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

# Guidelines

> Guidelines are conditional rules for handling soft edge cases during a conversation — situations that might not fit neatly into your main conversation flow. 

## What are Guidelines?

Guidelines define condition-response pairs for handling edge cases:

* **Condition** — When this situation occurs...
* **Response** — ...the assistant should respond this way

Guidelines are best suited for behavioral guidance (how to handle a situation), not factual information (what to say). Put specific data like pricing, hours, or product details in the Knowledge Base. Guidelines are a fallback for information not available in the Knowledge Base.

For specific facts, data, or detailed information the assistant should know, use the [Knowledge Base](/scripts/knowledge-base) instead.

***

## How Guidelines Work

Guidelines can be input from the script builder under the "Guidelines" section. When generating a response, the AI considers all active guidelines and applies relevant ones based on the conversation context.

Guidelines directly influence the assistant's behaviour, shaping tone, personality, and how it handles specific situations. However, some behaviours have inherent limitations — see [Assistant Behaviour](/conversation-dynamics/assistant-behaviour) for details.

***

## Writing Guidelines

Each guideline has two parts:

| Field         | Purpose                                    | Example                      |
| ------------- | ------------------------------------------ | ---------------------------- |
| **Condition** | Describes when the guideline applies       | "User asks about pricing"    |
| **Response**  | Describes how the assistant should respond | "Direct them to the website" |

### Examples

**Handling Pricing Questions:**

```text theme={null}
Condition: User asks about pricing or costs
Response: Explain that pricing varies by needs and offer to have a specialist follow up with detailed pricing information
```

**Technical Questions:**

```text theme={null}
Condition: User asks highly technical questions beyond scope of knowledge
Response: Note their question and offer to have a technical specialist contact them with detailed information
```

**Emotional Situations:**

```text theme={null}
Condition: User expresses frustration or anger
Response: Acknowledge their feelings, apologize for any inconvenience, and focus on finding a resolution
```

***

## Best Practices

Guidelines shape [assistant behaviour](/conversation-dynamics/assistant-behaviour) including tone and response style, but some behaviours have limitations. Keep these tips in mind:

1. **Be specific with conditions** — Vague conditions may trigger unexpectedly or be missed entirely
2. **Keep responses actionable** — Tell the assistant what to do, not just what to avoid
3. **Don't duplicate flow logic** — Use guidelines for cross-cutting concerns, not primary conversation paths
4. **Test edge cases** — Verify guidelines work as expected in real conversations
5. **Prioritize safety** — Include guidelines for handling sensitive topics appropriately

***

## Common Use Cases

| Use Case             | Example Condition                            | Example Response                                                                    |
| -------------------- | -------------------------------------------- | ----------------------------------------------------------------------------------- |
| **Scope boundaries** | User asks about unrelated services           | Politely explain our focus and offer relevant alternatives                          |
| **Compliance**       | User requests medical/legal/financial advice | Clarify we cannot provide professional advice and recommend consulting a specialist |
| **Escalation**       | User requests to speak with a human          | Acknowledge the request and initiate warm handoff                                   |
| **Data privacy**     | User asks what data we collect               | Explain our privacy practices and offer to send our privacy policy                  |

***

## Next Steps

* [Call Objectives](/scripts/call-objectives) — Define what your assistant should accomplish
* [Knowledge Base](/scripts/knowledge-base) — Give your assistant reference information to draw from
* [Assistant Behaviour](/conversation-dynamics/assistant-behaviour) — Understand how guidelines shape the assistant's tone and responses
