nVoq Voice CAIHL draft report

HugoScore CAIHL Draft Report: nVoq Voice

• Status: Draft for human review
• Last reviewed: 2026-07-12
• Review method: Public-source review of nVoq product pages, current support documentation, privacy and security materials, HIPAA documentation, customer quality-improvement studies, and a 2026 partner announcement. No deployed-system access, ambient recording test, clinician or patient interview, consent-form review, customer contract or BAA review, SOC 2 or penetration-test inspection, model audit, accessibility test, independent validation, or legal/regulatory determination.
• Service: nVoq Voice
• Vendor: nVoq Incorporated
• URL: https://www.nvoq.com/
• Category: Ambient scribe

1. Executive Summary

nVoq Voice is an enterprise clinical-documentation platform aimed especially at home health, hospice, palliative care, and other post-acute settings. It converts clinician dictation or clinician-patient conversations into transcripts, summaries, formatted clinical notes, and documentation-completeness checks. nVoq says structured OASIS and other form completion is coming in its Performance+ subscription. The product can place clinician-reviewed content into existing EHR workflows.

Under CAIHL, nVoq is institutional and clinician-mediated. Healthcare organizations and technology partners choose, configure, and pay for the platform. The immediate measures of success are less documentation time, clinician satisfaction, note completeness, audit readiness, and revenue protection. Patients may benefit indirectly when clinicians pay more attention during visits and when the record better represents their circumstances, but nVoq is not a patient-facing understanding or advocacy tool.

nVoq's ambient workflow contains meaningful safeguards. Current Windows and iOS instructions require the clinician to confirm permission from everyone before recording. The vendor describes dictation as the fallback when a patient or caregiver refuses ambient recording. Clinicians can review the transcript and audio, edit generated text, and are told they remain responsible for accuracy before EHR entry.

Important agency limits remain. Public documentation does not show that patients receive the transcript, summary, recording, source traceability, an AI-use label in the chart, or a direct correction and deletion channel. A broad “zero retention” statement in the product FAQ also needs qualification: the AI-security document applies zero retention to third-party generative-AI processing, while current Voice Assistant documentation says recordings, transcripts, and summaries remain in nVoq cloud storage under customer-configured retention settings.

• Agency posture: Mixed, institution-directed
• Agency axis position: 28 of 100
• Confidence: Medium draft, strong official documentation and vendor-customer studies, limited independent evidence

2. CAIHL Question

Who does nVoq Voice serve?

nVoq primarily serves healthcare organizations, clinical teams, administrators, and EHR technology partners seeking faster, more complete, reimbursement-ready documentation. It serves patients indirectly through the clinical encounter and medical record. Patients are data subjects and participants in ambient capture, not product administrators or the principal owners of the workflow.

CAIHL classification: Institutional, clinician-mediated clinical documentation AI.

3. What The Service Does

nVoq Voice has three advertised subscription levels. Premium generates structured clinical notes from clinician dictation. Performance adds ambient conversational capture and note summaries. Performance+ is described as coming soon and is intended to add structured form completion, including OASIS auto-population.

The platform records voice, transcribes or summarizes it, applies clinician- and customer-configured structure, checks documentation completeness, and returns content to an EHR or other system of record. Current documentation supports transcript and audio review, summary editing, feedback, and deletion. The vendor's current iOS Voice Assistant is described as limited early access, while Windows Conversation Capture is documented as available when enabled by an organization.

4. Patient-Impact Pathway

1. A healthcare organization or technology partner procures and configures nVoq Voice. 2. A clinician uses dictation or selects ambient Conversation Capture / Voice Assistant. 3. For ambient recording, the clinician must confirm permission from all participants; nVoq describes dictation as an alternative when permission is withheld. 4. Audio is captured locally and synced to nVoq cloud infrastructure for transcription and summarization. 5. Proprietary models and, for some generative features, secure third-party services such as OpenAI process spoken clinical information. 6. The clinician reviews audio, transcript, summary, or formatted note; edits are possible in documented workflows. 7. The clinician copies or inserts the final text into the EHR. Planned features may populate structured OASIS fields. 8. The resulting record affects care-team communication, hospice eligibility narratives, audits, reimbursement, and potentially subsequent care decisions. 9. Patient agency depends on notice, refusal without care penalty, accurate representation, access to the resulting record, correction rights, retention controls, and equitable speech performance.

