Help Center

The Knowledge Base Retrieval Tester is a tool in your CRM that allows you to preview how your Knowledge Base content will perform when answering real user questions through AI features. Before your AI tools—such as chatbots or virtual assistants—go live, this tester provides a controlled environment to verify that the information stored in your Knowledge Base can be accurately retrieved and used.

Your CRM Knowledge Base can include various types of content, such as documents, rich-text articles, tables, and web-crawled pages. The system processes this content by extracting, structuring, and breaking it into searchable segments. The Retrieval Tester lets you validate this process by entering sample questions and reviewing exactly which content pieces are returned in response.

Key Benefits of Using the Retrieval Tester

Using the Retrieval Tester shifts your approach from assumption to verification, ensuring your Knowledge Base is reliable before customers interact with it.

Real-Time Validation: Enter a question and immediately see how your Knowledge Base responds.
Source Transparency: Review which specific content sources or segments were used to generate each answer.
Proactive Troubleshooting: Identify missing, outdated, or poorly structured content before it impacts live interactions.
Content Improvement: Use test results to refine documents, tables, articles, and other source types.
Safe Testing Environment: Conduct read-only tests without altering your live Knowledge Base or affecting active conversations.
Rapid Iteration: Update your content and retest quickly to confirm improvements.

When to Use the Retrieval Tester

Incorporate the Retrieval Tester into your workflow whenever you need confidence that your AI tools will retrieve accurate and relevant information. Key moments to use it include:

After adding new content, such as documents, PDFs, rich-text articles, tables, or web sources.
When updating existing Knowledge Base entries.
Before launching an AI agent, chatbot, or voice assistant.
If you notice AI responses are incomplete, irrelevant, or based on incorrect sources.
To compare how different phrasings of a question retrieve different content segments.
To check for duplicate or conflicting information within your Knowledge Base.

Understanding Retrieved Sources and Content Segments

When you run a test, the results show which sources from your Knowledge Base were deemed relevant to your question. The CRM often breaks content into smaller, searchable segments (sometimes called chunks) to improve retrieval accuracy. For example, documents may be split by headings, and tables may be indexed by row.

When reviewing test results, pay attention to:

Whether the expected source appears in the results.
If the returned segment directly answers the question.
Whether any irrelevant or outdated content is being retrieved.
If similar sources contain conflicting or duplicate information.
Whether the content is clear, specific, and uses terminology your customers are likely to use.

How to Set Up and Use the Knowledge Base Retrieval Tester

To get meaningful results, ensure your Knowledge Base contains the content you want to test and that it has finished processing.

Navigate to AI Agents > Knowledge Base.
Open the Knowledge Base you wish to test.
Verify that it includes the sources you want to validate—documents, rich-text, tables, web content, etc.
Wait for any new or updated content to complete indexing.
Open the Knowledge Base Retrieval Tester.
Enter a realistic question a customer might ask, such as “What is your return policy?” or “How do I reset my password?”
Review the generated answer and the list of retrieved sources.
If the results are incorrect, update your Knowledge Base content and retest.

Best Practices for Testing

Effective testing involves simulating real user behavior and varied question styles.

Start with simple, direct questions.
Use natural language and customer phrasing, not just internal jargon.
Phrase the same question in multiple ways to check consistency.
Include questions that should trigger specific documents, table rows, or policy sections.
Test edge cases and high-stakes questions that could impact customer satisfaction.
Always review the sources behind the answer, not just the answer itself.
Remove or update outdated content that might conflict with newer information.
If retrieval is poor, consider breaking long or complex content into clearer sections.
Retest after every significant update to your Knowledge Base.

Troubleshooting Retrieval Results