used by Cochrane, the WHO and most major guideline developers.
Why a search company teaches
The strategic logic is sound. A clinical question is a revealed learning need: the moment a doctor asks about a drug interaction is the moment that doctor is maximally receptive to learning about it. Capturing that moment converts a utility into a habit, deepens engagement, and creates products, CME among them, that societies and sponsors value. It also answers a competitive truth the whole category is waking up to: answer quality is converging, so the durable differentiation shifts to what the platform does for the clinician's development over time.
Learning and deciding are different jobs
Here is where doctors should keep their thinking sharp. Point-of-care decision support and education overlap but are not the same activity. Decision support optimises for the fastest safe resolution of today's question. Education optimises for the clinician's performance on the next hundred questions, including the ones never asked. A platform can serve both, and OpenEvidence clearly intends to, but the mechanisms differ: reading an excellent answer resolves uncertainty, while durable learning requires retrieval, feedback, spacing and effort, none of which happen automatically when an answer is consumed.
The risk of learning exclusively through search
Cognitive science is blunt about this. Recognising a good answer feels like knowing it, and that feeling is unreliable. Retrieval practice, being made to produce the answer yourself, outperforms re-reading for long-term retention, which is why exam candidates who read extensively can still underperform candidates who tested themselves relentlessly. A clinician whose entire learning diet is fluent AI answers risks accumulating familiarity rather than capability. That is not an argument against AI search; it is an argument for pairing it with deliberate practice, a distinction we unpack in Searching Isn't Studying.
Where deliberate practice and Socratic methods fit
The educational forms that survive this transition will be the effortful ones: question banks that force retrieval, spaced repetition that schedules it, and tutoring that asks before it tells. Socratic questioning in particular converts an answer engine's weakness into a strength, because the value is no longer in delivering the answer but in making the learner reconstruct the reasoning toward it.
The commercial layer underneath
It is worth being clear-eyed about why the education layer is commercially attractive, because the incentives shape the product. CME and CPD are budgeted, recurring obligations, which makes education one of the few places a free-to-clinician platform can build revenue that does not depend on advertising alone: accredited modules can be paid for directly or underwritten by sponsors in exchange for engagement, and professional societies gain a distribution channel that reaches clinicians at the exact moment of relevance. The same content agreements that power the answers, NEJM, JAMA, NCCN, Wiley and Cochrane among them, become the raw material of teaching. None of this is sinister; it is the standard economics of medical education moving to a new venue. But it does mean clinicians should keep asking the old questions in the new setting: who funded this module, and does the learning serve my gaps or someone's reach?
What this means in the UK
UK clinicians cannot currently use OpenEvidence, but the direction it signals applies fully here: clinical AI and education are merging, and the platforms worth adopting will treat both seriously. iatroX was built on that premise from the start, pairing Ask iatroX, a clinical question tool grounded in UK national guidance, with adaptive question banks, spaced repetition and a Socratic Tutor that diagnoses the misconception behind a wrong answer rather than simply displaying the correct one. The education turn OpenEvidence is making in the US is, in that sense, a validation of building learning and answering as one system.