5. Evidence Table

| Source | Evidence | CAIHL relevance |
| --- | --- | --- |
| nVoq homepage, accessed 2026-07-12: https://www.nvoq.com/ | Describes voice-first documentation that captures clinical speech and delivers structured content into healthcare systems. | Establishes product purpose and institutional workflow. |
| nVoq FAQ, accessed 2026-07-12: https://www.nvoq.com/nvoq-faqs/ | Defines Premium dictation, Performance ambient summaries, and planned Performance+ structured form completion; describes clinician oversight and a broad zero-retention claim. | Product scope, clinical control, roadmap, and retention claim. |
| Windows Conversation Capture guide, accessed 2026-07-12: https://support.nvoq.com/docs/voice-help-windows-conversation-capture | Requires confirmation of permission from all participants; supports transcript and summary review, copying, feedback, and customer-configured retention. | Notice, refusal, correction, EHR entry, and retention evidence. |
| iOS Voice Assistant guide, accessed 2026-07-12: https://support.nvoq.com/docs/voice-ios-help-voice-assistant-conversation-capture | Describes permission confirmation, background recording, audio/transcript/summary review, summary editing, deletion, local and cloud storage, and cloud availability after local cleanup. | Core consent, visibility, correction, deletion, and storage evidence; feature is limited early access. |
| Ambient workflow article, accessed 2026-07-12: https://www.nvoq.com/a-day-in-the-life-with-ambient-ai-documentation-in-post-acute-care/ | Vendor scenario says the clinician explains ambient AI, obtains informed consent, reviews and edits output, and switches to dictation when consent is refused. | Choice and refusal evidence; illustrative vendor scenario, not deployment audit. |
| AI data security and privacy explainer, dated 2025-09-15: https://www.nvoq.com/wp-content/uploads/2025/09/nVoqDataSecurity.pdf | Says no PHI trains nVoq models; some features use services such as OpenAI through zero-retention APIs; identifies personal and clinical data processed and regional data residency. | Model training, third-party processing, retention, and residency evidence. |
| Security and compliance page, accessed 2026-07-12: https://www.nvoq.com/secure-ai-documentation/ | Claims HIPAA-ready safeguards, SOC 2 Type II, U.S. hosting, encryption, role-based access, and fairness monitoring. | Security and governance claims; underlying reports and subgroup results were not public. |
| nVoq privacy policy, updated 2026-03-12: https://www.nvoq.com/privacy-policy/ | Covers recordings and transcripts, R&D use of personal information and de-identified data, service-provider sharing, and purpose-based retention. | General service-level privacy disclosure; customer PHI may also be governed by contracts and BAAs not reviewed here. |
| nVoq and HIPAA, dated 2023-02-27: https://www.nvoq.com/wp-content/uploads/2023/03/nVoq_HIPAA_Compliance_2023.pdf | Describes tenant databases, AWS regions, encryption, SOC 2 audits, penetration testing, and support-personnel handling of PHI. | Security architecture and human-access evidence; audit artifacts require NDA. |
| Amedisys quality-improvement case study, published by nVoq in 2025: https://www.nvoq.com/wp-content/uploads/2025/02/CaseStudy_Amedisys29JAN25_v2-copy.pdf | Reports 7,321 narratives, greater inclusion of hospice documentation elements, 42% longer narratives, error-rate reduction from 15% to 3%, and 50% less documentation time. | Operational evidence; vendor-customer produced and not an independent patient-outcome evaluation. |
| MatrixCare announcement, 2026-03-24: https://www.nvoq.com/press-release/nvoq-expands-ai-powered-clinical-documentation-capabilities-to-matrixcare-users/ | Announces availability within a post-acute EHR ecosystem and describes ambient summaries and clinician oversight. | Deployment and integration context; company announcement, not independent validation. |

6. Mixed HugoScore Profile

Who does this AI serve?

Institutional and clinician-mediated. Buyers and administrators seek faster, more compliant documentation and revenue protection. Clinicians gain reduced charting burden. Patients may gain more attentive encounters and fuller records but do not control the platform.

Can patients tell AI is involved?

Partial. Ambient recording requires the clinician to confirm permission, and nVoq's scenario describes an explanation and informed consent. Ordinary dictation, formatting, completeness checks, and the resulting EHR note may not visibly disclose AI involvement.

Can patients meaningfully choose?

Partial for ambient recording. Recording cannot begin until the clinician confirms permission from all participants, and dictation is described as the alternative when consent is refused. The public evidence does not establish how refusal is documented, whether consent can be withdrawn mid-visit across deployments, or whether refusal ever burdens access to care. Patients cannot choose the vendor or control non-ambient documentation functions.

Can patients correct or challenge what the AI produces?

Partial and clinician-mediated. Clinicians can listen, review, edit, compare, delete, and give feedback. No nVoq-specific patient interface or correction route is documented. Patients must depend on the clinician and the care organization's ordinary medical-record access and amendment procedures.

Does it help patients understand or act?

Not designed for direct patient action. The platform creates clinician documentation rather than patient explanations, questions, choices, appeals, or navigation support. Indirect benefits are plausible but patient-agency outcomes have not been evaluated publicly.

