The term "AI scribe" is no longer specific enough. In 2024, it meant one thing: an ambient tool that listens to a clinical consultation and generates documentation for the clinician. In 2026, the category has split into at least two distinct product types with different users, different outputs, different governance requirements, and different safety profiles. Treating them as the same category creates confusion about what each tool does and does not do.
Definition: Patient-Facing AI Scribe
A patient-facing AI scribe is a tool initiated by the patient — not by the healthcare organisation — that records a clinical consultation, processes the audio through transcription and AI summarisation, and produces a patient-readable output. Typically this includes a plain-English summary of what was discussed, extracted action items and next steps, sharing options for family members or caregivers, and preparation tools for future appointments.
The patient controls the recording, owns the output, and decides who to share it with. The output does not enter the clinical record. The clinician may or may not be aware that the recording is happening — depending on local custom, the patient's communication, and whether the recording is overt or covert.
How It Differs from Clinician Ambient Scribing
| Feature | Clinician scribe | Patient-facing scribe |
|---|---|---|
| Initiated by | Healthcare organisation/clinician | Patient independently |
| Output destination | Electronic health record | Patient's personal app |
| Primary user | Clinician | Patient and caregivers |
| Clinical review | Clinician reviews before saving | No clinical review of summary |
| Regulatory framework | Medical device, DTAC, clinical safety | Consumer app, personal data privacy |
| Governance | Organisational (DPIA, IG, safety case) | Patient-initiated (personal use) |
| Error consequence | Wrong entry in permanent record | Patient acts on inaccurate summary |
| Examples | Heidi, Accurx Scribe, Tortus, Dragon | Kin Health, Aide Health Mirror |
The governance distinction is critical. When an NHS practice deploys Heidi or Accurx Scribe, the organisation is the data controller, must complete a DPIA, must ensure DCB 0129/0160 compliance, and must inform patients. When a patient independently uses Kin on their personal phone, the legal framework is different — patient-initiated recording for personal use, potentially outside UK GDPR scope under the personal/household activity exemption.
Kin Health (US)
Kin Health (Los Angeles, $9M seed, 18 May 2026) is a free patient app that records medical visits, generates plain-language summaries with action items, supports sharing with a "care circle" of family and caregivers, and helps patients prepare questions for future visits. Founded by physician brothers Arpan and Amit Parikh alongside Kyle Alwyn (co-founder of HeyDoctor, acquired by GoodRx). GoodRx co-founders Doug Hirsch and Trevor Bezdek are founding partners. Revenue model: free for patients, monetised via downstream referrals, labs, and prescriptions — the GoodRx thesis applied to consultation intent. Not HIPAA-certified (patient-facing), claims comparable privacy standards.
Aide Health Mirror (UK)
Aide Health Mirror (UK, launched October 2025) similarly produces plain-English consultation summaries, lets patients ask questions grounded in the consultation transcript, and supports sharing with family and caregivers. Aide Health states its first product is used by thousands across NHS England for chronic disease self-management. Mirror was designed for GP surgeries, hospitals, pharmacies, and care settings. Inspired partly by the founder's experience supporting his father with early-stage Alzheimer's during hospital care. Recordings are automatically deleted to maintain privacy. Users are reminded that AI can make mistakes and are encouraged to confirm uncertainty with healthcare professionals.
Benefits and Risks
Benefits: better recall (addressing the 49% forgetting rate), improved adherence to action items, caregiver involvement in care, health literacy through plain-language translation, and longitudinal care tracking.
Risks: AI hallucination (summaries including recommendations never made), over-simplification (clinical nuance lost in plain-language translation), misplaced trust (patients treating the AI summary as a complete and authoritative record), consent uncertainty (is the clinician aware of recording?), accent and audio issues (transcription errors propagating through the summary), and privacy (audio data sensitivity even with encryption).
What Clinicians Should Know
Patient-facing AI scribes do not require clinician setup, EHR integration, or organisational governance approval. They are patient-initiated tools that exist outside the clinician's control. The practical response: communicate clearly, structure safety-netting explicitly, and accept that consultation content may be captured and summarised by tools the clinician does not control — which is additional reason to ensure verbal communication is as clear and accurate as possible.
iatroX supports clinicians with the clinical knowledge behind their recommendations — so the verbal advice captured by any patient recording is guideline-grounded and evidence-based.
Use iatroX for clinical knowledge, not just consultation memory →