> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cura.so/llms.txt
> Use this file to discover all available pages before exploring further.

# Advanced AI instructions

> When tone and length aren't enough, Advanced AI instructions let you guide your AI in your own words.

When tone and length aren't enough, **Advanced AI instructions** let you guide your AI in your own words. Think of it as the extra coaching you'd give a new front-desk hire — how to phrase things, when to call a manager, and any rules unique to your business.

By the end of this guide, you'll have given your AI clear, specific guidance for the situations that matter most to you.

## Before you begin

* You'll need **Admin** or **Manager** access.
* Set your [tone and length in Rules](/ai-agent/ai-tone-and-length) first — instructions build on top of them.
* Instructions apply to your whole account — the same guidance applies across every workspace and channel. (Facts like prices and FAQs are still set per workspace.)

***

<Steps>
  <Step title="Open System Instructions">
    * Go to **Cura AI** and open the **System Instructions** tab.

    <Frame>
      <img src="https://mintcdn.com/cura-2d9288ab/mERlrNpxELq5pJa2/images/ai-agent/ai-instructions.png?fit=max&auto=format&n=mERlrNpxELq5pJa2&q=85&s=41804ee5628dcec0c1ea66f82fe9cf48" alt="The Guidance tab in Cura where you write advanced AI instructions" width="2880" height="1560" data-path="images/ai-agent/ai-instructions.png" />
    </Frame>
  </Step>

  <Step title="Set your communication style">
    The **Communication style** field is where you describe *how* the AI should talk, beyond tone and length.

    * Use plain instructions, e.g. *"Always greet customers warmly and use their first name. Use emojis sparingly. Avoid medical jargon."*
    * Keep it specific and short — a few clear sentences work better than a long essay.
  </Step>

  <Step title="Set escalation & handoff guidance">
    The **Escalation & handoff** field tells the AI *when* to step back and bring in a human.

    * Spell out the moments that should always go to a person, e.g. *"Hand off if a customer is upset, asks about surgical complications, or requests a refund."*
    * The AI also escalates automatically when a customer asks for a human or when it can't help — this field adds *your* extra triggers on top.
  </Step>

  <Step title="Add any other guidance">
    The **Other** field is for anything that doesn't fit the first two.

    * Business-specific rules, e.g. *"We don't quote prices for laser packages over the phone — invite the customer to book a consultation instead."*
    * Things to avoid, promotions to mention, or how to handle particular requests.

    <Check>Your AI now follows your custom guidance in every conversation, across all workspaces.</Check>
  </Step>

  <Step title="Save and test">
    * Save your changes, then send a test message to a connected channel to confirm the AI behaves as you intended.

    <Tip>
      Write instructions the way you'd brief a person. The AI reads them as guidance, so clear, concrete examples ("do this", "never do that") work far better than vague goals.
    </Tip>

    <Note>
      Instructions shape behaviour, not facts. Prices, policies, and treatment details still come from your Services, FAQs, and Knowledge Base.
    </Note>
  </Step>
</Steps>

***

## Troubleshooting

* **The AI isn't following an instruction.** Make it more specific and concrete. If it conflicts with your Knowledge Base or Services, the facts there still win.
* **The AI escalates too often (or too rarely).** Tighten the wording in the Escalation & handoff field to describe exactly when to hand off.

***

### Related articles

* [Setting your AI's tone and length (Rules)](/ai-agent/ai-tone-and-length)
* [Escalation: when the AI hands off to a human](/ai-agent/escalation-handoff)
* [How your AI agent works](/ai-agent/how-your-ai-agent-works)