Best Practices for AI Agents

Write Clear Instructions

Your agent's instructions define its personality, knowledge scope, and response style. Well-written instructions are critical for accurate, on-brand responses.

Instruction Template

Tip: Start simple, then iterate based on real conversations. Review chat logs to see where your agent struggles, then refine instructions accordingly.

Structure Training Data for Clarity

The quality of your agent's responses depends entirely on the clarity of your training data. AI can't accurately answer questions about ambiguous, unclear, or poorly formatted information.

Key principle: Specific, clear information → accurate agent responses. Ambiguous information → confused agent responses.

Train Your Agent to Include Links

If you want your agent to recommend specific pages or resources, you must explicitly teach it those links. The agent doesn't automatically learn URLs—it learns from your training content.

How to Add Links

Create a text document in your training data that maps resources to page names:

Example:

Product Documentation: https://docs.example.com/products
Pricing Page: https://example.com/pricing
Contact Support: https://example.com/support
Help Center: https://example.com/help

When your agent has this information in its training data, it will naturally recommend the appropriate link when relevant.

Iterative Improvement

The best agents are built through continuous improvement:

1. Deploy and monitor — Review conversations in the Activity tab
2. Identify gaps — Look for questions the agent answers poorly
3. Improve training data — Add Q&A pairs, clarify existing information
4. Test changes — Use the Playground to verify improvements
5. Deploy updates — Retrain and push live

See Activity to review conversations and improve responses.