--- title: "In Healthcare's Reference Layer, Hospital Integration Outranks the App and the Corpus" canonical: "https://anchor.is/on-deck-vvSq/issue/17/vertical-ai/article/51079/in-healthcares-reference-layer-hospital-integration-outranks-the" publication: "On Deck" publisher: "Ground Truth" language: "en-US" guide: "https://anchor.is/on-deck-vvSq/llms.txt" format: "Vertical Competitive Read" author: "Reva Sandoval" author_url: "https://anchor.is/on-deck-vvSq/team/143" issue: 17 section: "vertical-ai" section_name: "Vertical AI" published: 2026-10-22 description: "Free AI answer apps, licensed UpToDate content, and hospital contracts ranked: durable advantage goes to whoever integrates hospital protocols inside the EHR." word_count: 2299 source_links: 30 --- # In Healthcare's Reference Layer, Hospital Integration Outranks the App and the Corpus *By [Reva Sandoval]()* *October 22, 2026* [Read on the website]() **The read, if you stop here.** Free AI answer apps hold physicians' attention today, but the questions are starting to move into the EHR (the electronic health record). Expert medical content is now licensed to nearly every surface, so it no longer sets anyone apart. The durable position is the health-system license, and only where that license carries a hospital's own protocols and patient context into the answer. A license that only gates access can be cancelled, and one was in June 2026. This week Wolters Kluwer used its hospital contracts to put UpToDate inside a competitor's app. That is the strongest gate in the market, and so far there is no public evidence that it carries hospital-specific context. - Signal status: **developing**. - Confidence in the ranking: **high**. - Confidence that Wolters Kluwer keeps the top spot: **moderate**. ## The players these events touch | Company | Surface | Who pays | Where answers come from | |---|---|---|---| | **OpenEvidence** | Standalone AI answer app, free to clinicians verified by their NPI (the federal provider ID) | Advertisers; hospital deals starting | Licensed journals: NEJM, JAMA, Wiley/Cochrane | | **Doximity** | Physician network; **Ask** is its AI answer tool | Subscriptions for marketing, hiring and workflow tools, [93% of June-quarter revenue](https://www.sec.gov/Archives/edgar/data/1516513/000151651326000040/docs-20260630.htm); pharma's share not broken out | Its own system, now plus UpToDate at some hospitals | | **Wolters Kluwer** (owns **UpToDate**) | The incumbent paid reference. **Expert AI**, its conversational version, launched October 2025 | Institutions and enterprises, [~80% of UpToDate revenue](https://assets.contenthub.wolterskluwer.com/api/public/content/2971261-2025-09-11-wolters-kluwer-uptodate-presentation-f335034a04?v=147fabc2) per a September 2025 investor deck | Its own expert-written content | | **Suki, Abridge, Microsoft Dragon Copilot** | Ambient scribes that draft the visit note inside the EHR | Hospitals | UpToDate, licensed in; Microsoft also works with OpenEvidence | ## What happened, October 8–20 **The lawsuit ended (October 16).** In June 2025 OpenEvidence sued Doximity, alleging Doximity staff [impersonated physicians](https://www.fiercehealthcare.com/ai-and-machine-learning/openevidence-doximity-reach-settlement-contentious-ai-lawsuit) using their NPI credentials and fed OpenEvidence prompts designed to extract its system instructions. Doximity denied it and countersued for false advertising and defamation. Both sides dismissed with prejudice, so neither can refile, and each paid its own legal bills. No payment, license or testing agreement is public. Doximity's [fiscal 2026 10-K](https://www.sec.gov/Archives/edgar/data/1516513/000151651326000025/docs-20260331.htm), filed in May, names OpenEvidence as a competitor in Marketing Solutions, the business that sells to pharma marketers, and I've seen nothing suggesting the settlement changed that. On its own, the settlement moves little. The underlying fight is over whose answers clinicians trust inside hospitals, which is decided in hospital contracts, and four days later Doximity announced a move built on exactly those contracts. **UpToDate appeared inside Doximity Ask (October 20).** Doximity [announced](https://blog.doximity.com/articles/doximity-and-wolters-kluwer-sign-partnership-to-bring-uptodate-to-doximity-ask) that clinicians whose institution uses both Doximity and UpToDate will see an UpToDate answer alongside Ask's own, "in the coming weeks." [Wolters Kluwer's release](https://www.wolterskluwer.com/en/news/doximity-wolters-kluwer-uptodate-doximity-ask) confirms the UpToDate answer comes from Expert AI. Neither company has said which contracts qualify, how eligibility gets checked, what everyone else sees, or which direction the money flows. Twelve days earlier, Suki launched [its own product called Ask](https://www.suki.ai/press-releases/suki-launches-ask-accelerating-complex-clinical-decision-support-powered-by-wolters-kluwer-up-to-date-expert-ai/), which Suki describes as the first product integration of the UpToDate Expert AI API. It runs inside Epic, Oracle Health, athenahealth and MEDITECH and pulls the patient's chart into the query. Within a month, Wolters Kluwer put the same answer engine into a consumer-style physician app and into the EHR. **JAMA published OpenEvidence usage data (October 15).