A conversation with Ro’s AI support agent does not automatically become a message to a clinician or an entry in the medical record. Ro’s published policy describes a separate step: asking for a support request, reviewing its summary and sending it to a person. That distinction matters even when a chat feels like part of the care service. Ro’s AI agent policy
The document concerns a support tool for signed-in account holders. It is not evidence that every message has been reviewed by the care team.
A reply and a handoff are different events
Ro says the agent can answer educational and service questions and help draft messages. Its responses are generated without a person reviewing them before the user sees them. The policy also says the chat is not monitored in real time.
The supported handoff is therefore narrower than “I told Ro.” Information in a conversation, a request prepared from it and a request actually sent to a person are different states.
That is an information-flow distinction, not evidence that a particular message has been lost. The policy does not establish that a clinician has read a conversation or promise a particular response time.
The summary is a point of control
The policy describes a summary that the user can review and edit before sending. In this design, the summary is the bridge from an AI conversation to a human support request.
A concise summary can be easier to review than a long conversation, but concision also makes its contents important. The policy does not establish that every detail in the original exchange automatically reaches a clinician elsewhere in the system.
Nor does the ability to ask for a person make the AI itself a licensed provider. Ro explicitly distinguishes its support agent from diagnosis, prescribing and professional medical advice.
Read the actual communication boundary
This is a different task from authorizing an external AI assistant to retrieve a health record. Here, the question is whether information travels from a support conversation toward the care team.
The public policy supports that explanation; it does not provide independent evidence of clinical accuracy, the completeness of any patient record or a service-level guarantee. The useful reading rule is concrete: an AI response and a submitted human-support request should not be treated as the same event.