Wrong tone, writing style, or general behavior
If the AI uses the wrong tone of voice, inappropriate language, or doesn’t follow your brand guidelines:- Review your Company settings. Check that your global guidelines cover tone, style, response structure, and boundaries.
- Check channel-specific configuration in Email Configuration or Chat Configuration.
Incorrect handling of specific topics
If the AI mishandles a particular type of request (e.g., cancellations, refunds, shipping inquiries), review or create a Specialist. Specialists define how the AI handles specific question categories, so ensure the relevant one has clear instructions for the scenario.Wrong actions taken
If the AI performs the wrong action, skips an action, or takes actions it shouldn’t:- Review your Actions configuration. Verify the action description clearly explains when to use it.
- Check the Specialist instructions. Actions must be added using the
@tool picker to be available.
Incorrect or missing information in responses
If the AI provides wrong facts, outdated information, or can’t answer questions it should know:- Review your Knowledge sources. Ensure the correct information is available and the source’s “When to use” description matches the scenarios where it’s needed.
- Check access levels to verify the information can be shared with customers.
After making changes, use the Simulator to validate the AI’s behavior before going live.