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

# AI Behavior

> Troubleshoot and tune your AI assistant's conversation behavior — fix misunderstandings, unnatural flow, and repetitive responses.

Your AI assistant's behavior is shaped by two things: the **engine mode** you've selected and the **system prompt** you've written. Most behavior problems can be resolved by adjusting one or both.

## Engine Modes

DialogBot offers three engine modes. The mode you choose affects how your assistant processes speech and generates responses.

<CardGroup cols={3}>
  <Card title="Fast Engine" icon="bolt">
    Quick, conversational responses. May occasionally miss context on complex inputs. Best for simple interactions and outbound sales calls. Pairs well with Filler Audio for a natural flow.
  </Card>

  <Card title="Standard Engine" icon="gauge">
    Balanced speed and understanding. The recommended starting point for most new assistants and a good default for general use cases.
  </Card>

  <Card title="Accurate Engine" icon="brain">
    Deep context understanding and more considered responses. Best for complex support scenarios or detailed information gathering. Expect slightly longer pauses between turns.
  </Card>
</CardGroup>

## Common Issues and Fixes

<AccordionGroup>
  <Accordion title="AI misunderstands what the caller said">
    The assistant may be using a faster engine that prioritizes speed over accuracy.

    **Steps to fix:**

    1. Switch to **Accurate Engine** in your assistant's AI Engine Configuration settings.
    2. Add more context to your system prompt — explain who the caller is likely to be and what they typically ask.
    3. Include example conversations in your prompt to show the assistant how to handle ambiguous inputs.
  </Accordion>

  <Accordion title="Conversation feels unnatural or robotic">
    Unnatural flow is often caused by processing pauses or overly rigid prompt phrasing.

    **Steps to fix:**

    1. Enable **Filler Audio** in your assistant settings so the AI fills silence with natural sounds ("hmm", "one moment") while processing.
    2. Switch to **Fast Engine** for quicker back-and-forth exchanges.
    3. Add sample conversation snippets to your prompt to demonstrate a more natural cadence.
  </Accordion>

  <Accordion title="AI gives the same response repeatedly">
    Repetitive responses usually indicate the prompt is too narrow or the engine is over-fitting to a short instruction set.

    **Steps to fix:**

    1. Review and expand your system prompt — add variation examples and cover edge cases.
    2. Explicitly instruct the assistant to vary its phrasing when repeating information.
    3. Consider switching from Fast to **Standard** or **Accurate** engine for more varied output.
  </Accordion>

  <Accordion title="AI response is too slow">
    Slow responses can frustrate callers, especially in sales or high-volume scenarios.

    **Steps to fix:**

    1. Switch to **Fast Engine** mode.
    2. Enable **Filler Audio** to mask processing time with natural sounds.
    3. Simplify your system prompt — long, complex prompts increase processing time.
  </Accordion>
</AccordionGroup>

## Optimization by Use Case

Different call types benefit from different configurations:

| Use case           | Recommended engine | Additional tip                              |
| ------------------ | ------------------ | ------------------------------------------- |
| Outbound sales     | Fast               | Enable Filler Audio for engagement          |
| Customer support   | Accurate           | Add detailed context and FAQs to the prompt |
| Lead qualification | Standard           | Good balance for mixed conversation types   |

## Testing Your Changes

After adjusting engine settings or your prompt, validate the changes before deploying to live calls:

<Steps>
  <Step title="Run a test call">
    Use the **Test Call** feature in your assistant settings to simulate a real inbound or outbound call.
  </Step>

  <Step title="Try multiple scenarios">
    Test edge cases — unusual questions, interruptions, unclear speech — not just the happy path.
  </Step>

  <Step title="Review recordings">
    Listen to the call recording and review the transcript to check whether the behavior has improved.
  </Step>
</Steps>

<Tip>
  Start with **Standard Engine** and only switch modes once you've identified a specific problem. Changing both the engine and the prompt at the same time makes it harder to know which change had the effect.
</Tip>