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

# Best Practices

> Get the most accurate, on-brand answers from your agent

Setting up an agent takes minutes, but a few habits make a large difference in answer quality and customer experience. Use these as you build and refine.

## Build a strong knowledge base

Your agent answers from the knowledge you give it, so the quality of that content sets the ceiling on answer quality.

<AccordionGroup>
  <Accordion title="Prepare documents carefully" icon="file-check">
    Upload files that are clear, structured, and complete: product details, FAQs, policies, and guides. Well-organized source material produces more precise answers.
  </Accordion>

  <Accordion title="Add relevant links" icon="link">
    Add links to your website, blog, and policy pages. Make sure they point to accurate, up-to-date content, and remove outdated or broken links so the agent doesn't serve old information.
  </Accordion>

  <Accordion title="Use Q&A for gaps" icon="question">
    When customers ask things your documents don't cover well, add them as specific Q\&A pairs. This gives the agent an exact, brand-approved answer for those questions.
  </Accordion>

  <Accordion title="Retrain after changes" icon="rotate">
    After adding or editing knowledge, click **Train agent** so the changes take effect. On the Scale plan you can enable automatic daily retraining.
  </Accordion>
</AccordionGroup>

## Shape behavior with instructions and skills

* Write clear **instructions** in **Settings → AI**: the agent's role, tone, and what it should and shouldn't do.
* Use **[Skills](/dashboard/agent-skills)** for situation-specific behavior (returns, upsell, troubleshooting) instead of overloading one big instruction block.
* Keep the **temperature** low for support and factual answers so responses stay consistent.

## Test before and after you ship

* Use the **[Overview](/dashboard/overview)** live preview to try questions exactly as a customer would.
* Use **Compare** to run the same prompt against different models, temperatures, or instructions side by side, and keep the setup that performs best.

## Monitor and improve

* Check **[Analytics](/dashboard/analytics)** for engagement, resolution, sentiment, and the questions customers ask most.
* Review **[Conversations](/dashboard/conversations)** regularly. When an answer isn't ideal, use **Revise Answer** so the agent uses your improved version for similar questions.
* Turn recurring **knowledge gaps** (questions the agent couldn't answer well) into new knowledge or Q\&A.

## Track business impact

If you run a store, watch add-to-cart events, checkout starts, and attributed revenue in **Analytics → Revenue & Commerce**, and watch captured leads from your **[AI Actions](/dashboard/ai-actions)**. This shows the real impact your agent has, not just message volume.