Test an agent

Test an agent before you implement it on your website and after you change its guidelines or connected topics. Testing helps you find scope, content, and answer-quality issues before visitors encounter them.

Test an agent:

• after creating it and adding its topics
• after changing Answer Guidelines or the Default Response
• after adding, removing, or changing a monitored topic
• after important website content changes are processed by Content Intelligence
• when reviewing a conversation reveals an unexpected answer.

1. Open Conversations Agents.
2. Select the agent you want to test.
3. Click the test action available for the agent.
   1. If the test interface is not available, confirm that the agent has been saved and that you have access to Conversations.
      
4. Enter a question in the test interface.
5. Submit the question and wait for the answer.
6. Review the answer, sources, and any confidence or validation information shown.
7. Ask follow-up questions if your website implementation will support threaded conversations.

Use questions that reflect how visitors are likely to search. Include:

• direct questions about information in each connected topic
• questions using different words for the same concept
• follow-up questions that rely on the previous answer
• questions that are outside the agent’s scope
• questions for which the website does not contain an answer
• ambiguous or incomplete questions

Do not use real personal, confidential, or sensitive information in test questions.

A useful answer should:

• address the question directly
• be consistent with the source pages
• use the configured tone and structure
• stay within the agent’s connected topics
• provide useful source links where available
• use the Default Response rather than inventing information when the question cannot be answered

If an answer is out of scope, check the connected topics and the question’s intended scope. If the content is correct but the presentation is unsuitable, revise Answer Guidelines. If the answer is inaccurate or incomplete, review the source content and the topic’s AI Readiness issues before changing the agent.

Repeat the test after each change and record important questions as a regression set for future checks.