AI symptom checkers: Intended purpose matters more than the chatbot label
Appointment support and individual medical decision support are different uses. Define and verify the scope before putting an assistant on a website.
Editorial review:
Background & source findings
An opening-hours assistant differs from software assessing individual symptoms for diagnostic or therapeutic decisions. MDCG guidance examines medical intended purpose and function; a non-binding disclaimer does not remove that assessment. [1]
Where software qualifies as a medical device, applicable classification rules consider its purpose and the consequences of decisions based on its output. A website deployment is not itself an exemption. [1]
Health-data processing additionally requires its own purpose, legal basis, necessary scope and recipient assessment. Medical-device status does not resolve GDPR obligations. [2]
Scope & limitations
Not every chatbot is a medical device or has the same risk class. No automatic AI Act high-risk classification is asserted; that requires a separate current assessment of the specific system.
In plain language
Fictional example: A practice approves an appointment assistant, but testing reveals individual symptom assessment. The extra behaviour triggers restriction or reassessment, not an informal expansion of scope.
Practical recommendations
Our recommendation: Align permitted questions, marketing, manufacturer documents and actual outputs. Obtain intended-purpose and conformity evidence for medical functions and seek competent regulatory review before processing patient inputs.
Provide clear emergency and human-contact routes. Minimise data and prevent sensitive inputs reaching advertising or unnecessary analytics services. Test scope boundaries with fictional questions. Warnings complement, but cannot replace, suitable system design.