** The [study](https://jamanetwork.com/journals/jama/article-abstract/2855491) covers January 2024 to June 2026: 952,008 health workers, 556,200 of them physicians, 271 million questions. ## Treat the JAMA figures as usage data Discount this as evidence that the tool works, for four reasons: - **Three of seven authors are OpenEvidence employees**, including its chief medical officer. - **The company supplied the data.** - **JAMA licenses its content to OpenEvidence**, which makes the journal a commercial partner. - **The [editor's note](https://jamanetwork.com/journals/jama/fullarticle/2855492) says plainly** that the study did not assess answer quality or patient outcomes. Repeat use shows clinicians value the tool; it says nothing about whether they're right more often. The headline numbers also count different things from Doximity's. It's the same gap between registered reach and usage depth I flagged in [the legal AI piece](https://anchor.is/on-deck-vvSq/issue/15/vertical-ai/article/42747): | Figure | What it counts | |---|---| | OpenEvidence: 556,200 physicians | Anyone who used it once in 2.5 years, residents included | | OpenEvidence: 703,265 monthly users | June 2026, all clinician types | | Doximity: 85%+ of U.S. physicians | Profiles on the network, **not** Ask users | | Doximity: AI search queries up 25% QoQ | Query growth, no user count disclosed | For scale, 556,200 is roughly half of the 1.03 million practicing U.S. physicians counted by the [Association of American Medical Colleges](https://www.aamc.org/data-reports/data/2025-key-findings). The study counts trainees and the AAMC doesn't, so don't read it as an adoption rate. What the study does establish is bottom-up adoption at national scale, plus the mix of questions. Physicians ask about **treatment** (33.1%) far more than **diagnosis** (13.9%). My inference, moderate confidence: treatment is where institutional protocols most often diverge from general references, through formulary choices, antibiotic stewardship rules and local pathways. If that's right, the question mix argues for the hospital layer. ## Where durable advantage sits, ranked ### 1. The health-system license, as the route to integration *High confidence this is the strongest position. Moderate confidence Wolters Kluwer keeps it.* A hospital license does two separate jobs, and only one of them is hard to cancel. **As an entitlement gate, it decides who sees what.** In the announced rollout, a physician sees UpToDate's answer inside Ask only if her health system uses both Doximity and UpToDate. Whether individual subscribers qualify hasn't been disclosed. The effect is that UpToDate becomes something her institution buys and that follows her into every surface it's licensed to. Wolters Kluwer's [half-year report](https://www.globenewswire.com/news-release/2026/08/05/3339046/0/en/wolters-kluwer-2026-half-year-report.html) says more than 90% of its U.S. Enterprise Edition customers, roughly 2,500 hospitals, had "signed up to adopt" Expert AI by July 31, 2026. Read that wording as commitment, not deployment. A hospital can walk away from a gate. In June 2026 the University of Arkansas for Medical Sciences (UAMS) [let UpToDate lapse](https://libguides.uams.edu/c.php?g=1526391&p=11432929), citing needs and budget, and named OpenEvidence and Doximity its primary point-of-care tools. It's the first public case I've found of free apps displacing the paid reference. Every app that shows UpToDate only to paying institutions adds to the functionality a hospital loses by cancelling. Read Wolters Kluwer's distribution push as retention first. **As integration, it carries the hospital's own context into the answer.