Who is left out or burdened?

Public support documentation lists dictation topics as U.S. English. Ambient performance is not publicly stratified by accent, language, speech disability, cognitive impairment, quiet speech, background noise, interpreter use, or overlapping speakers. Windows guidance recommends no more than four speakers and avoiding overlap. nVoq claims fairness monitoring but publishes no subgroup methods or results. No patient-facing accessibility or low-literacy consent materials were identified.

What happens to patient data?

nVoq says audio and clinical content can be processed in tenant-isolated, encrypted U.S. or Canadian environments. It says no PHI trains nVoq models and that third-party generative-AI providers such as OpenAI operate under zero-data-retention and no-training terms.

The product itself is not uniformly zero retention. Current iOS documentation says recordings sync to the cloud and that audio, transcripts, and summaries remain available after local copies are removed. Windows documentation says original input and summaries are retained according to group settings. Exact default and maximum periods, backups, deletion propagation, administrator access, feedback retention, and patient request handling are not publicly specified. The general privacy policy allows R&D use of personal information and derived de-identified data, while the AI explainer says PHI is not used for model training; contracts and data-category boundaries need human review.

Are the clinical boundaries clear?

Partial. nVoq consistently requires clinician review and states that the clinician is responsible for accuracy. But the generated documentation can affect care-team understanding, hospice eligibility, audits, reimbursement, and planned structured assessments. Public evidence does not establish independent ambient-summary accuracy, hallucination rates, safety escalation, or how omissions are detected across patient subgroups.

Who defined what good looks like?

Mostly the vendor, clinicians, customer organizations, and reimbursement rules. The Amedisys work measures required hospice content, error rates, documentation time, clinician experience, and revenue-related completeness. No independent patient-experience, patient-agency, clinical-outcome, subgroup-fairness, or ambient-summary validation was identified.

7. Key Unknowns

• Deployment-specific patient notice, consent wording, withdrawal, and refusal-without-penalty practices.
• Whether the EHR note discloses AI assistance and whether patients receive recordings, transcripts, or summaries.
• Default, minimum, and maximum cloud retention by feature; backup deletion and administrator deletion controls.
• Whether deleting from the clinician app deletes every cloud, administrator, feedback, and backup copy.
• How patient record-access and amendment requests reach nVoq-held source material.
• Exact data sent to OpenAI or other providers for each feature and whether all providers receive zero-data-retention treatment.
• How the privacy policy's R&D language applies to customer recordings, transcripts, feedback, and PHI.
• Ambient-summary accuracy, omission, hallucination, speaker attribution, and structured-field extraction performance.
• Performance by language, accent, race, age, sex/gender, disability, care setting, microphone, and background noise.
• Accessibility, interpreter, caregiver/proxy, pediatric, and capacity/consent workflows.
• Independent patient-safety, experience, clinical-outcome, and agency evaluation.
• Current commercial availability and validation of Performance+ and structured OASIS/HOPE completion.

8. Patient Agency Interpretation

nVoq's strongest agency feature is the ambient refusal pathway: recording requires confirmed permission, and clinicians can fall back to dictation. Its second strength is clinician reviewability—audio and transcripts can make errors visible to the person signing the note.

That is not the same as patient authorship. The institution chooses the system, the clinician edits the representation, and the final note enters a record that can shape eligibility, reimbursement, care coordination, and future decisions. A patient-centered deployment would make AI use visible, offer a genuine refusal without care penalty, provide access to the resulting note, make amendment routes easy, minimize and time-limit source recordings, and monitor error rates across speech and language groups.

9. Publication Recommendation

Ready for human review as an AI-assisted, source-backed draft under Ambient scribe. Keep the CAIHL classification “Institutional, clinician-mediated clinical documentation AI,” the agency posture “Mixed, institution-directed,” and confidence at medium draft. Do not mark reviewed or verified. Human follow-up should prioritize the zero-retention/product-storage distinction, patient consent and refusal in live deployments, cloud deletion, patient correction rights, subgroup accuracy, and independent ambient-summary validation.

Review Provenance

• Criteria: HugoScore patient agency framework derived from CAIHL, using the same public questions and mixed answer types applied to every tool.
• Reviewer: AI-assisted public-source draft prepared in OpenAI Codex; no named human reviewer is recorded.
• AI / model: OpenAI Codex / GPT-5.
• Human review: No comprehensive human review has been completed or claimed.
• Review date: 2026-07-12.
• Limitations: No deployment access, ambient recording test, vendor or user interview, patient consent-form review, customer contract or BAA review, SOC 2 or penetration-test inspection, model or bias audit, accessibility testing, independent validation, or legal/regulatory determination.