Wave 2 of the Algorithmic Authority Index measured Share of Model across 120 healthcare B2B technology vendors on ChatGPT, Perplexity, and Google AI Overview. 86% receive zero citations, the highest rate in the Index. The platforms treat healthcare as a your-money-or-your-life domain, and the trust thresholds show it.
Healthcare is the Index's first your-money-or-your-life vertical. The platforms apply stricter trust behavior here than anywhere else tested. They refuse, they disclaim, and they demand clinical evidence as the price of a citation. That changes what visibility costs and who can afford it.
Wave 2 evaluated 120 vendors across four tiers. Tier 1 holds ten legacy incumbents including Epic, Oracle Health, and Optum. Tier 2 holds ten mature healthtech pure-plays. Tier 3 holds ten specialized clinical AI vendors. Tier 4 holds 90 emerging B2B startups.
The study executed 25 buyer-intent queries per sub-category across 12 sub-categories: clinical AI and decision support, ambient clinical documentation, healthcare NLP and data platforms, revenue cycle management, EHR and interoperability, patient engagement and intake, telehealth infrastructure, remote patient monitoring, medical imaging AI, clinical trials technology, compliance and credentialing, and dental practice technology.
Queries simulated six buyer personas, from Chief Medical Information Officers to dental practice owners, across clinical and administrative workflows. The configuration yielded 900 documented query sessions across ChatGPT, Perplexity, and Google AI Overview.
Two measurements are new to this vertical. First: platform refusal and disclaimer rates on clinical queries. Second: a clinical-evidence authority test measuring whether peer-reviewed publications or funding announcements move citations. Six findings surface across the layers of the Algorithmic Authority Stack.
Share of Model measures the percentage of AI responses in a query set that name a specific vendor. In healthcare the metric splits by workflow type. Epic and Oracle Health dominate EHR queries, and Optum and Waystar lead revenue cycle. The clinical categories belong to specialized pure-plays with dense published evidence.
The clinical side rewards evidence density. John Snow Labs takes healthcare NLP at 24.5% Share of Model on dense, published clinical evidence. Microsoft Dragon Copilot leads ambient documentation at 18.5% on deep EHR integrations. Weave wins dental practice technology at 18.2% on explicit care-setting specificity.
| Sub-Category | Rank 1 Winner | SOM | Rank 2 Winner | SOM |
|---|---|---|---|---|
| Clinical AI / Decision Support | Epic | 12.4% | OpenEvidence | 8.2% |
| Ambient Clinical Documentation | Microsoft Dragon Copilot | 18.5% | Abridge | 14.2% |
| Healthcare NLP / Data Platforms | John Snow Labs | 24.5% | Oracle Health | 4.8% |
| Revenue Cycle Management | Optum | 15.2% | Waystar | 11.8% |
| EHR / Interoperability | Epic | 32.1% | Oracle Health | 18.4% |
| Patient Engagement / Intake | Phreesia | 14.6% | Tebra | 7.2% |
| Telehealth Infrastructure | Teladoc Health | 22.4% | Doximity | 12.8% |
| Remote Patient Monitoring | Philips | 11.2% | GE HealthCare | 8.5% |
| Medical Imaging AI | Siemens Healthineers | 13.5% | Philips | 9.1% |
| Clinical Trials Technology | Veeva Systems | 28.4% | McKesson | 4.2% |
| Compliance / Credentialing | Definitive Healthcare | 6.5% | Veradigm | 4.1% |
| Dental Practice Technology | Weave | 18.2% | Tebra | 4.5% |
A clinical buyer asks AI for a shortlist. AI answers with the vendors holding the deepest published evidence and the oldest institutional footprints. The startups building clinical AI are invisible in the category they created.
L1 L5 Category Absence means a Share of Model score below 1%. In healthcare, 86% of tested vendors sit below that line, the deepest deficit in the Index to date. Among Tier 4 startups the rate reaches 92%, and 100% of Tier 4 startups fail Layer 4 on training-ready content.
The mechanism is compounded trust. Healthcare buyers cluster around validated vendors, and the platforms mirror that caution with their own YMYL behavior. ChatGPT refused 42% of clinical queries on safety grounds. Google AI Overview attached disclaimers to 56%. A conservative platform names fewer vendors, and the ones it names carry evidence.