** No publisher can license a hospital's protocols or patient records to a rival, and once those run through one vendor's answers, the hospital becomes expensive to leave. This is where OpenEvidence's strongest evidence sits. Cedars-Sinai [gave clinicians enterprise access](https://www.beckershospitalreview.com/healthcare-information-technology/ai/cedars-sinai-gives-clinicians-enterprise-access-to-openevidence/) in May 2026, combining literature with patient-record context, with local protocols planned. Johns Hopkins [announced on October 13](https://www.fiercehealthcare.com/ai-and-machine-learning/johns-hopkins-medicine-taps-openevidence-design-ai-tools-medical-training) that it will co-develop tools drawing on patient histories and Hopkins's own practices. Terms are undisclosed, and it isn't a system-wide deployment yet. I found no public evidence of Wolters Kluwer running a hospital's own protocols through Expert AI. The chart context in Suki's Ask comes from Suki. That gap is why my confidence in Wolters Kluwer stops at moderate. | | Entitlement gate | Hospital-specific integration | |---|---|---| | **Wolters Kluwer** | Strongest: ~2,500 U.S. hospitals signed up | Through partners (Suki's chart retrieval); no local protocols shown | | **OpenEvidence** | Weak: ad-supported; ad-free model in development | Cedars-Sinai live, protocols planned; Hopkins co-development | | **Doximity** | Borrowed from Wolters Kluwer | None disclosed | ### 2. The authoritative corpus *High confidence it's becoming a licensed commodity.* - **OpenEvidence licenses** from [NEJM](https://www.fiercehealthcare.com/ai-and-machine-learning/jama-signs-multi-year-deal-openevidence-inform-ai-powered-medical-search), JAMA and Wiley. Wiley's [SEC exhibit](https://www.sec.gov/Archives/edgar/data/107140/000010714026000009/exhibit992-q326earningsp.htm) discloses a five-year deal including Cochrane, plus a small equity stake. - **The journal deals aren't exclusive.** [Abridge announced](https://www.healthcaredive.com/news/abridge-partners-new-england-journal-medicine-nejm-jama-network-clinical-decision-support/817630/) NEJM and JAMA content partnerships in April 2026, on top of the UpToDate content it was already running. - **UpToDate is live in Suki and [Abridge](https://www.healthcaredive.com/news/abridge-partners-new-england-journal-medicine-nejm-jama-network-clinical-decision-support/817630/)**, coming to Doximity, and part of Microsoft's [Dragon Copilot work](https://www.healthcaredive.com/news/microsoft-expands-dragon-copilot-ai-assistant-nurses/802883/), while Wolters Kluwer also sells Expert AI directly. - **Elsevier went hybrid.** Its [ClinicalKey AI](https://www-prod.elsevier.com/about/press-releases/elsevier-health-launches-clinicalkey-ai-the-most-advanced-gen-ai-powered), launched in 2024, was built in [collaboration with OpenEvidence](https://www.fiercehealthcare.com/ai-and-machine-learning/elsevier-health-unveils-ai-driven-clinical-decision-support-tool) but keeps Elsevier's content inside an Elsevier-branded product: it borrows the answer technology and keeps the corpus behind its own front end. Nobody has shown that an expert corpus produces better answers. A June 2026 [Nature Medicine study](https://www.nature.com/articles/s41591-026-04431-5) had blinded physicians rate answers to 100 real clinician questions. OpenEvidence and UpToDate Expert AI didn't differ significantly, and general-purpose frontier models scored higher. Critics [challenged the methods](https://www.nature.com/articles/s41591-026-04638-6), and the authors partly conceded. The [NOHARM preprint](https://arxiv.org/pdf/2512.01241v4), which isn't peer-reviewed, found tools grounded in curated sources made the fewest severe-harm errors, with no significant spread among them. None of these studies measures patient outcomes. For now, corpora compete on brand and habit. Wolters Kluwer's [annual report](https://assets.contenthub.wolterskluwer.com/api/public/content/3210406-wolters-kluwer-2025-annual-report-pdf-8bb01b4e47?v=ddef685b) lists AI as a risk to demand for its products, and licensing UpToDate outward is how it's responding. ### 3. The answer surface *Moderate confidence.* Two national surfaces own physician habit and chase overlapping pharma budgets. **OpenEvidence** raised $250 million at a [$15 billion valuation](https://www.beckershospitalreview.com/healthcare-information-technology/ai/openevidence-raises-250m-at-15b-valuation/) in September 2026, led by Andreessen Horowitz and Byers Capital, on an ad model. **Doximity** said on its [August call](https://www.fool.com/earnings/call-transcripts/2026/08/13/doximity-docs-q1-2027-earnings-call-transcript/) that it had signed AI search deals but booked no AI revenue in the June quarter. The bigger threat to both is where the question gets asked. Suki's Ask answers from UpToDate inside the EHR with the patient's chart already in the query, while a standalone app makes the physician leave the record and describe the patient again. As more answer engines move into the EHR, the standalone apps are left offering habit without patient context. Ads are a second weak point: they last only as long as pharma keeps paying and hospitals tolerate ads in clinical tools. OpenEvidence CEO Daniel Nadler [told Fierce Healthcare](https://www.fiercehealthcare.com/ai-and-machine-learning/openevidence-launches-hands-free-voice-ai-feature-expands-hospital) in May 2026 that a non-ad-supported enterprise model is in development, which tells you where he expects stable revenue to come from. ## Nobody has said who arbitrates the two-answer screen *Read this one as a practitioner.