Rejoy Health

The artificial intelligence is the product. Every line Rejoy sells is a model application: RejoyGPT for clinical questions, notes, letters and patient education; an AI scribe that turns the visit into structured notes; an AI coder that suggests ICD-10 and CPT codes; an AI receptionist that answers calls and books appointments; an AI interpreter for live interpretation in several languages; clinical decision support offering differentials, treatment suggestions and lab recommendations; a medical search engine; and an AI study companion. Remove the model and nothing remains to sell: no authored content library, no workflow system of record, no device.

The clinical answer and search tools form one side of the suite, and the scribe, coder, receptionist and interpreter form the other.

Oversight is asserted largely as a phrase, with one concrete limit. The home page repeats "You stay in charge" and says the AI takes on documentation, calls and clinical lookup and then hands decisions back to the clinician; the pricing page adds the one described control: nothing is signed or billed without the clinician, the user reviews and signs, and Rejoy does not bill payers or write orders on its own.

What is not described is how that control reaches the rest of the surface. The clinical decision support module offers differentials, treatment suggestions, lab recommendations and next step guidance while the patient is in the room, with nothing published on citations shown, confidence, abstention or what happens when the evidence is thin. The AI receptionist answers patient calls, handles questions and books appointments around the clock without a clinician present, and no routing rule, escalation criterion or crisis handoff is published for that channel. A public symptom checker sits among the free resources with no stated audience or limit.

No absolute accuracy claim is made; "more accurate than GPT-5 on MedQA" and "clinical-grade accuracy" are comparative and quality claims.

Proprietary models are asserted rather than explained. The home page says the company is building clinical grade models hospitals can rely on, calls the study companion its most accurate healthcare models, and describes RejoyGPT answers as journal backed and grounded in current medical evidence and the search engine as covering guidelines, journals and drug data. That is the architecture in general terms.

No model class, provider, version, retrieval or grounding method, training or fine tuning description, or release record is published. The MedQA comparison against GPT-5, GPT-4 and GPT-4o names the benchmark and nothing else: no date, split, sample or protocol.

Nothing establishes who else sits between a patient's conversation and an answer. The scribe, receptionist, interpreter and coder process patient information by design, and no model provider, hosting provider, speech or telephony provider, subprocessor list or data flow is named on the home page, pricing page, privacy policy or terms. Google, Microsoft and NVIDIA appear on the home page as supporters of the company, not as named processors of its data.

The published privacy policy refers to third party providers for payments, accounting software connections and advertising, and does not describe the clinical product at all, so it cannot answer the question for recordings, notes or prompts.

Vendor reported figures with no method, denominator or reference standard, and no named customer. The home page shows Rejoy at 99.4 percent on MedQA, the USMLE style question benchmark, against GPT-5 at 97, GPT-4 at 89 and GPT-4o at 88 percent, with no evaluation date, split, sample, method or citation, and the competitor figures carry no source. RejoyGPT is said to save about 12 hours of administrative work each week, with no study or cohort described. The home page says more than 40,000 clinicians at leading US hospitals and health systems trust the product, and names no institution; earlier company material gave 20,000 professionals and, before that, 50,000 physicians, and the figures are not reconciled.

No peer reviewed or independent evaluation of any Rejoy product is published. A clinician registering for a tool is not an institution deploying it, and nothing published distinguishes the two.

General assurances that do not answer the questions AI raises, on a product that processes a great deal of patient information. The scribe turns patient conversations into notes, the receptionist answers patient calls, the interpreter sits inside the consultation, and the coder reads the visit note. Against that, the home page says the product is HIPAA compliant by design and that patient data stays protected with healthcare grade security, encryption and access controls, and the pricing page says paid and free clinical workspaces are HIPAA compliant.

The published privacy policy does not describe this product: it covers business registration, linking accounting software such as QuickBooks, FreshBooks and Xero, business information for funding approval, purchases and cookies, and says nothing about patient conversations, recordings, notes, prompts or model use. No retention period for recordings or notes, no statement on whether clinical inputs train any model, no de identification practice and no call recording policy is published. The privacy policy does allow third party providers to collect browsing data for interest based advertising.

No safety engineering is published: no guardrails, no hallucination mitigation and no process for handling a safety event.

Compliance is claimed without the underlying document, and the agreement sits on one tier only. The home page says Rejoy is HIPAA compliant by design, and the pricing FAQ says paid and free clinical workspaces are HIPAA compliant. A business associate agreement is listed as an Enterprise feature, alongside SSO and an uptime SLA; no agreement, template or terms are published.

The tier matters. The scribe, coder and interpreter, which handle patient information, are sold on the Pro and Practice plans, and nothing published says those plans carry a business associate agreement, so a clinic on them relies on the compliance claim alone. The published privacy policy does not describe the product or mention HIPAA, so no substantive privacy document reaches the service.

Controls are asserted with nothing independent behind them. The home page says patient data is protected with healthcare grade security, encryption and access controls, and names no standard, control framework or auditor. No SOC 2 report, ISO 27001 certificate, HITRUST certification, trust center, penetration testing statement or vulnerability disclosure program appears on the home page, pricing page, footer, privacy policy or terms of service.

