Qualitative Analysis: AI-Drafted Patient Messages

A July 16, 2026 qualitative study of 40 patients found that patients will accept AI-drafted portal messages only if clinicians review them and patients are informed that AI assisted. The evidence points teams toward piloting AI for low-stakes, high-volume messages first, adding EHR safeguards and testing disclosure language. Use purposive interviews plus cross-segment frequency analysis to build defensible policies and operational workflows.

• A qualitative sample of 40 patients (reported July 16, 2026) showed acceptance of AI drafts conditioned on clinician review and disclosure.
• Patients view portal messages as transactional for refills and scheduling, while preferring clinician-written or in-person communication for uncertain or sensitive results.
• One cited study reported a 7% response-time reduction (about 20 seconds) when AI assisted, highlighting operational gains for routine tasks.
• Deployment needs: clinician-review workflows, disclosure UX testing, KPIs, and patient advisory involvement to reduce risk.

A concise summary: a qualitative study of 40 patients at an academic medical center (reported July 16, 2026) found patients accept AI-drafted portal messages provided clinicians review drafts and patients are told AI assisted.

TechTarget covered the JAMA Network Open study.

• Patients view portal messages as transactional (refills, scheduling, quick answers) rather than relationship-building.
• Nearly all participants insisted on clinician review before messages are sent.
• Patients prefer clinician-written or in-person communication for higher-stakes or uncertain results.
• The study informed practical deployment tips: KPIs, clinician-review workflows, disclosure guidelines, and patient advisory involvement.

Researchers conducted semi-structured interviews with 40 patients to explore expectations around AI-drafted portal messages.

The interviews revealed that patients saw potential benefits such as faster replies and reduced clinician load, but almost uniformly required human review and explicit disclosure that AI generated a draft.

• Tone matters: clerical tone is acceptable for routine tasks (refills, scheduling), while an empathetic, careful tone is expected for sensitive topics.
• Workflow safeguards are essential: policies and EHR constraints should prevent unreviewed sends and define responsibility pathways for content.
• Patients did not agree on exact disclosure wording or timing, indicating disclosure UX needs testing across populations.

This pilot checklist lists seven practical steps to test AI-drafted portal messaging while capturing robust qualitative evidence.

• 1) Define scope & KPIs: response time, edit time per message, patient trust scores, and safety incidents.
• 2) Recruit & segment participants: aim for diverse portal users; include high-anxiety cases separately (n≥40 per pilot recommended for initial qualitative depth).
• 3) Collect artifacts: export portal messages, AI drafts, clinician edits, and post-message interviews or surveys.
• 4) Ingest data into Evidano: transcripts and message text → auto-code then review; use custom dictionaries for clinical terms.
• 5) Run theme and frequency analysis: identify where patients accept drafts versus demand clinician messaging.
• 6) Prototype disclosure variants and simulate EHR constraints; gather user reaction data.
• 7) Report: deliver a concise decision memo with quantified themes, representative quotes, and recommended EHR rules.

The July 16, 2026 qualitative evidence shows patients expect clinician review, tailored tone, and transparent disclosure for AI-drafted portal messages.

AI can speed responses for routine portal tasks, but the July 16, 2026 qualitative evidence shows patients expect clinician review, tailored tone, and transparent disclosure. For teams running qualitative analysis of AI-drafted patient portal messages, combine purposive interviews with cross-segment frequency analysis to build defensible policies.

• Ready to pilot? Ingest transcripts, run thematic and cross-segment analyses, and produce stakeholder-ready visuals with Evidano: Try Evidano for free.
• Primary source: TechTarget (coverage of JAMA Network Open study).