* When Ask's answer and UpToDate's answer **disagree**, neither company has said what happens: which takes precedence, whether the conflict gets flagged, or what the clinician is supposed to do next. Both describe expert-written source material, but neither says a clinician reviews each generated answer. In [the Figma piece](https://anchor.is/on-deck-vvSq/issue/16/horizontal-ai/article/45954) I argued that recorded and reviewed are different states, and that generated work picks up legitimacy by sitting next to approved work. On this screen the effect runs in both directions. Ask borrows credibility from sitting beside UpToDate, and UpToDate becomes the reference clinicians check Ask against. Each plausible layout picks the arbiter: - **UpToDate leads, Ask supplements.** Wolters Kluwer is the arbiter by design. - **Both shown, disagreement flagged.** The clinician resolves it, probably toward the brand her hospital pays for, and Doximity's misses are visible. - **Both shown, no comment.** Ask keeps apparent parity, which helps Doximity, but the hospital inherits unmanaged conflicts inside a clinical tool. - **Ask leads, UpToDate one tap away.** Best for Doximity, and the version I'd expect Wolters Kluwer's terms or a paying hospital to push back on (speculative). Picking among these is a clinical-governance decision. Watch who makes it: Doximity's product team, Wolters Kluwer's contract terms, or the hospital. ## Who is most exposed, ranked 1. **Doximity, most exposed.** It has the widest reach and no booked AI revenue as of its last report, and inside paying hospitals its edge over OpenEvidence now runs through Wolters Kluwer's customers and terms. A design leader there would be shaping an answer surface whose most credible answer belongs to someone else. 2. **OpenEvidence, structurally split.** Ads and institutional trust pull in opposite directions. Cedars-Sinai accepted the ad-supported product, and the ad-free model is still unlaunched and unpriced. 3. **Wolters Kluwer, exposed to cancellation** until it shows integration of its own. On a CFO's spreadsheet next to a free app, UpToDate looks weakest. ## What to watch - **Doximity rollout screenshots, next few weeks.** Do individual UpToDate subscribers qualify? How are disagreements rendered? - **Doximity's September- and December-quarter results.** Management guided most contracted AI Search revenue to the December quarter. Look for the first booked AI revenue, and whether Doximity discloses Ask users. - **A priced, ad-free OpenEvidence hospital tier.** - **Wolters Kluwer's full-year report.** Whether "signed up" hospitals actually deployed, and any sign of hospital protocols running through Expert AI, directly or via a partner. - **A second UpToDate cancellation**, or UAMS coming back. ## Where this leaves the vertical **Observable now:** the corpus licenses outward, the surfaces duplicate each other, and hospital contracts decide who sees what. **My projection, moderate confidence:** within 18 months the winning position goes to whoever runs a hospital's protocols and patient context through the answer inside the EHR, which is where Suki already puts the question. Today Wolters Kluwer holds the gate, and OpenEvidence is further along on integration. ## Things to follow up on... - **The benchmark methods dispute:** Beaulieu-Jones and Nemati's [critique of the Nature Medicine comparison](https://www.nature.com/articles/s41591-026-04638-6) argues that benchmark contamination and limited real-query scope constrain its finding that general models beat OpenEvidence and UpToDate, and the authors' reply partly concedes the point. - **A preview leaderboard ranking:** The [ARISE MAST technical leaderboard](https://www.arise-ai.org/mast/technical) currently puts Doximity Ask 6.1 at 84.5% and OpenEvidence at 80.0% on its NOHARM v2 measure, but labels itself a preview whose scores may change. - **The first public cancellation:** The UAMS library's [notice on moving to OpenEvidence and Doximity](https://libguides.uams.edu/c.php?g=1526391&p=11432929) shows how one academic medical center justified dropping UpToDate while keeping Lexidrug and Micromedex. - **Wolters Kluwer's own risk language:** The [2025 annual report](https://assets.contenthub.wolterskluwer.com/api/public/content/3210406-wolters-kluwer-2025-annual-report-pdf-8bb01b4e47?v=ddef685b) names generative and agentic AI, changing revenue models, and customer preferences as threats to demand, and describes partnerships into customer workflows as the response. ## More - [Vertical AI — Issue 17]() - [Issue 17]() - [Publication guide]()