Truace
Subscribe

Regulating the drafting fiction: A qualitative content analysis of official guidance and regulator documents on AI scribe adoption and use in healthcare

Artificial intelligence scribes use automatic speech recognition and generative language models to document clinical consultations, but official expectations for safe adoption are scattered across privacy, professional, medical device, safety and health-system guidance. We conducted a qualitative content analysis of 74 unique English-language official documents issued by regulators, government bodies, medical colleges and health-system organisations. Documents were identified through a staged grey literature sea…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1814· cite & verify
Regulating the drafting fiction: A qualitative content analysis of official guidance and regulator documents on AI scribe adoption and use in healthcare
BOU ELEC 3620 - Policy Analysis and Implementation - Module 06 - Instruments of Government Policy by Dave Barrows. CC BY 4.0 · https://creativecommons.org/licenses/by/4.0

In brief

Researchers reviewed 74 English-language official documents from regulators, government bodies, medical colleges and health-system organisations about AI scribes, which transcribe and draft clinical consultation notes. They coded the documents and described a pattern they call the drafting fiction: AI-generated text stays provisional until a clinician reviews and authenticates it.

The pattern matters because governance is being built around human review, yet the guidance leaves unclear what that review should involve in practice. The study is a document analysis, so it describes what guidance says rather than measuring how well scribes or safeguards perform in clinics.

Main points

  1. The analysis covered 74 unique English-language official documents found through a staged grey literature search.
  2. Transparency about AI scribe use is common across guidance, while consent requirements are uneven.
  3. Privacy governance ranges from secure storage to the identifiability of voices.
  4. Medical device classification shifts when a tool moves from transcription to summarisation.
  5. Recommendations include reasonable verification standards, protected review time, consent refusal pathways and monitoring across accents, language and speech differences.

The problem

Official AI scribe guidance treats clinician verification as the central safeguard but does not specify the encounter-level support clinicians need to catch hallucination, omission, automation bias or model drift.

The rundown

The documents came from privacy, professional, medical device, safety and health-system sources, so expectations for safe adoption are scattered. The authors coded them with a hybrid deductive and inductive framework.

Verification by the clinician is treated as the central safeguard in direct AI-scribe guidance. Organisational safety cases, vendor due diligence and data protection assessments reduce product-level uncertainty but do not define support for individual encounters.

The authors conclude that AI scribe governance is forming a sociotechnical architecture. They call for guidance that allocates protected review time, resources pathways for patients to refuse consent, and requires performance monitoring across accents and speech differences.

Sources

  1. Peer-reviewedPLOS Digital Health2026-10-06

The debate