Rejoy Health

The one axis where this record is unambiguous. Every product line is a language model application: chat over clinical questions, generated documentation, a voice agent taking patient calls, real time interpretation, generated treatment suggestions, search and synthesis, and examination practice. Remove the model and nothing remains, not a content library, not a workflow tool, not a device.

That matters for how this record should be read. This index rejects vendors whose artificial intelligence turns out to be absent, and four names on the same source list were rejected on exactly that ground. This one passes that test cleanly and is graded low on other axes instead. The scope test asks whether there is an artificial intelligence product; it does not ask whether the vendor is a strong one. Those are different questions and this record keeps them apart.

No oversight mechanism of any kind was described in two retrieval passes, on a product surface that carries more autonomy than anything else in this category.

The assistant is said to give journal backed answers, but no citation mechanism is described, no source linking is shown, and nothing states whether a clinician can trace a statement to the document it came from. There is no abstention behaviour, no confidence signal, no human review layer and no statement of what happens when the system does not know.

The surface is the concern. A decision support module offers diagnostic insights, treatment suggestions and recommended laboratory tests. An artificial intelligence receptionist answers patient calls, books appointments and answers questions without a clinician present. A symptom checker sits in the public resources alongside the clinician tools with no described boundary between the two audiences. Each of those acts carries a different and higher oversight requirement than a reference lookup, and none is addressed.

No disclaimer, scope statement or limitation of any kind was located on the product material. Two other vendors in this category publish precise statements of what their tools do not do; one states plainly that it does not account for variation among patients. Nothing comparable exists here.

The only technical disclosure is a benchmark result, and it does not survive examination.

The homepage reports 99.4 percent accuracy on a named medical question answering benchmark derived from United States licensing examination material, presented above three named competitor models at 97, 89 and 88 percent. Naming the benchmark is more specific than several vendors in this category manage, and that is credited. Everything a reader would need in order to believe it is absent: no methodology, no evaluation date, no sample or split described, no citation, no paper and no independent replication. The competitor figures carry no source either.

The figure also sits far above published results for that benchmark, which makes an unsourced claim harder rather than easier to accept. This index has recorded the same shape elsewhere and treats it consistently: a precise comparative number that no reader can reproduce is the weakest form of transparency, because it invites a comparison while providing nothing to check it against.

Beyond that number there is nothing. No model class, no provider, no retrieval or grounding architecture, no training or fine tuning description and no evaluation harness. A vendor asserting the most accurate healthcare artificial intelligence has published less about how its system works than any other record in this category.

Nothing measuring clinical performance was located in two retrieval passes. No peer reviewed publication, no independent evaluation, no accuracy study, no error rate, no named health system deployment and no third party assessment.

The quantified claims that exist are unsourced. A benchmark figure with no methodology. A saving of 12 hours per week of administrative work with no study, cohort or measurement described. And a count of more than 20,000 professionals at leading centres, displayed above the marks of eight prominent academic medical centres.

That count deserves a specific note because it moved in the wrong direction. Material published in 2025 claimed use by more than 50,000 physicians; the current homepage claims more than 20,000 professionals. Both are vendor figures, they are not reconcilable, and a user count that falls by more than half while the marketing grows more confident is the kind of inconsistency this index records rather than resolves.

An institutional mark on a customer panel is not evidence of an institutional deployment. Individual clinicians registering for a tool are not their employer becoming a customer, and nothing published distinguishes the two.

This product touches more patient data than anything else graded in this category, and describes its handling least.

A scribe generates clinical notes from patient conversations. A receptionist answers patient calls and books appointments. An interpreter carries clinical conversation across languages in real time. Each of those processes identifiable patient information by design, and together they represent a substantially wider exposure than a reference lookup.

What is published against that is a hero line stating the product is compliant with the federal health privacy rule and that patient data is protected, and a feature panel stating that data stays protected with strong security and healthcare level privacy standards. Neither names a control. No retention period, data residency position, encryption practice, recording policy for calls, or statement on whether clinical inputs are used to improve the models was located.

Graded on what was found rather than as a finding that controls are absent. A privacy policy and terms of service are published and were identified but not opened this pass; both are where a fuller answer would sit, and both should be read on refresh. The question to ask first is what happens to a recorded patient call and a generated note after the encounter ends.

An assertion exists and nothing sits underneath it. The homepage states the product is compliant with the federal health privacy rule, positioned as a headline reassurance alongside claims about accuracy.

No business associate agreement was located, no terms of such an agreement, no covered entity or business associate language, no execution path and no statement of the tier at which an agreement becomes available.

That gap matters more here than for a pure reference product. A scribe and a voice agent handling patient calls place the vendor squarely in a business associate relationship with any practice that deploys them, so an agreement is not optional for the customer even if it is unmentioned by the vendor. A clinic adopting this product without one would be the party carrying the exposure.

Graded on what a buyer can find. The privacy policy and terms were identified and not opened this pass and could carry relevant commitments. Ask for the agreement before any pilot involving real patients, and note that the smallest vendor graded in this category publishes its agreement in full at a public address.

No certification was located in two retrieval passes: no SOC 2 report of either type, no ISO 27001, no HITRUST, no trust centre, no penetration testing statement, no vulnerability disclosure programme and no named encryption practice or hosting provider.

