Administrative/access AI
Avie
Avie’s clinic-paid WhatsApp receptionist may reduce the burden of arranging care. Clinic control and patient-facing convenience do not establish patient-directed AI. Refusal, human-request and correction controls remain unverified. Mixed and unscored.
Hugo Campos authorized publication of the completed review. No comprehensive human verification or hands-on product testing is claimed.
Summary judgment · Unscored in current comparisons
Clinic-configured Atende reception and access workflow. No adopted numerical calculation or calibrated comparative position.
Mixed
Documented booking, rescheduling and cancellation may expand practical task agency. The clinic governs the workflow. Patient refusal, human-request and correction controls remain incompletely established, not proven absent.
Patient agency
How this tool changes agency
Documented scheduling preferences, booking changes and cancellation could help patients pursue chosen administrative goals outside reception hours.
Patient-initiated human access, equivalent non-AI access, correction of inferred context and downstream portability are not established. These are evidence gaps, not findings of absent controls or harm.
Patient agency assessment
Who sets and changes the goal?
The clinic purchases and configures facts, tone, routing and rules. Patients can request appointment alternatives within that workflow.
Can patients tell AI is involved?
Vendor pages identify AI, but a mandatory patient-facing notice across clinics is unverified. Human-style reply pacing creates a documented transparency concern, not proof of deception.
Can patients meaningfully choose?
Booking alternatives, cancellation and some outreach stop requests are documented. Universal AI refusal with equivalent access and patient-triggered human transfer are not established.
Can patients correct or challenge what the AI produces?
Staff can edit or stop replies, and privacy rights include correction. Direct patient inspection of inferred context and reliable correction propagation were not verified.
Does it help patients understand or act?
Documented booking, rescheduling and cancellation could support the patient’s chosen task. Vendor reply-time observations do not demonstrate patient empowerment or correct resolution.
Can patients evaluate the conditions of use?
Public policies name data roles and providers, but patient-visible action receipts, configuration-specific notices and practical access to derived context need verification.
Text findings
What happens to patient data?
Published rights and vendor controls, implementation unverified
Clinic/controller and Avie/processor roles, OpenAI/Google providers, stored conversation context and rights requests are described. Content-specific retention, training restrictions and patient control of inferences remain unresolved. Notes audio limits do not govern every data class. See the linked report for sources and version differences.
Who is left out or burdened?
Deployment evidence needed
Brazilian Portuguese and WhatsApp may reduce some barriers. Screen-reader use, low literacy, shared devices, caregiver permissions, connectivity and equivalent non-WhatsApp access were not tested. Clinic subscription costs do not establish a separate patient subscription.
Are the clinical boundaries clear?
Documented, behavior untested
Reception help excludes diagnosis, medication guidance and symptom-severity assessment, with staff routing for sensitive topics. Routing itself can affect access. After-hours acknowledgment is not a guarantee of immediate human or emergency care. Notes is a separate beta documentation workflow.
Who defined what good looks like?
Vendor-run operational evidence
Índice Avie v3 reports small nonrandomized clinic observations of first replies from AI or staff. It does not establish causal clinical benefit, patient-defined success, equitable access or independent validation.
Review provenance
Criteria
CAIHL-derived HugoScore framework and current qualitative review method. Draft numerical rubric anchors are not adopted.
Reviewer
AI-assisted public-source research, adapted for public HugoScore publication
AI / model
OpenAI GPT-6 Astra (gpt-6-astra)
Human review
Hugo Campos authorized publication of the completed review. Publication authorization is not comprehensive human verification.
Review date
2026-10-18
Limitations
Public-source documentary review. Some sources were recovered through search-index extraction. No account, conversation, booking, health-data upload, clinical evaluation, security audit or vendor interview. Vendor observations are not independent validation. Configuration-specific behavior and patient controls remain unverified.
Review method
CAIHL public-source documentary review. Published claims, vendor observations, inferences and unknowns distinguished. No interaction testing.
AI-assisted draft · Medium for public documentation, lower for implementation and patient outcomes (AI-assisted draft)