The published support route is WhatsApp alongside a general email address, which for a product that handles patient calls and clinical notes means support conversations about it will tend to involve specifics.

Clinical claims with no regulatory position stated anywhere. The home page markets clinical decision support that gives evidence based differentials, treatment suggestions and next step guidance while the patient is in the room, and suggests relevant diagnostic tests; a public symptom checker sits among the free resources. No clearance, submission, device analysis, intended use statement or limitation is published, and the terms of service and privacy policy do not describe these products.

Nothing published shows that a differential or treatment suggestion links to a source a clinician can open; the home page describes answers as journal backed and grounded in current medical evidence without describing a citation mechanism. The pricing page's statement that Rejoy does not write orders on its own is a workflow limit, not a regulatory position. The result is diagnosis and treatment marketed with neither a stated position nor a shown basis.

Nothing is published on how model behavior is governed or tested. The home page, pricing page, privacy policy and terms of service carry no AI principles, responsible use framework, clinical advisory board, named clinical leadership, evaluation method, bias assessment, subgroup analysis or error taxonomy. The one quality statement is the company's own benchmark figure, which carries no method.

Two presentational choices concern how the company represents itself to a clinical buyer and are recorded as displayed. A panel headed "Backed by the teams that build AI" says the company is supported by Google, Microsoft, NVIDIA and Y Combinator as it builds clinical grade models; the nature of each relationship (investment, partnership or a startup program) is not stated. And the home page says the product is trusted by more than 40,000 clinicians at leading US hospitals and health systems, with no institution named as a customer; earlier company material gave 20,000 professionals and, before that, 50,000 physicians.

One mechanism lets a clinician catch an error before it takes effect, and nothing stands behind the output. The pricing page says every plan keeps the clinician in charge, that nothing is signed or billed without the clinician, that the user reviews and signs, and that Rejoy does not bill payers or write orders on its own; the home page describes RejoyGPT answers as journal backed and grounded in current medical evidence. Review before signing is a real control on notes and codes.

No accuracy or error rate with a method is published, and no warranty, indemnity or correction commitment attaches to an answer, a note or a code. The published terms of service (last updated 19 February 2021) address orders, refunds and cancellation for goods and subscriptions and contain nothing about AI output. The home page's MedQA comparison (99.4 percent against GPT-5 at 97, GPT-4 at 89 and GPT-4o at 88) carries no method, date or split, so it is not an error rate.

Integration is claimed with no system named and nothing to verify. The pricing page lists EHR integration as an Enterprise feature and not on the Free, Pro or Practice plans. No record system, scheduling system, marketplace listing, API or health data exchange standard is named anywhere.

Two products depend on integration to do what they claim. A scribe note that cannot be written back to the record moves the work rather than removing it, and a receptionist that books, reschedules and cancels appointments must reach a scheduling system; on the plans a single clinic would buy, neither connection is described.

A single hosted option with location implied rather than committed. Rejoy is a hosted web product with one login for a single clinic or a network of locations, a San Francisco office address, and an AI receptionist that runs a clinic phone line; Enterprise adds multiple sites, SSO and an uptime SLA.

No hosting provider, cloud region, data residency statement, tenant isolation model or on premises option is published, and the privacy policy does not describe the product. Voice handling and live interpretation raise questions text does not, including call recording and where audio is processed, and none is answered.

Published tiers with figures, a stated unit of charge and a route to start without a sales conversation. The pricing page lists four plans, monthly or annual with 20 percent off annual. Free, for students and light clinical use: 25 RejoyGPT questions a day, medical search, the study companion and question bank, no card required. Pro, for individual clinicians: $63 per clinician per month billed annually, with unlimited RejoyGPT, AI scribe, AI coder, clinical decision support, AI interpreter, search and study. Practice, for clinics up to five clinicians: $199 per clinic per month billed annually, adding one AI receptionist line and a shared notes and coding queue. Enterprise: custom, with unlimited seats and locations, EHR integration, multiple receptionist lines, SSO, a business associate agreement and an uptime SLA. The FAQ states there is no per visit or per note billing.

The business associate agreement appears only in the Enterprise column, while the scribe, coder and interpreter that handle patient information are sold on Pro and Practice. The published terms of service (last updated 19 February 2021) also describe order cancellation and refunds for goods rather than these subscriptions.

Coverage is claimed broadly without specifics. The stated audience runs across doctors, nurses, medical assistants and front desk staff, from solo practice to health system, with students served by the study companion and exam question banks, including a NEET PG bank for Indian postgraduate entrance exams. Live interpretation is offered in English, Spanish, French, Chinese, Korean and more, which is a real access feature.

No specialty or care setting is enumerated beyond the clinic, no geographic scope is stated, and no evidence of usefulness is published for any setting. Seven product lines spanning documentation, coding, telephony, interpretation, decision support, search and education are each a distinct clinical safety problem, and nothing published shows performance on any of them.