chathq.ai logo
chathq.ai

Staying on topic

A chatbot that answers anything will eventually answer something wrong, in your name. A support assistant for a bank should not be writing code or giving medical advice, however nicely it is asked.

ChatHQ assistants stay inside the scope you give them.

What the assistant declines

  • Unrelated requests — Homework, programming questions, recipes, opinions on the news. The assistant replies briefly that it can only help with topics related to your business and points back to what it can do.
  • Topics outside your content — Questions about things your knowledge base does not cover are answered with your fallback message rather than with general knowledge. See Evidence verifier.
  • Requests to change its role — Asking the assistant to "act as" something else does not work. See Resisting manipulation.

How scope is defined

Scope comes from two things you control:

  • Knowledge base — The documents, pages and FAQs you add define what the assistant knows and therefore what it will talk about. See Knowledge base.
  • Instructions — The role and boundaries you write tell the assistant how to handle everything else. See Instructions & behavior.

A short boundary line in your instructions is usually enough:

You are the support assistant for Acme Travel. Only answer questions about
Acme Travel's bookings, policies and services. For anything else, say politely
that you can only help with Acme Travel topics.

Test it

Use the test chat on the assistant's page and ask a few off-topic questions: "write me a poem", "what is the capital of Peru", "help me debug this code". The assistant should decline each one and steer back to your topics.

Related

Was this article helpful?

Build AI assistants you can actually rely on

Turn your company's knowledge into assistants your team and customers can rely on.