Best Practices

Get the most out of your AI agent with these proven strategies for training, configuration, and ongoing optimization.

Training Content Tips

Start with High-Impact Content

Focus on content that addresses the most common questions:

• FAQ pages and help articles
• Product documentation
• Pricing and feature comparisons
• Getting started guides

Keep Content Focused

Quality beats quantity. An agent trained on 10 relevant pages outperforms one trained on 100 loosely related pages.

Use Q&A Sources for Critical Answers

For questions requiring exact responses (pricing, policies, legal info), use Q&A sources to define specific question-answer pairs.

Update Regularly

Enable auto-retraining (paid plans) to keep web sources current. For manual updates, retrain after significant content changes.

Writing Effective Instructions

The AI instructions (in the AI tab) shape how your agent responds. Here's a template:

You are a helpful assistant for [Company Name].

Your role:
- Answer questions about [product/service]
- Be friendly but professional
- If you don't know something, say so honestly

Guidelines:
- Keep responses concise (2-3 paragraphs max)
- Include links to relevant pages when helpful
- Never make up information not in your training data

Good vs. Poor Instructions

Poor: "Be helpful"

Good: "You are a customer support assistant for Acme Software. Answer questions about our project management tool. Be concise and friendly. If asked about pricing, direct users to our pricing page at /pricing."

Choosing the Right AI Model

Recommendation: New agents on every plan, including Free, start with GPT-6 Sol. In our evaluation, it was the most reliable at answering from your content and at running actions and multi-step tools. Existing agents keep their selected model. Claude Sonnet 5 and Gemini 3.1 Pro are strong alternatives. Use GPT-6 Luna, Claude Haiku 4.5, or Gemini 3.7 Flash for high-volume use cases.

Plan access: Free accounts can use GPT-6 Sol, GPT-5.6 Terra, GPT-6 Luna, Claude Haiku 4.5, and Gemini 3.7 Flash. All paid plans (Hobby, Standard, Pro, Enterprise) include ten core models, including GPT-6 Sol, GPT-6 Luna, and Claude Opus 5.5. GPT-6 Astra and Claude Fable 5.1 require Pro or higher.

Testing and Iteration

Test Regularly in Playground

Use the Playground to test common questions before and after making changes.

Compare Models Side-by-Side

Use the Compare feature to see how different models answer the same question.

Review Chat Logs

Check Chat Logs weekly to identify:

• Questions your agent struggles with
• Missing content topics
• Opportunities to improve instructions

Common Issues and Fixes

"The agent doesn't answer my question correctly"

• Add more relevant training content covering the topic
• Create a Q&A source with the exact question and answer
• Refine your instructions to guide the AI's behavior

"Responses are too long/short"

• Add length guidance in your instructions: "Keep responses to 2-3 sentences" or "Provide detailed explanations"

"The agent makes things up"

• Ensure your instructions say "Only answer based on your training data"
• Add Q&A sources for critical/factual information
• Consider using a more capable model (GPT-5.6 Terra or Claude Sonnet 5 vs the fast tiers)

Next Steps

• Configure AI settings to customize behavior
• Set up lead collection to capture visitor info
• Monitor chat logs to track performance