What exists is a feature panel asserting strong security and healthcare level privacy standards without naming a standard, a control or an auditor.

One operational detail is recorded without being graded, because it is an observation rather than a control. The published support route is a consumer messaging application, alongside a general enquiries email address. That is an ordinary choice for a small company and it is a notable one for a vendor whose products generate clinical notes and handle patient telephone calls, since support conversations about those products will tend to involve specifics.

Recorded as a retrieval outcome rather than as a finding that controls are absent, and flagged for refresh.

The weakest regulatory position in this category, and the reason is not the absence of a clearance but the absence of any statement at all against a product surface that plainly calls for one.

No clearance, no submission, no device analysis, no disclaimer, no intended use statement and no limitation of any kind were located. Against that, the published product set includes a decision support module offering diagnostic insights, treatment suggestions and recommended laboratory tests, and a symptom checker made available among general resources.

Diagnostic insight and treatment suggestion are patient specific outputs, which is the act that separates decision support from reference, and this index has treated that boundary as decisive throughout the category. A symptom checker reaches further still, since it is used without a clinician interposed at all, and nothing published states who it is for or what it is not for.

The contrast within the category is direct. A professional society publishing a comparable assistant states plainly that its tool is a voluntary educational resource that does not account for variation among patients and does not substitute for professional judgement, and was graded up for it. The same statement, or any statement, is available to this vendor and has not been made.

No governance artefact of any kind was located: no artificial intelligence principles, no responsible use framework, no clinical advisory board, no named clinical leadership, no external validation membership, no bias evaluation, no subgroup analysis, no error taxonomy and no evaluation methodology.

Two presentational choices belong on this axis because they concern how the vendor represents itself to a clinical buyer, and both are described here strictly as displayed rather than characterised.

A panel headed with a statement about being backed by world class technology leaders displays the marks of four major technology companies. No investment, partnership or commercial relationship with any of them was located in two passes, and the nature of each relationship is unstated. Programmes offering credits or tooling to early stage companies are common and are not backing in the sense the heading conveys.

A second panel displays the marks of eight prominent academic medical centres above a count of professionals using the product. Nothing distinguishes individual clinicians registering for a tool from an institution adopting it, and no institutional customer was named or verified.

A buyer evaluating a clinical artificial intelligence product needs the vendor's claims about itself to be as reliable as its claims about medicine. Establish the nature of each displayed relationship in writing before proceeding.

No integration of any kind was located: no named electronic health record, no marketplace listing, no application programming interface, no conformance statement for a health data exchange standard, no practice management or scheduling system and no published integration path.

This is a capability gap rather than a disclosure gap, which is why it grades below the vendors in this category that simply do not integrate. A reference tool used alongside the record loses little by standing apart from it. Two of the products offered here cannot do what they claim without integration. A scribe that generates a clinical note but cannot write it back to the record has moved the work rather than removed it. A receptionist that books, reschedules and cancels appointments must reach a scheduling system to do so, and none is named.

The product material states that the tools fit into existing workflows and work across single or multiple clinics. Nothing published explains how.

Delivered as a hosted web application with registration and login, positioned for single practices and multi location groups, with a demonstration booked through a shared calendar link.

Nothing further is described. No hosting provider, no cloud region, no data residency statement, no retention position, no uptime or availability commitment and no alternative deployment model for an organisation requiring processing inside its own environment. No geographic availability statement exists either, which is a live question in this category since three other vendors publish explicit limits on where their generative products may be used.

The telephone and interpretation products raise a further question nobody has answered: real time voice handling has infrastructure and jurisdictional implications distinct from text, including call recording, and none of it is addressed.

A dedicated pricing page exists and is linked from both the main navigation and the footer, which is more than most vendors in this category offer and is credited on that basis. Its contents were not retrieved this pass, so no figure, tier structure or basis is on this record; pull it on refresh, since this axis could move.

What can be established without it: registration is open and self serve, so an individual clinician can reach the product directly, and the enterprise route runs through a demonstration booked on a shared calendar link.

What is unclear is how a wide product set is priced. Seven distinct product lines are offered, from documentation to telephone reception to interpretation to examination preparation, and nothing published indicates whether they are bundled, sold separately, metered by usage or limited by volume. For products with genuine per unit cost, such as voice minutes and real time interpretation, the pricing basis is the first thing a buyer needs and the least likely to be answered by a single page.

Breadth is claimed at an unusual scale and evidenced nowhere. The stated audience runs across doctors, nurses, medical assistants and front desk staff, in single practices and multi location groups, with a further student and trainee segment served by examination preparation tools. Interpretation is offered across several named languages, which is a genuine access feature and is credited.

No specialty is enumerated, no care setting is characterised beyond the clinic, no geographic scope is stated and no evidence of usefulness is published for any of it.

The breadth is itself worth weighing rather than simply recording. Seven product lines spanning documentation, telephony, interpretation, decision support, search and education represent several distinct engineering and clinical safety problems, each of which the more focused vendors in this category treat as a whole company. A buyer should test the specific product they intend to use rather than assuming that a platform claiming everything performs evenly across it.