Funding does nothing to offset it. Heavily funded startups sit at zero visibility while clinically-validated vendors outperform. The authority currency in clinical categories is peer-reviewed evidence. Ventures carrying the DiMe Seal average 3.9 peer-reviewed publications against a US ecosystem average of 1.3. The citation pattern follows the publications.
The counter-pattern is vertical specificity. Vendors with explicit care-setting positioning achieve 31% higher Share of Model than horizontal generalists. Weave wins dental practice technology queries on exactly that signal while horizontal patient-communication platforms collapse into generic results.
Startups deploying horizontal, abstract language fail to establish a stable identity node. That is a structural failure at Layer 1 (Market Identity Clarity) and Layer 5 (Algorithmic Touchpoint Presence). 82% of tested vendors compound it by relying entirely on their owned website.
A funded clinical AI startup with a horizontal pitch loses the citation to a 40-year incumbent. A niche vendor with one explicit care-setting signal and published evidence beats both inside that setting.
L3 Category Confusion occurs when distinct categories collapse into a single entity in AI classification. In healthcare, 71% of clinical queries returned vendors from an adjacent category. In this vertical the misclassification is not only a marketing problem. It is a clinical risk surface.
One session illustrates the stakes. ChatGPT was asked for specialized ambient clinical documentation software for a 12-provider cardiology group. The engine misclassified the specialized tools and recommended general data integration systems instead of dedicated AI scribes.
A second session asked Perplexity for a clinical decision support system for oncology treatment paths. The system returned an administrative billing platform. Recommending billing systems for oncology treatment paths creates clinical risk under the EU AI Act, not just a lost deal.
The collapse stems from Semantic Drift at Layer 3: 74% of tested companies use inconsistent terminology for their own core concepts across surfaces. The model averages the conflicting signals and files the vendor into the wrong competitive set.
Seven out of ten clinical queries surfaced a vendor from the wrong category. The buyer cannot see the misclassification. In healthcare, neither can the patient downstream of it.
L1 A Zombie Rebrand is a rebrand alive to humans and dead to machines. Healthcare's merger-heavy history makes it the worst vertical measured for this pattern. Across ten tested rebrands, AI engines returned the old name 81% of the time, the highest persistence in the Index.
Cerner is the flagship case. Oracle completed the acquisition in June 2022. 44 months later, Cerner still returns in 84% of relevant queries and dominates ChatGPT clinical queries. The old domain and legacy pages stay indexed, and the model trusts decades of consensus over a corporate announcement.
| Former Identity | New Identity | Old-Name Return Rate | Recognition Lag |
|---|---|---|---|
| Change Healthcare | Optum | 89% | 41 months |
| Cerner | Oracle Health | 84% | 44 months |
| Vera Whole Health | Apree Health | 83% | 42 months |
| Livongo | Teladoc Health | 82% | 21 months |
| Castlight Health | Apree Health | 81% | 42 months |
| MDLIVE | Evernorth | 78% | 58 months |
| Allscripts | Veradigm | 76% | 38 months |
| Ginger | Headspace Health | 75% | 52 months |
| Nuance DAX | Microsoft Dragon Copilot | 71% | 11 months |
| MTBC | CareCloud | 68% | 59 months |
Every case is a Layer 1 Market Identity Clarity breakdown running on the Two-Loop Problem, diagnosed below. The healthcare twist: clinical compliance processes slow content updates, which extends the correction latency past the standard window.
Oracle paid $28 billion for Cerner and has spent 44 months unable to buy the name back from the machines. The rebrand budget cannot outspend parametric memory.
L6 When brand-owned content fails extraction, AI substitutes third-party sources it can extract from. That mechanism is Trust Seed Substitution. Healthcare is the first vertical where the routing splits cleanly in two. Clinical queries route to validated evidence. Administrative queries route to review platforms and communities.
| Substitute Source | Clinical Sub-Categories | Administrative Sub-Categories |
|---|---|---|
| Peer-reviewed literature (PubMed, JAMA) | 42% | 2% |
| FDA & ONC databases | 24% | 1% |
| KLAS Research | 18% | 12% |
| Industry trade press | 13% | 22% |
| G2 & Capterra | 2% | 38% |
| Reddit & peer forums | 1% | 25% |
The split rewrites the playbook. A clinical decision support vendor running a review-platform strategy spends on sources the platforms ignore for its queries. A patient intake vendor ignoring G2 and community forums is invisible on the exact sources that decide its shortlist.
