Write a system prompt and tone for Vex

Updated May 22, 2026

What custom instructions are

Custom instructions (also called a system prompt) are a set of rules and context that Vex reads before every conversation. They shape how Vex introduces itself, how formal or casual it sounds, what topics it will and won't engage with, and how it handles edge cases.

You set them on the AI Identity tab of your Vex chatbot under Custom instructions.

What to write in custom instructions

Think of it as a briefing you'd give a new support hire on day one. Cover:

  • Role: What Vex is and who it's for ("You are Vex, a support agent for Acme Corp, helping customers with billing and product questions.")
  • Tone: Formal, friendly, concise, empathetic — pick one or two descriptors and stick to them.
  • Topics to avoid: Competitors, legal advice, medical advice, anything off-limits for your context.
  • Things to always do: Sign off with a contact email, always acknowledge the visitor's name, always suggest the knowledge base first.
  • What to say when you don't know: Though the "when Vex doesn't know" behavior is also set separately (see AI Identity → When AI Doesn't Know Answer), you can reinforce the message here.

Tone settings beyond the system prompt

Two quick toggles in AI Identity affect tone without writing a word of instructions:

  • Emoji usage: Enable to let Vex sprinkle emojis naturally. Disable for a more formal voice.
  • Company name: Always fill this in — Vex uses it in replies to refer to your brand correctly.

When Vex doesn't know the answer

In the AI Identity tab, under When AI Doesn't Know Answer, choose one of three behaviors:

  • Politely decline (recommended): Vex says it doesn't have that information and suggests contacting a human. Safest for most setups.
  • Transfer to human agent: Vex automatically escalates the conversation when it can't answer. Use this if response coverage is critical.
  • Answer anyway: Vex uses its general LLM knowledge to reply even without a source match. Useful for general-knowledge questions, but risks hallucinated answers. Use with caution.

Tips for effective instructions

  • Be specific, not vague. "Be friendly" is hard to act on. "Use first names when the visitor introduces themselves, never use jargon like 'ticketing system', and keep responses under 4 sentences" is actionable.
  • List what NOT to do. Restrictions often have more impact than positive instructions.
  • Keep it short. Aim for under 300 words. Vex reads the full prompt on every message — very long prompts slow responses and dilute the most important rules.
  • Test after every change. Use the Preview tab to check that Vex actually behaves the way you intended.

Example instructions

You are Vex, a support agent for Acme Corp. Your job is to help customers with questions about our project management software.

Tone: Friendly but professional. Use first names when offered. Keep replies concise — 2–3 sentences max unless a step-by-step answer is needed.

Always: Suggest the knowledge base first. If you can't find the answer, say so clearly and offer to connect the visitor with a human agent.

Never: Discuss competitors. Give pricing advice beyond what's in the knowledge base. Make promises about features that aren't documented.

If asked about your technical nature, say: "I'm Vex, Acme's AI support assistant."

Next step

Once you've written your instructions, test them with the Preview before going live: Test your chatbots before going live. To control when Vex escalates to a human, see Control when Vex hands off to a human.

Did this article answer your question?