How to Write an AI Use Disclosure for Patients: Consent and Notice Templates

Patient AI disclosure components

A patient-facing AI disclosure is most useful when it connects plain-language notice, consent expectations, and operational governance. The disclosure should be understandable to patients while still matching the way the AI workflow actually functions.

Start with the workflow, not with generic AI language

1. Map the specific patient-facing workflow

   A useful patient AI disclosure begins with a precise description of the workflow. Healthcare AI may support ambient documentation, clinical decision support, triage, scheduling, billing, patient messaging, care navigation, or agentic task execution. These uses do not create the same patient experience, data exposure, or governance burden. A short general notice may be sufficient for routine assistive uses that remain under staff review, while a workflow-specific notice may be appropriate when AI is visible to the patient or affects timing, routing, recommendations, or documentation.

2. Use a workflow-specific notice for patient messaging, triage, or documentation

   Workflow-specific notices are better when AI is visible in an interaction or affects routing, documentation, messages, or recommendations.

Notice versus consent: choosing the right disclosure format

General notices are appropriate for routine assistive uses. Workflow-specific notices are better when AI is visible in an interaction or affects routing, documentation, messages, or recommendations. Affirmative consent should be reserved for uses required by law, policy, research rules, contractual commitments, or higher-impact workflows where patient data use or AI autonomy goes beyond routine expectations.

Disclosure template types

The supplied disclosure framework distinguishes between a general patient AI use notice, workflow-specific notice language, and affirmative patient consent for an AI-enabled activity.

Template 1: general patient AI use notice

A general patient AI use notice is appropriate for routine assistive uses when the disclosure can explain where AI may be involved, what data may be used, what the AI does and does not do, and who remains responsible for care decisions.

Template 2: workflow-specific notice for patient messaging, triage, or documentation

A workflow-specific notice is appropriate when AI is visible to the patient or affects timing, routing, recommendations, messages, documentation, or similar workflow outcomes.

Template 3: affirmative patient consent for an AI-enabled activity

Affirmative consent should be reserved for uses required by law, policy, research rules, contractual commitments, or higher-impact workflows where patient data use or AI autonomy goes beyond routine expectations.

Drafting rules for compliance-ready AI disclosure in healthcare

• Use factual verbs: Prefer words such as drafts, summarizes, routes, organizes, suggests, or assists. Avoid vague statements such as AI improves care unless the claim is validated and approved.
• Separate assistive AI from decision support: Clinical decision support language should make clear whether clinicians can independently review the basis for a recommendation and whether the AI output is advisory.
• Address agentic actions explicitly: If an AI agent can call tools, send messages, schedule visits, or trigger workflow actions, disclose the role of human approval and maintain tool-call controls.
• Coordinate with health IT transparency obligations: Where certified health IT predictive decision support is involved, ensure internal review has access to source attributes and risk management summary information.
• Keep urgent-care pathways visible: Patient messaging and triage workflows should include clear instructions for urgent, emergency, medication, diagnosis, or self-harm concerns.
• Review language after material changes: Changes to data sources, autonomy, workflow placement, vendor terms, or patient impact should trigger review before the old notice is reused.