Spaces

Do and don't guide for a hospital AI agent

This guide lists plain do and don't rules for an AI agent answering hospital messages. It keeps the agent useful on practical questions and silent on anything clinical or private.

Runs in your browser. Nothing you enter is sent anywhere or saved.

Before you start

A good hospital agent is like a helpful volunteer at the front door. It knows the building and the rules, it is patient with worried people, and it knows when to fetch a nurse. A bad one guesses, soothes with false comfort or shares what it should not. These rules draw the line between the two.

Share this guide with your administrators, your privacy officer and a senior clinician, and edit it to fit your site. Then turn each line into a written instruction for the agent. Read it aloud at a team meeting, so everyone who takes handovers knows what the agent will and will not do.

Tick each one as you go. 0 of 10 done.

[ ] Do answer visiting hours, entrances, parking and clinic locations from your written notes.
[ ] Do give the emergency number or route when a message hints at danger.
[ ] Do hand every clinical, urgent or sensitive chat to a person.
[ ] Do speak kindly and in short sentences.
[ ] Do say plainly when you do not know and a person will confirm.
[ ] Don't confirm or deny that a named person is a patient.
[ ] Don't give medical advice, reassurance or opinions on symptoms.
[ ] Don't promise waiting times, bed places or appointment slots you cannot see.
[ ] Don't ask for records, ID numbers or full dates of birth in chat.
[ ] Don't argue with an upset person, and don't copy a complaint into a public reply.

How the rules look in practice

A relative writes: how is my mother, Ana Reyes? The agent replies that patient information comes only from staff, shares the ward desk contact and hands the chat over. It does not say yes, no or maybe. That one habit protects both the patient and your hospital.

A visitor writes: where do I park for the cancer clinic? The agent answers fully, with the car park name, the entrance and the accessible route. This is where the agent shines, so do not restrict it more than the limits above require.

Keep the guide alive

Revisit these rules every few months and after any incident. Add a new do or don't whenever a real chat shows a gap, and retire any line that no longer matches your policy.

FAQ

Can the agent comfort someone who is worried about a relative?

It can be kind and brief, but it cannot reassure about a person's condition. The best reply is warm, explains who can help and passes the chat to staff straight away.

Why are these rules stricter than for a shop or a clinic?

A hospital holds the most sensitive information and the highest risks. A wrong answer about a patient or an emergency can do serious, lasting harm, so the limits are firm.

Who should have the final say on these rules?

Your hospital's own leaders, with advice from a clinician and your privacy officer. The agent follows what you write, so the responsibility for the rules stays with your team.

More free tools for hospitals and hospital admin teams

Read the guide: An AI agent for hospital enquiries and administration.