Independent third-party presence increases citation likelihood 2.8 times across the vertical. Without it, vendors face a Trust Gap at Layer 6 that no ad budget closes, because the substitute sources carry editorial and scientific gatekeepers.
In clinical categories the citation graph runs through peer review and regulators. The vendors who never publish evidence have opted out of the graph without knowing it.
L4 AI citation patterns are the inverse of what healthcare vendors publish. Clinical evidence sits in gated PDFs. Marketing copy is compliance-flattened until it says nothing extractable. Bottom-line-up-front structure earned a 34% Share of Model lift in testing. Fluffy narrative text dropped 18%.
| Corporate Output Pattern | Machine Extraction Requirement | Impact on Share of Model | Layer Failure |
|---|---|---|---|
| Gated clinical evidence PDFs | Un-gated, semantic HTML with open schemas | Citations drop toward zero | L4 Citation Invisibility |
| Compliance-flattened marketing copy | BLUF structure with direct answers | +34% BLUF vs -18% narrative | L4 Citation Invisibility |
| Funding announcements as authority | Peer-reviewed clinical evidence | Excluded from clinical queries | L6 Trust Gap |
| Inconsistent name variations | Unified cross-platform entity nodes | Confident misclassification | L1 Identity Fragmentation |
The Volume Paradox from Wave 1 replicates here: more unstructured content decreases machine visibility. 100% of Tier 4 startups fail Layer 4 regardless of publishing volume, because standard blog formatting produces Citation Invisibility that quantity cannot fix. Only 5% of tested teams track their Share of Model, so the other 95% cannot see any of this happening.
The clinical evidence behind your lead-gen form contains exactly the data AI needs to cite you. The gate that captures the lead blocks the citation.
AI systems process healthcare identity across two loops. Loop 1 is Live Retrieval: it updates in two to four weeks by scanning schemas and current web pages. Loop 2 is Parametric Memory: it hardcodes data into model weights and updates only during core training cycles of 18 to 24 months. Healthcare compliance processes push corrections toward the far end of that window.
A rebrand breaks visibility in both loops simultaneously, and the sequence above is the only order that works. Schema without bridge pages leaves retrieval ambiguous. Bridge pages without third-party coverage never reach parametric memory. This is Parametric Decay: every quarter of delay hardens the old identity deeper into the next model generation.
The healthcare vendor that starts the displacement campaign this quarter enters the next training cycle under its own name. The one that waits gets re-memorized as its former self.
The Healthcare Industry Snapshot names the pattern at category scale. The AI Visibility Snapshot names the pattern for your specific brand. Your Share of Model, the layer failure producing your Category Absence, and the structural gap between you and the vendors holding your slot.
Wave 2 uses the same measurement framework as Wave 1 of the Algorithmic Authority Index, adapted for category-level analysis. A standard prompt library of 25 buyer-intent queries was designed per sub-category, covering direct shortlist prompts, comparison questions, and problem-language queries. Queries simulated six buyer personas across clinical and administrative workflows. All sessions were executed from US-based IPs between January 15 and March 1, 2026, using default platform models. Two vertical-specific measurements were added. The first: platform refusal and disclaimer rates on clinical queries. The second: a clinical-evidence authority test comparing peer-reviewed publications against funding as authority currency.
Share of Model was calculated by aggregating Mention Share, Citation Share, and Recommendation Share per vendor. Category Winners hold a composite score above 5%. Category Absence means a composite score below 1%. The resulting dataset comprises 900 fully documented query sessions across 120 vendors and 12 sub-categories.
Category sources: KLAS Research coverage patterns. PubMed and JAMA citation routing. FDA and ONC database records. Digital Medicine Society (DiMe) Seal publication data. Oracle-Cerner acquisition records and legacy brand documentation. Practitioner commentary: Kateryna Lee (K Digital, John Snow Labs).