Spaces

Do and don't guide for your agent: programme offices

This guide gives plain do and don't rules for an AI agent working in a programme office. It keeps answers helpful and keeps eligibility, health and appeals with caseworkers.

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

Before you start

A good agent for a programme office behaves like a patient front-desk clerk who knows the published guides well and knows exactly where their job stops. It explains steps and points to the right form. It does not decide who qualifies, read a result or promise an outcome.

Use these rules as the base for your agent's instructions. Edit them to match your office, add examples from your own chats, and share the final version with every caseworker, so they know what the agent has been told to do.

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

[ ] Do answer from your published guides and say where the answer comes from.
[ ] Do ask one clear question to find the right programme.
[ ] Do use plain words and short sentences.
[ ] Do tell people when a caseworker will take over.
[ ] Do be warm when someone shares bad news.
[ ] Don't decide whether someone is eligible for a benefit or service.
[ ] Don't quote a payment amount or date for a specific person.
[ ] Don't give medical advice or comment on a test result.
[ ] Don't promise that an application or appeal will succeed.
[ ] Don't ask for case numbers, ID copies or health details unless a person asks.

Show the rules with examples

Rules stick better with a sample. Add a few before-and-after replies to your agent's setup. Show a good answer to an enrolment question, and a good hand-over for an eligibility question, so the agent can copy the shape.

Include one example of a calm reply to an angry message. Programme offices get frustrated writers, and the first sentence matters a great deal.

Review the rules each term

Programmes change, so the rules should too.

  • Update the do list when a guide or form changes.
  • Add to the don't list whenever a chat shows a new risk.
  • Ask caseworkers which rule they would add.

FAQ

Can the agent say a person probably qualifies?

No. Even a soft yes can be heard as a promise. The agent can explain the published rules in general and then pass the personal question to a caseworker.

Should the agent tell people it is an AI?

Yes, when asked, and many offices say so up front. Honesty builds trust and sets clear expectations about what the agent can do and when a person will step in.

How long should the rule list be?

Short enough that staff can read it in a few minutes. Ten to fifteen clear rules work better than a long manual that nobody opens.

More free tools for public programme and benefits offices

Read the guide: Answer benefit, clinic and enrolment questions for public programme offices.